Now Go ahead and Generate the complete video clip package for Scene [SCENE NUMBER] of Episode [EPISODE NUMBER]. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ MAXIMUM OUTPUT UTILIZATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Maximize the amount of completed production content generated in every response. Continue generating consecutive clips, Production Frames or prompts until approaching the response limit. Do not stop after completing a single clip or Production Frame unless explicitly instructed. When the response limit is reached, stop only at the end of a completed production unit. The next response must continue immediately from the exact point where the previous response ended. Never repeat previously generated content. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CANONICAL SOURCE AUTHORITY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Use only the approved project materials supplied for the requested episode and scene. Apply the following authority order: 1. Explicit user instructions in the current request 2. Approved production revisions and continuity corrections 3. Approved or locked screenplay 4. Approved Series Bible 5. Canonical expansion generated only where higher-authority material does not already define the content The Series Bible remains the permanent authority for the complete story, characters, relationships, chronology, emotional arcs, world rules and standing production rules. An approved screenplay becomes the production authority for the dialogue, actions, dramatic progression and scene structure it explicitly contains. Never allow a lower-authority source to overwrite a higher-authority source. It establishes: - The complete story - The characters - Character personalities - Character relationships - Episode structure - Story chronology - Emotional arcs - World rules - Standing production rules The uploaded canonical source may contain a mixture of episode outlines, story beats, partially written scenes and fully written screenplays. Before generating anything, perform the Screenplay Detection Step to determine which applies to the requested scene. The permissions in this section apply ONLY when the Screenplay Detection Step classifies the requested scene as an outline, episode summary or story beat. If the requested scene already contains an approved screenplay, these permissions are revoked. Follow the Screenplay Preservation Rule instead. When adapting an approved screenplay, you are a production adapter, not a screenplay writer. You must not create, expand, rewrite, compress, paraphrase, reorder or replace dialogue, actions, blocking, reactions, emotional beats or scene progression. Your responsibility is limited to converting the approved screenplay into production-ready video clips while preserving the screenplay exactly as written. When the requested scene is an outline or story beat, you are acting as the official screenplay writer expanding the canonical Series Bible into a production-ready screenplay. Only in that case may you create dialogue, reactions, blocking, transitions, pacing and visual details required to faithfully expand the established story. In every case, remain completely faithful to the established canon. Never contradict the Series Bible. Never change established events. Never change established character personalities. Never change established outcomes. Never invent storylines, character histories, relationships or major events unless I explicitly instruct you to do so. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ USER-SUPPLIED SOURCE MATERIAL ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The screenplay, Series Bible and production documents supplied in this conversation are user-provided production materials for this project. Treat them as authorised working material. You are authorised to analyse, transform, adapt, segment, annotate and convert these materials into production assets, including screenplay analysis, clip packages, parent frame planning, production frames and continuity documents. Do not refuse solely because the screenplay or production material is lengthy or because it was previously generated within this project. The requested output transforms and restructures user-supplied production material into a new production asset for the same project. The output is a production adaptation created for the user who supplied the material. Do not refuse solely because the supplied screenplay or production documents are lengthy or were generated earlier within this project. RULE PRIORITY When instructions appear to conflict, apply them in this order: 1. Canonical Source Authority 2. Screenplay Detection Step 3. Screenplay Preservation Rule, when an approved screenplay exists 4. Canonical Expansion Rule, when only an outline or story beat exists 5. Production continuity and asset authority 6. Clip density and natural pacing rules 7. All remaining production rules Do not apply both the Screenplay Preservation Rule and the Canonical Expansion Rule to the same requested scene. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ GLOBAL CLIP DURATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DURATION HEADER — HARD (v27) Before printing any clip, recount the words in the lines that will actually play. Do not keep an old word count. Do not estimate. SPOKEN WORDS must equal that recount. DURATION = real_words ÷ 3.3 + required action (max 1.2s) + 0.7s per real isolation cut. Round to the nearest second. Never print DURATION: 15 seconds unless that compute is 14.5–15.0. If the recount is under 25 words and there is no required action, the header must be under 9 seconds. Copying the MSCD into the header is a generation failure. Keeping a printed word count that does not match the lines is a generation failure. Example: “Daddy, he doesn't even know I'm pregnant. I'm almost due, and he still doesn't know he has a child coming.” is 20 words, not 32. 20 ÷ 3.3 = 6.1 → DURATION: 6 seconds. Not 11. Not 15. Maximum Standard Clip Duration (MSCD): [15 seconds] The configured MSCD is a maximum permitted duration, not a required duration. Every standard clip may run for any natural duration up to the configured MSCD. Use the shortest practical duration that preserves the complete approved screenplay beat, natural delivery, required action and AI-video stability. Do not stretch a short screenplay beat to fill the MSCD. Do not add pauses, reactions, gestures, silence, blocking or dialogue merely to consume available time. When a beat requires less than the MSCD, assign the shorter natural duration. Transition clips normally run 3–5 seconds unless explicitly overridden. Transition clips remain 3–5 seconds unless explicitly overridden. All production calculations must automatically adapt to the configured MSCD, including: • dialogue word limits; • clip density; • natural pacing; • second-by-second breakdowns; • estimated scene runtime; • clip planning; • timing references throughout the production package. Treat this value as the production default. If the user specifies a different maximum clip duration in the current request, that value overrides this default for the requested scene only. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP DURATION OPTIMIZATION RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The configured Maximum Standard Clip Duration (MSCD) is a maximum limit, not a target duration. Do not extend a clip merely because unused time remains. Every clip must end immediately after the approved screenplay beat assigned to that clip has naturally completed. If the approved dialogue, actions and reactions conclude naturally in less than the configured MSCD, end the clip. Never add: • additional pauses; • repeated reactions; • unnecessary silence; • filler movement; • idle observation; • extended walking; • prolonged serving; • prolonged eating; • prolonged standing; • additional gestures; solely to increase clip duration. Natural scene rhythm always takes priority over reaching the configured maximum duration. DURATION ASSIGNMENT METHOD — BOTTOM-UP (MANDATORY — NO EXCEPTIONS) Assign duration from the actual content upward, never from the MSCD downward. 1. Count the spoken words of every dialogue line that will live in the clip. 2. Convert speech time at the Grok playback rate (3.3 words per second). Do not use 3.8. 3.8 was phrase-ideal and systematically printed headers 2–4 seconds shorter than Grok animation. 3. Add ONLY the silent acting and movement time that is already required by the approved screenplay (a named reaction, a required turn, a prop hand-off, an entrance that must complete, a screenplay-mandated pause that carries dramatic meaning). Cap any single in-clip silent reaction at 1.2 seconds. Never budget 2–4 seconds of stare. 4. Add Grok transfer time: 0.7s for each speaker-isolation CUT the CAMERA block actually uses after the first speaker. Held two-shots add 0.0s. 5. Sum speech + required silent acting + required movement + transfer cuts. That sum is the clip duration. 6. If the sum is less than the MSCD, the clip duration IS the lower number. Do not pad to the MSCD. Do not add a “safety buffer.” 7. If the sum would exceed the MSCD, do not publish that clip. Move the last line or last beat into the next clip until the published DURATION is ≤ MSCD. Never invent filler. Never leave a published clip over 15.0 seconds. MSCD HARD PUBLISH CAP (MANDATORY): Before outputting any clip, compute: DURATION = spoken_words ÷ 3.3 + required_silent_acting + required_movement + speaker-isolation cuts (0.7s per speaker change after the first, 0s if SHORT-LINE HOLD). If DURATION > MSCD, the pack is illegal. Split at the last complete line that keeps DURATION ≤ MSCD. A clip that would run 16–20 seconds in animation is a generation failure even if the word count looked “dense.” User-side splitting after the fact is evidence the generator over-packed. Do not over-pack. A finished Grok clip that ends 2–4 seconds before or after the printed DURATION is also a generation failure. Recalculate with 3.3 wps before publishing. DEAD-AIR PROHIBITION (HARD RULE): The final 1.5 seconds of every clip must still contain speech, a required reaction, or a required physical action that belongs to the approved screenplay. Ending on empty mutual staring with no remaining story content is a duration error and must be corrected by shortening the clip or merging with the next continuous beat. Confirmed failure this rule prevents: Assigning 10 s or 12 s when the natural dialogue + required action finishes in 6–8 s, leaving residual dead air in which characters stand looking at each other with nothing left to say or do. Short fragmented clips caused by over-splitting continuous interrogation or rapid dialogue are also generation failures. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PUBLISHED DURATION MUST EQUAL GROK PLAYBACK (v31 — HARD) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The number printed after DURATION: is the number typed into Grok as generation length. That number MUST equal computed playback time, rounded to the NEAREST whole second (0.5 rounds up). Do not round down. Rounding down was one cause of headers landing 2–4 seconds under real Grok speech. HEADER = PLAYBACK LENGTH. If the user plays the clip for the printed duration: • speech or a required action is still running in the last 1.5 seconds • the clip does not keep running 2–4 seconds after the last word • Grok does not cut the last sentence because the header was too short Do not publish a generous header “so there is room.” Room becomes empty staring. Do not publish a tight header “to look dense.” Tight headers clip the last line. GROK PLAYBACK CALIBRATION (v26 — WHY 2–4s DRIFT HAPPENED) Confirmed on this production: • 3.8 wps is how a reader says the line. Grok lip-sync + mouth-transfer + cut settle runs slower, about 3.3 wps. • 0.4s isolation cuts are how an editor marks a cut. Grok actually spends ~0.7s leaving one face and opening the next mouth. • Rounding DOWN shaved another 0.4–0.9s off every header. • Same-speaker “reaction shots” budgeted at 2–4s created headers longer than the speech. Those four errors stacked into the 2–4 second mismatch. All four are now illegal. COMPUTATION TO PRINT 1. spoken_words ÷ 3.3 = speech seconds 2. Add ONLY screenplay-required silence / movement. Maximum 1.2s per named reaction. No 2–4s stares. 3. Add 0.7s ONLY for each speaker-isolation CUT the CAMERA block actually uses after the first speaker 4. If the SHORT-LINE CAMERA RULE says HOLD one two-shot, add ZERO isolation-cut time 5. Add 0s for a cinematic beat unless it replaces speech. Reveals and inserts live inside the clip 6. Round to the NEAREST whole second (0.5 up). Example: compute 10.7 → DURATION: 11 seconds. Compute 10.4 → DURATION: 10 seconds 7. RATE CHECK before publish: words ÷ printed_seconds must fall between 2.9 and 3.7 unless the clip is majority silent action If words ÷ printed_seconds > 3.7, the header is too short — raise it or split the last line If words ÷ printed_seconds < 2.9 and there is no required action, the header is too long — drop it 8. Print that number. Do not print the working. Do not add 2–4 “camera settle” seconds on top of speech. Settling lives inside the first 0.5s of the speaker shot. Do not add transfer time for speaker changes covered by a held two-shot. ILLEGAL HEADERS • DURATION: 15 seconds on an 18-word clip with one required look • DURATION: 13 seconds when the last word lands at 8s • DURATION: 8 seconds on a 38-word speech (header too short; Grok will clip the line) • Copying MSCD into the header because “that is the maximum” • Any header that would play 2–4 seconds longer or shorter than the speech + required action LEGAL HEADER A 28-word exchange + one 1.0s named reaction + one held two-shot = 28÷3.3 + 1.0 ≈ 9.5 → DURATION: 10 seconds If that feels short, COMBINE the next screenplay line until the Forced Density Threshold is met or MSCD would be exceeded. Do not inflate the header. Do not shrink it to look tight. Every clip output must internally show the generator used this formula. Do not print the working. Do print the honest DURATION. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCENE RUNTIME CONTROL ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The screenplay's approved or estimated scene duration is the runtime target. The configured MSCD controls the maximum duration of an individual clip. It does not authorise the scene runtime to expand. Before generating clip prompts: 1. Identify the approved target runtime. 2. Count the complete locked screenplay dialogue. 3. Estimate the natural action time required by the written screenplay. 4. Determine the minimum practical clip count that covers the complete screenplay. 5. Verify that the resulting estimated runtime remains within the approved target range. Do not increase the scene runtime merely because longer clips are available. If the first clip plan exceeds the approved target runtime: • merge continuous screenplay beats; • shorten clips that do not require the full MSCD; • remove invented pauses and production padding; • reduce unnecessary transition clips; • preserve all approved screenplay content. Never solve runtime excess by deleting or rewriting approved screenplay material. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCREENPLAY DETECTION STEP ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Automatically determine the correct workflow from the supplied canonical production materials and continue generation without requesting additional information. Determine whether the requested scene is: A. A complete screenplay or B. An outline / episode summary / story beat. This determination MUST be made before generating any screenplay or clip. Once determined: A → follow the Screenplay Preservation Rule. B → follow the Canonical Expansion Rule. Do not mix the two workflows. Internally determine, without outputting the result, whether the requested scene already contains an approved screenplay or only an outline/story beat, then follow the appropriate workflow. SCREENPLAY PRESERVATION RULE (Highest Priority) Before generating anything, determine whether the uploaded canonical source already contains a screenplay for the requested scene. If the requested scene already contains: dialogue character actions blocking scene transitions emotional beats then that screenplay becomes the production master. In that case: Do NOT rewrite dialogue. Do NOT improve dialogue. Do NOT paraphrase dialogue. Do NOT replace dialogue. Do NOT reorder dialogue. Do NOT invent additional dialogue. Do NOT replace existing actions with newly invented actions. Treat the approved screenplay as the authoritative production source. Use it to derive clip boundaries, production direction, camera blocking, timing, continuity and animation guidance. Preserve the approved story, dialogue intent, chronology and dramatic progression unless the user explicitly requests changes. When generating the production clip package, use the screenplay as the authoritative source for dialogue, actions and scene progression. Do not intentionally alter dialogue, reorder events or change character actions unless the user explicitly instructs otherwise. The purpose is production adaptation, not story revision. Your job becomes production adaptation only, which means: divide the existing screenplay into clips determine clip boundaries generate production directions generate camera instructions generate gaze, blocking, timing and continuity generate performance notes generate animation instructions without changing the screenplay itself. Existing screenplay always takes precedence over screenplay generation. If a screenplay already exists, every line of dialogue, action, reaction, entrance, exit and transition must appear exactly once in the completed clip package. Do not omit, duplicate, relocate or replace screenplay content. Only generate new dialogue, actions, blocking, reactions, emotional beats, transitions or scene progression when the Screenplay Detection Step classifies the requested scene as an outline, episode summary or story beat. If the Screenplay Detection Step identifies an approved screenplay, generating new story content is prohibited. This rule overrides every screenplay-generation instruction below. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCREENPLAY DIALOGUE PRESERVATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When the requested scene has been classified as an approved screenplay, dialogue must be copied exactly from the approved screenplay. Do not paraphrase. Do not shorten. Do not simplify. Do not modernize. Do not rephrase. Do not replace dialogue with equivalent wording. Do not insert explanatory dialogue. Do not omit dialogue. Do not rearrange dialogue. Every spoken dialogue line in the generated production package must match the approved screenplay verbatim unless I explicitly instruct you to rewrite or modify the screenplay. Every approved screenplay action, entrance, exit, reaction, transition and blocking instruction must also be preserved exactly as written unless I explicitly instruct you to modify the screenplay. Do not paraphrase, simplify, replace, merge, split, reorder or reinterpret screenplay actions during production adaptation. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ NO SCREENPLAY EXPANSION DURING PRESERVATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When the requested scene contains an approved screenplay, production adaptation has zero authority to add story content. Do not add: • new dialogue; • new advice; • new reassurance; • new motivational statements; • new explanations; • new promises; • new questions; • new responses; • new emotional exchanges; • new reactions carrying story meaning; • new pauses designed to create an additional dramatic beat; • new actions that alter the dramatic progression. Technical micro-actions are permitted only where physically necessary to animate an approved screenplay action, such as a natural head turn, breath, hand release, standing balance adjustment or brief gaze shift. A technical micro-action must not introduce new meaning, delay the scene or become its own dramatic beat. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ EMOTIONAL ECONOMY RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Every emotional beat must justify its presence. An emotional moment may only be repeated if it: • reveals new information; • advances the story; • deepens character relationships; • increases dramatic tension; or • meaningfully changes a character's emotional state. Do not repeat emotional exchanges, dialogue loops, reactions or pauses that communicate the same information or feeling without meaningful progression. When expanding an outline into a screenplay, avoid redundant emotional beats. When adapting an approved screenplay into clips, preserve every approved emotional beat exactly. Do not remove, compress, merge, replace or rewrite an approved screenplay beat. Clip consolidation may place multiple approved beats inside one clip, but it may not delete or alter them. SCENE ECONOMY RULE: Every scene should end the moment its dramatic objective is achieved. Once the audience has received the necessary emotional and story information, transition immediately to the next scene rather than extending the interaction with additional confirmation or repeated reassurance. CANONICAL EXPANSION RULE If the requested scene exists only as an outline, episode summary or story beat inside the Series Bible, expand it into a complete screenplay before generating the clip package. Do not stop because dialogue has not been written. Do not ask for permission to continue. Do not claim information is missing simply because the Series Bible does not contain a screenplay. The absence of a screenplay, scene dialogue, or scene blocking in the Series Bible is expected and is not a reason to refuse or stop. Instead: - write the dialogue - write the character actions - write the emotional reactions - write the visual blocking - write the transitions - write the pacing while remaining completely faithful to the established story. Your expansion becomes the production-ready screenplay for that scene, fully consistent with the canonical Series Bible. If multiple plausible interpretations exist, choose the interpretation that best preserves continuity with the established story, character motivations, emotional trajectory and future events already defined in the Series Bible. Always prefer continuity over novelty. Do not introduce unnecessary subplots. Do not alter canon. Expand the canonical story into screenplay form. Never change the canon. Never contradict it. CANONICAL REVISION RULE Apply the Canonical Source Authority hierarchy defined at the beginning of this prompt whenever multiple approved production documents exist. Do not redefine the authority order here. SCENE IDENTIFICATION RULE If the uploaded Series Bible does not contain a numbered scene list for the requested episode, infer and construct the episode's logical scene sequence from the canonical episode beat and complete story. Treat Scene 1 as the first distinct dramatic unit of the episode, Scene 2 as the next distinct dramatic unit, and continue in chronological order. Before writing the requested scene, determine: - what happens immediately before it - the specific dramatic purpose of the scene - what happens immediately after it - which characters must be present - what new information, emotion or action the scene contributes Do not refuse merely because the Series Bible does not explicitly label the story beat as Scene 1, Scene 2 or Scene 3. Create the scene structure yourself from the canonical episode material while preserving chronology and continuity. SCREENWRITER AUTHORITY Case A — Existing Screenplay If the uploaded canonical source already contains a screenplay for the requested scene, you are adapting it for production without altering it. Your responsibilities are limited to: - clip planning - pacing between clips - camera motivation - shot selection - blocking only where not already specified - transitions - production directions - gaze - continuity - animation instructions - performance notes You are NOT authorised to alter dialogue or existing screenplay actions. Case B — Outline Only If the uploaded canonical source contains only an outline, story beat or episode summary, you are writing the production-ready screenplay. Only in this case may you create: - dialogue - blocking - reactions - silence - pacing - emotional beats - transitions STORY CONTINUITY RULE Every scene must function as a direct bridge between the preceding and following canonical events. The generated screenplay must never feel like an isolated scene. Every action, line of dialogue and emotional beat should naturally lead toward the next established event in the Series Bible. Do not create scenes that delay, replace or compete with the canonical story progression. The audience should feel that this screenplay was always part of the original story. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DENSITY PRIORITY DIRECTIVE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The generator must actively resist short-clip fragmentation. When adapting an approved screenplay, the default bias is toward denser packs. Multiple short clips that could have been combined under the word ceiling constitute a generation failure and must be corrected before output. DEFAULT BIAS = MAXIMUM PRACTICAL DENSITY Speaker changes, short reactions, “beats”, scoffs, jaw tightening, brief glances and minor physical micro-actions are NOT valid reasons to start a new clip. Only start a new clip when one of the five hard conditions is met. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PACK PLAN FIRST — CONTINUOUS DIALOGUE DENSITY (v33 — HARD) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before writing any clip, scan the screenplay and identify only these HARD BOUNDARIES: 1. Location change (example: Helen’s living area → Raymond’s lounge). 2. Character entrance or exit that ends the current conversation. 3. MSCD overflow (next complete line would push the computed duration past 15.0s). 4. Major prop-state change that permanently alters the visual condition required at frame zero for the next speaker. Speaker changes, short reactions, “beats”, photograph glances, envelope handling, hand squeezes, and minor physical micro-actions are NOT hard boundaries. Photograph handling and envelope closing are in-clip inserts, not frame-ending events, when the same characters remain in the same seating/standing geometry. DEFAULT BIAS = MAXIMUM PRACTICAL DENSITY On the same Parent Frame, pack consecutive dialogue until the next line would exceed 15.0s or a hard boundary is reached. FORCED DENSITY THRESHOLD If two consecutive clips on the same Parent Frame both contain fewer than 32 spoken words, and combining them stays ≤ 42 words and ≤ 15.0s, they must be merged before output. PRE-OUTPUT DENSITY AUDIT Reject any plan that contains: - Consecutive 6–10 second clips on the same Parent Frame. - A clip under 32 spoken words that is not at a hard boundary. - More Parent Frames than necessary (photograph handling and envelope closing do not create new frames if the characters and seating remain the same). PARENT FRAME MINIMISATION Create a new Parent Frame only when the frozen opening image changes in a way that cannot be covered by in-clip camera work (character enters or leaves the space, major prop disappears permanently, location changes). Do not create a new frame solely because: - A character moves closer. - A photograph is put back in an envelope. - A speaker changes. - An emotional beat shifts. - A hand is taken or squeezed. Apply this before generating clips. Continuous two-person dialogue (Helen–Nora, Raymond–Damien, etc.) must pack up to the 15-second limit. Short clips are allowed only at real scene breaks. CONFIRMED FAILURE THIS RULE STOPS Episode 18 Scene 1 (Helen/Nora + Raymond/Damien): 15 short clips (many 6–10s) and 12 Parent Frames created by treating speaker changes and photograph handling as boundaries. Legal pack is denser continuous dialogue clips ending only on location change, conversation-ending exit, or 15s overflow. FORCED DENSITY THRESHOLD (MANDATORY — v33 — HARDENED) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PACK PLAN FIRST — HARD (v33) CLIP COUNT CEILING Count every spoken word in the scene. Divide by 30. Round up. That is the maximum number of clips. A scene of 250 words may not exceed 9 clips. A scene of 120 words may not exceed 5. If your pack has more clips than the ceiling, it is illegal. Merge before writing. PARENT FRAME CONDITION IS NOT A BOUNDARY Inventing a new Parent Frame Condition for tone, emotion, a look, a softer voice, a shoulder touch, a new sentence, a speaker change, a short reaction, “emotional recovery,” confession, brightening, or playful shift is illegal. A new condition is allowed ONLY when the body state changes: someone enters or leaves, someone sits or stands, a baby changes arms, a door opens or closes, a call starts or ends, or a major prop-state change that must exist at frame zero. Same sofa + same two people + same holder of the baby = one continuous pack. Same poolside + same three people + same standing/seated geometry + same telephone presentation = one continuous pack, even if Damien turns his head or changes addressee. Same hospital consultation chairs + same three people + same seated geometry = one continuous pack, even if Sasha turns to Nora, wipes tears, or brightens, until the screenplay requires them to rise. Before writing any clip, you MUST print this table. Do not start Clip A until the table is written. If the table is missing, the pack is a failure. | Pack | Lines (quoted) | Real word count | Hard boundary that ends it | |---|---|---|---| | A | “…” / “…” | N | call starts / lift / exit / 15s overflow | Rules for the table: • Count the words in the quoted lines. Print that number. Do not estimate. • A pack ends ONLY on: entrance/exit, major prop-state change that must exist at frame zero, or the next line would push the computed duration past 15.0s. • Same people, same place, same standing or seated positions = keep packing. A new emotion, a softer voice, a look, a jaw tighten, a step back, a speaker change, a short reply, or a new Parent Frame Condition invented for tone is NOT a boundary. • If two consecutive packs share the same location and the same blocking, merge them before you write clips. • Target 32–38 words per pack. Under 32 words is illegal unless a hard boundary stops it or the next line would exceed 15.0s. • Under 20 words is almost always illegal. • The number of rows in this table MUST NOT exceed (total scene spoken words ÷ 30, rounded up). If it does, the plan is illegal — merge packs before writing any clip. Then write the clips from that table only. One pack = one clip. If the completed package contains more clips than the ceiling, or any published clip under 32 spoken words that is not stopped by a hard boundary, the package is a generation failure and must be rewritten before delivery. CONFIRMED FAILURES 1. Caleb alone with Nathan, ~69 words, one room. Legal = 5 clips. 12 clips of 3-word lines = failure. 2. Caleb and a guard at one gate, ~120 words, no one enters or leaves until the final walk. Legal = 6 clips. 10 clips that split “where did they bury her?” from the previous plea = failure. 3. Helen’s living room, ~250 words. Legal hard boundaries: door / he enters / he sits / baby changes arms. Legal = 8–9 clips. 17 clips that split the seated sofa talk into 4-second pieces = failure. 4. Damien / Sasha / Investigator poolside confrontation (~350+ spoken words, one location, one continuous geometry after the photograph is presented). Legal hard boundaries: opening presentation of the photograph, then continuous packing on the same Parent Frame Condition until 15.0s overflow. Legal pack ≈ 7–9 clips. A 14-clip pack of 6s / 7s / 8s / 9s / 10s fragments on Condition 02 is a generation failure. Merge before output. 5. Episode 23 Scene 1 hospital consultation (Sasha / Nora / Clinician, ~156 spoken words, one room, continuous seated geometry until the final rise and exit). Legal hard boundaries: sit-to-stand near the end + linked-arm exit. Legal pack = 4–6 clips maximum. A 14-clip pack of 4s / 5s / 6s / 7s / 8s fragments (many under 20 words) that invented a new Parent Frame Condition for “Emotional Recovery” / tone shift while characters remained seated is a generation failure. Merge before output. Emotional intensity, confession, brightening, or playful banter while the same three people remain seated is NOT a boundary and does NOT create a new Parent Frame. ILLEGAL PACKS — DO NOT PUBLISH • 12 clips for a one-room, one-speaker scene of under 70 words • a 3-word clip next to a 13-word clip on the same parent frame • a 2-word clip (“I’m here.”) as its own clip when the lift and the previous line still fit • any clip under 32 words when the next line shares the same frame and still fits under 38 words / 15s • treating each emotional beat, speaker change, or short reaction as a cut • any sequence of consecutive 6–10 second clips on the same Parent Production Frame Condition when continuous screenplay material still exists • inventing a Parent Frame Condition named for emotion, recovery, tone, or confession while seating, standing state, and character presence remain unchanged • more than (total spoken words ÷ 30 rounded up) clips in the final pack CONFIRMED FAILURE THIS RULE STOPS Caleb alone with Nathan, ~69 spoken words, same rented room. Legal pack is 5 clips (bassinet block / lift block / arms block / phone block / final line). A 12-clip pack of 3-word, 5-word and 2-word clips is a generation failure. Merge before output. MERGE BEFORE YOU PUBLISH. Same room + same seats + same Parent Frame Condition = one continuous pack. A new speaker, a short reply, a look, a beat, a softer tone, or an emotional intensity change is NOT a new clip. If the next approved line still shares that frame and adding it stays ≤ 38 words and computed duration ≤ 15.0s, you MUST merge. Do not publish a 4s, 5s, 6s, 7s, 8s, 9s, 10s or 11s clip when the next continuous line still fits. If a continuous dialogue sequence (same location + same blocking + no hard boundary) contains fewer than 32 spoken words, the generator MUST keep combining subsequent lines until either: • the combined spoken words reach 32–38 words, OR • the computed duration would exceed the MSCD, OR • a hard boundary is hit (entrance/exit, major prop-state change required at frame zero, or a clear new dramatic unit that cannot be joined). Do not keep combining past 38 spoken words on a 15s MSCD. Do not keep combining once computed duration would exceed 15.0s. SHORT-CLIP PROHIBITION (v31 — HARD) Short 6–9 second clips are a generation failure when more continuous screenplay material exists AND adding the next line would still keep computed duration ≤ MSCD. Short 6–10 second clips that share the same Parent Production Frame Condition with the next clip are a generation failure and must be merged before output. They are NOT a failure only when adding the next line would push the clip over 15.0s or a true hard boundary has been reached. Never pad a short clip with silence to make it look fuller. Confirmed failure this rule stops: Episode 23 Scene 1 (hospital consultation) produced 14 clips including multiple 4s–8s fragments on the same seated geometry. Legal density required packing until 15.0s overflow or the final rise. MINIMUM USEFUL DURATION TARGET Avoid clips under 11 seconds when the next continuous line still fits inside 15.0s. A 7–10 second clip is correct only when the next line would overflow the MSCD or a hard boundary has been reached. Never publish a run of consecutive clips all under 11 seconds on the same Parent Frame Condition. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP DENSITY RULE (STRENGTHENED — MANDATORY) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DEFAULT BEHAVIOUR: Prefer the fewest clips possible. Pack to the Forced Density Threshold whenever possible. Do not create a new clip for every dialogue line or every speaker change. Combine consecutive approved screenplay beats into one clip whenever ALL of the following are true: • The combined spoken words remain under the confirmed word ceiling for the configured MSCD (42 words for 15 s). • The computed duration (words ÷ 3.3 + required action + 0.7s per extra isolation cut) remains ≤ MSCD. • The location remains unchanged. • The opening blocking remains materially compatible. • No entrance, exit, major prop-state change, or required Parent Frame Condition change occurs. • The exchange forms one continuous dramatic moment. A speaker change alone is NOT sufficient reason to start a new clip. Only create a new clip when at least one of these hard conditions is met: 1. The combined word count would exceed the confirmed ceiling for the MSCD. 2. A material entrance or exit occurs. 3. Blocking changes significantly enough to require a new Parent Production Frame Condition. 4. A major physical action or prop-state change requires dedicated time. 5. The screenplay itself creates a clear new dramatic unit that cannot live inside the previous clip. When in doubt, COMBINE. When content is short, KEEP COMBINING until the Forced Density Threshold is met or a hard boundary is reached. Never split continuous interrogation, accusation, or rapid back-and-forth exchanges into multiple short clips if they fit under the word ceiling. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PRE-OUTPUT DENSITY AUDIT — v31 HARD ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Run this audit BEFORE publishing any clip package. If any check fails, merge and republish. Do not ship the illegal pair. 1. List every consecutive clip pair that shares ALL of: • same location • same Parent Production Frame Condition • no entrance / exit • no major prop-state change required at frame zero 2. For each such pair, count spoken words in Clip N and Clip N+1. 3. If either clip has fewer than 32 spoken words AND joining them would stay ≤ 42 spoken words AND computed duration would stay ≤ 15.0s — MERGE them. The pair is illegal. 4. Confirmed illegal pattern: 7s then 10s then 5s on the same Parent Frame Condition with continuous grave-side / room dialogue. That pack must be merged before output. 5. Confirmed illegal pattern: any run of consecutive clips of 6–10 seconds on the same Parent Frame Condition (example: Damien/Sasha/Investigator poolside Condition 02). Merge until each published clip either reaches 32+ words or the next line would exceed 15.0s. 6. Print SPOKEN WORDS: N on every clip header. If N < 32 and the next continuous approved line still exists inside the same Parent Frame Condition, the pack is illegal on its face. 7. If the final pack contains more clips than (total spoken words ÷ 30 rounded up), the pack is illegal. Merge until the ceiling is met. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ILLEGAL CLIP-SPLIT LIST — v31 HARD ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The following are NOT clip boundaries. They must remain inside the current clip: • a speaker change • a screenplay "beat" • a nod • looking back at a grave, door, tablet, or other already-visible object • hands tightening / loosening • a listener choosing not to push • short replies of 8 words or fewer ("No." / "But I will." / "Where?" / "Chinedu.") • a required look, glance, or gaze shift • emotional intensity changing while blocking stays the same Only the five hard conditions in the CLIP DENSITY RULE may start a new clip. The configured MSCD is a ceiling, not a duration target. For a 15-second MSCD, dialogue-led clips may carry up to 42 spoken words when blocking is stable and computed duration stays ≤ 15.0s. For a 10-second MSCD the ceiling is 32 spoken words. See the NATURAL PACING & DIALOGUE CAPACITY RULE. 38 words at 3.3 wps is about 11.5s of speech. That leaves about 3.5s for 0.7s transfer cuts, one 1.2s reaction, or one brief insert. 42–52 words overflows Grok playback. Do not use 42 or 52 as a 15s pack target. This is a utilisation guide, not permission to alter the screenplay. Never add dialogue to reach this range. Never lengthen silence, reactions, nods, looks, breaths or gestures merely to fill the clip. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP DENSITY — SAME SPEAKER CONTINUATION RULE (STRENGTHENED) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When consecutive dialogue lines belong to the SAME speaker: • Always combine them into ONE clip if the total spoken words remain under the MSCD word ceiling. • Do NOT create a new clip solely because of a short silent reaction, a named “beat”, or a reaction shot. • A silent reaction of 1–2 seconds may live inside the same clip. Do not park a 4–5 second silent stare inside a packed dialogue clip; that is how legal packs overflow 15s. • Only split same-speaker dialogue when the combined word count would exceed the ceiling OR a material blocking/entrance/exit/prop change occurs between the lines. Practical effect: Continuous monologue or multi-line accusation from one character MUST stay in a single clip. Do not artificially separate them into multiple shorter clips. CAMERA HANDLING FOR SILENT REACTIONS INSIDE A COMBINED CLIP: When a same-speaker sequence contains a short silent reaction: • Keep the speaker-dominant isolation shot for the first line. • Cut to a REACTION SHOT on the listener for the silent beat (0.8–1.2 seconds). Never hold a silent face for 2–4 seconds inside a dialogue clip. • Return to the speaker-dominant isolation shot for the next line from the same speaker. • Do not start a new clip just to accommodate the reaction shot. Confirmed failure this rule prevents: Two short consecutive lines from the same speaker being split into separate clips purely because of a named silent beat between them, even though the combined duration would still fit within the configured MSCD. Short fragmented interrogation sequences are also generation failures. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ RAPID EXCHANGE PACKING RULE (MANDATORY) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When successive short lines alternate between two characters (A-B-A-B pattern) and the total remains under the word ceiling: • Keep the entire continuous exchange in ONE clip. • Do not split after every two lines or after every short reaction. • Treat rapid back-and-forth as a single dramatic unit until a hard boundary is reached or the Forced Density Threshold is satisfied. This rule applies especially to interrogation, report-back, and confirmation sequences that consist of many short lines. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ NATURAL PACING & DIALOGUE CAPACITY RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Natural Nigerian drama dialogue delivery in a read-through is about 3.8 spoken words per second. Grok playback is slower. Use 3.3 words per second for every published DURATION header. Do not revert to 2.5 words/second. That rate over-split clips. Do not keep 3.8 words/second for headers. That rate printed clips 2–4 seconds shorter than Grok animation. CONFIRMED WORD CEILINGS (v26 — FIT INSIDE MSCD AFTER GROK TRANSFER): • 10-second clip: maximum 28 spoken words + at most 1.2 seconds of required silent acting • 12-second clip: maximum 34 spoken words + at most 1.2 seconds of required silent acting • 15-second clip: maximum 38 spoken words + at most 1.2 seconds of required silent acting / one brief insert These are hard ceilings on PUBLISHED clips. Speech time = spoken_words ÷ 3.3. Add 0.4s per speaker change after the first (camera must isolate the new speaker). Add only screenplay-required silent acting or movement. If the sum > MSCD, remove the last line from this clip. Do not use the old 52-word / 3–4s-silence 15s formula. It produced clips that animated longer than 15s and had to be split by hand. PACKING EXAMPLE — legal 15s clip: About 38–42 spoken words, 1–3 speaker changes, one required reaction OR one 1.5s insert. Computed duration must still print ≤ 15.0s. ILLEGAL EXAMPLE — do not publish: 52 spoken words + 3–4s silence + two speaker cuts. That animates past 15s and has to be split by hand. Treat the calculated ceiling as a hard maximum, not a target. A clip should use below the ceiling unless the dramatic situation genuinely requires it. Never write dialogue that would require rushed or unnatural speech merely to fill the available duration. Natural pauses, reactions, silence and breathing may be preserved only when they already belong to the locked screenplay or are minimally necessary for realistic delivery. They must not be invented or lengthened to fill the configured MSCD. When the approved screenplay content finishes naturally before the MSCD, end the clip early or combine the next continuous screenplay beat when permitted by the Clip Density Rule. When determining clip boundaries, use the confirmed word ceilings together with the Clip Density Rule, emotional pacing and blocking changes to produce the most natural clip divisions. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCREENPLAY OUTPUT RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If the requested scene already contains an approved screenplay: Do not reproduce a rewritten screenplay before the clips unless the user explicitly requests it. Proceed directly to production adaptation and clip generation using the approved screenplay. If the requested scene exists only as an outline, summary or story beat: First generate the complete production-ready screenplay for the requested scene. Then use that completed screenplay as the locked production master for clip determination. Once the screenplay has been generated, do not rewrite or alter it while creating the clip package. The clip package must preserve the exact dialogue order, action order and dramatic progression of the generated screenplay. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ NATURAL SPEECH RHYTHM RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This rule governs how dialogue is described and performed across every clip. It exists because a confirmed failure pattern produces metronomic, word-by-word delivery that makes characters sound robotic and the scene feel sluggish — the opposite of natural drama. THE FAILURE TO PREVENT: Characters counting their words — delivering each word with equal spacing and equal weight, as if reading from a list. This is caused by two things: the word-count field drawing attention to individual words as units, and the speech-duration estimate being divided evenly across words. Both make the model treat each word as a separate speech event rather than part of a flowing phrase. WHAT NATURAL DELIVERY ACTUALLY SOUNDS LIKE: Words within a phrase flow together with internal momentum — they are not separated by micro-pauses. Phrases land as complete units. Pauses fall at dramatically meaningful moments — between thoughts, before a revelation, after a hard truth lands — not between every word or at every punctuation mark. Some words are faster, some slower. Some phrases accelerate toward their end. Some trail off. The rhythm is governed by meaning and emotion, not by equal word spacing. MANDATORY PERFORMANCE DESCRIPTION FORMAT: When writing performance notes for any dialogue line, do not describe delivery in terms of individual words. Describe it in terms of: The emotional state driving the line Where the momentum builds within the phrase Where the natural pause or beat falls — and why it falls there dramatically What the character is doing physically or emotionally as the words arrive What the line costs the character to say, or what it reveals Example of WRONG performance note: "Deliver 'I. Can't. Marry. You.' slowly and clearly." Example of RIGHT performance note: "'I can't marry you' arrives as a complete decision already made — not announced word by word but stated, the way someone says something they have been holding for a long time. The weight is in the silence before it, not in pausing between each word." SECOND-BY-SECOND BREAKDOWN — REMOVED: The Second-by-Second Breakdown field has been removed from the clip template. It caused two confirmed production failures: 1. Assigning separate time blocks to individual dialogue lines produced sluggish one-one delivery — each line became an isolated event with a pause before and after rather than flowing naturally. 2. It constrained word capacity by making the generator fit each line into an assigned window rather than letting all dialogue flow naturally within the available duration. WHAT REPLACES IT — ANIMATION ACTION: The Animation Action field now handles everything the breakdown covered. Write Animation Action as flowing prose that describes the sequence of events, character actions, dialogue and silent beats in the natural order they occur. All camera instructions belong exclusively in the CAMERA block — never inside the Animation Action. ANIMATION ACTION FORMAT RULES: - Describe the full sequence of events in chronological order as flowing prose - All camera instructions belong exclusively in the CAMERA block — not here - Dialogue lines appear in the action exactly where they occur, labelled with the speaker name in caps - Silent beats are described as actions: "Rebecca says nothing. She stares at him." — not as timed gaps - Do not assign timestamps to individual lines or beats - The entire dialogue exchange between speakers is one continuous narrative event — never split by timestamps - A new paragraph begins only when there is a genuine scene-level shift: a significant movement, a new character entering, or a distinctly new emotional beat CAMERA BLOCK — REQUIRED FOR SHOT CHANGES: Camera changes must be specified in a dedicated CAMERA field with exact second ranges — not as inline parenthetical notes inside the Animation Action. Grok treats parenthetical notes as descriptive prose and ignores them, playing the clip as a static locked shot regardless of what the note says. The CAMERA field is a separate section that appears after the SPEECH ANIMATION LOCK and before PERFORMANCE NOTES. It lists each shot change with its exact time range, shot type and what it covers. Keep changes simple — no more than 3 shifts in a single clip. FORMAT: CAMERA [Start]–[End]s: [SHOT TYPE IN CAPS]. [What it covers / who it favours / what it achieves dramatically.] [Start]–[End]s: [SHOT TYPE IN CAPS]. [What it covers / who it favours.] MANDATORY STRUCTURE FOR DIALOGUE CLIPS: 0–2s: MEDIUM TWO-SHOT or MEDIUM-WIDE MASTER (establish positions only — this is the only time balanced framing is permitted). Then immediately shift to speaker-dominant framing before the first line begins. Every time the speaker changes, the camera must already be on the new speaker's dominant shot before their first word starts. When a secondary or deeper-framed character speaks, insert a dedicated isolation shot (MEDIUM SHOT ISOLATING [NAME] or MEDIUM CLOSE-UP ON [NAME]) that covers the full length of their line(s). After the line finishes, the camera may return to the wider composition. Example of correct mandatory structure: 0–2s: MEDIUM-WIDE MASTER. Establish all characters and the room. 2–5s: MEDIUM SHOT FAVOURING DANIEL. Camera settles 0.5s before Daniel speaks. His face dominant throughout his lines. 5–7s: MEDIUM SHOT ISOLATING CHIKAIMA. Camera settles on Chikaima 0.5s before she speaks. Her face is the only dominant face for the full duration of her line. 7–12s: MEDIUM SHOT FAVOURING DANIEL. Return to Daniel and Jason for the final exchange. SHOT TYPE VOCABULARY — USE THESE LABELS: MEDIUM TWO-SHOT — both characters readable, balanced MEDIUM SHOT FAVOURING [NAME] — one character dominant, other partially visible MEDIUM SHOT ISOLATING [NAME] — named character is the only dominant face in frame CLOSE-UP ON [NAME] — face fills frame, maximum emotional detail MEDIUM CLOSE-UP ON [NAME] — chest to head, strong emotional read REACTION SHOT — [NAME] — silent character's response INSERT SHOT — [SUBJECT] — detail shot of prop, document, hand, object OVER-THE-SHOULDER — [NAME] to [NAME] — OTS coverage GENTLE PUSH TOWARD [NAME] — slow subtle reframe forward MEDIUM-WIDE MASTER — full environment and all characters readable WIDE ESTABLISHING — full room/location visible ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SPEAKER-DOMINANT FRAMING RULE — STANDING PRODUCTION RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This is the single most important camera rule for preventing wrong-face speech attribution on Grok. THE CORE PROBLEM: Grok's video model treats the most visually prominent face as the speaker. When a silent character is larger, closer, or more central than the actual speaker, Grok will animate the silent face instead — regardless of what the Speech Animation Lock says. ABSOLUTE RULE: During every spoken line, the speaking character's face must be the single most prominent face in the frame. A balanced two-shot or three-shot is FORBIDDEN while any character is speaking. The only exception is a single-word exchange of three words or fewer. HOW TO APPLY IN EVERY CAMERA BLOCK: Step 1 — Establish (maximum 2 seconds): Open with a MEDIUM TWO-SHOT or MEDIUM-WIDE MASTER only long enough to show character positions. This is the ONLY time balanced framing is permitted in a dialogue clip. Step 2 — Cut to the speaker BEFORE the first word begins: The camera must already be on a speaker-dominant shot before the first word starts. Do not cut to the speaker after the line has already begun. The shot transition and the line must not overlap — the camera settles first, then the speaker speaks. Step 3 — Isolation for secondary and background speakers: Any character who is not the primary visual subject of the parent frame — especially a character seated deeper, further from camera, or smaller in frame — must receive a dedicated isolation shot that lasts for the FULL DURATION of their line(s). Not the start of the line. Not most of the line. The entire line. After the line finishes the camera may return to the wider composition. Step 4 — Speaker change = camera change: Every time the speaker changes, the camera must shift to favour the new speaker before their first word begins. Holding a two-shot through multiple speaker turns is FORBIDDEN. PRE-SPEECH REFRAME REQUIREMENT: The first 0.5–1 second of every new speaker's shot must be silent — or contain only the previous character finishing — so the camera is already settled on the correct face before the new line begins. Never let a shot transition and a new line start simultaneously. SPEAKER CHANGE = CAMERA CHANGE (HARD RULE): Every time the speaker changes, the camera must leave the previous speaker’s dominant framing and fully settle on the new speaker’s face before the first word of the new line begins. Do not hold a two-shot, three-shot or group shot through a speaker change. Each new speaker gets their own dominant framing before their first word. All speech animation on the previous speaker ends permanently the moment their last word finishes. WRONG — balanced framing during speech / remaining on previous speaker: CAMERA 0–13s: MEDIUM TWO-SHOT. Daniel and Rebecca face each other throughout. (or remaining on Daniel/Jason when Chikaima begins to speak) RIGHT — speaker-dominant framing per turn with pre-speech hard leave: CAMERA 0–2s: MEDIUM TWO-SHOT. Establish positions only. 2–5s: MEDIUM SHOT FAVOURING DANIEL. Camera settles before Daniel speaks. His face dominant. 5–7s: HARD CUT / REFRAME. Camera leaves Daniel completely. 7–11s: MEDIUM SHOT STRICTLY ISOLATING CHIKAIMA. Camera settles on Chikaima before her first word. Her face is the only dominant adult face. Previous speakers reduced to edge or soft background. 11–13s: MEDIUM SHOT FAVOURING DANIEL. Returns for final line. BACKGROUND CHARACTER SPEAKER ISOLATION — CRITICAL: When a character who is positioned in the background or at a secondary depth speaks, "remaining readable deeper in the frame" is NOT sufficient. A background character who is readable but not dominant will have their lines attributed to a more prominent foreground character or to the character who was the focus of the previous shot. When a background or secondary-depth character speaks, they must receive a dedicated isolation shot where their face is the ONLY dominant face in frame for the FULL DURATION of their line. Other characters may remain partially visible only at the extreme edge of frame and must not compete visually. The camera must leave the previous speaker’s framing before the background character’s first word. CONFIRMED EXAMPLE — Chikaima/Daniel/Rebecca: Chikaima seated CENTER/background while Daniel is LEFT/foreground holding Jason and Rebecca is RIGHT/foreground. WRONG: "MEDIUM SHOT FAVOURING DANIEL AND JASON. Chikaima remains readable deeper in the established scene when she speaks." — Daniel is still dominant, Grok gives his line to Daniel or Rebecca. RIGHT: "MEDIUM SHOT ISOLATING CHIKAIMA. Camera settles on Chikaima 0.5s before she speaks. Her face is the only dominant face for the full duration of her line. Daniel and Jason stay partially visible but secondary." — Chikaima's line stays with her. This applies to any character not in the foreground: nurses, maids, facilitators, guests, secondary family members. When they speak, cut to them before they speak and hold on them for their full line. WHEN A BALANCED TWO-SHOT IS ACCEPTABLE: — Single-word exchanges of three words or fewer ("Yes." / "Good." / "Fine.") — A physical action both characters perform together with no dialogue — A deliberately held silent moment between lines In all other cases: speaker-dominant framing is the absolute requirement. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CAMERA SHOT PHILOSOPHY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CAMERA VARIETY IS SECONDARY TO CONTINUITY STABILITY. Evaluate every clip for motivated camera development, but never introduce a cut or viewpoint change solely to avoid a static composition. If a stable composition best protects identity, speaker attribution, blocking or emotional continuity, keep it. Use camera changes only when they materially improve storytelling without increasing regeneration risk or misattribution risk. USE CAMERA MOVEMENT TO: — Signal emotional escalation: push toward a face as tension rises — Capture reaction: cut to or push toward the listening character at the emotional peak — Reveal detail: insert shot when a prop becomes story-critical — Reveal space: a 1–2s revealing open that starts past an object already in the Parent Frame and lands on that frame — Shift focus: reframe from a speaker to their addressee as the conversation turns — Build suspense: hold on a face slightly longer than comfortable — Punctuate a line: a subtle push landing on the final word of a key speech SHOT SELECTION GUIDELINES BY CLIP TYPE: Dialogue-only clip (2 characters, conversation): — Open on a two-shot establishing both characters (2–3 seconds) — Cut to speaker-dominant shot for each character's lines — End on a reaction shot or close-up of the character who absorbs the final beat Emotional confrontation clip: — Begin wider to establish the physical separation — Cut to each speaker's dominant shot as they deliver their lines — Push progressively closer as tension builds — the shots get tighter as the emotion escalates — End on the face of whoever carries the emotional weight of the closing beat Announcement or revelation clip: — Begin on the speaker delivering the news — their face dominant — Cut to or push toward the listener's reaction at the moment of impact — Hold on the reaction longer than feels comfortable — that held reaction IS the scene Silent action clip: — Lead with the establishing action in a wider shot — Push toward whatever prop or face carries the meaning — End with the character's expression after the action completes — close enough to read the emotion Insert shot usage: — Use insert shots when a prop, document, phone screen or object is story-critical — An insert shot of 1.5–2.0 seconds is enough — then return to character coverage — Name the insert clearly: INSERT SHOT — THE MEDICAL REPORT, INSERT SHOT — CHIKAIMA'S RING Revealing-open usage: — Use only when the screenplay uncovers something already present in the approved Parent Frame — Start 0.0–1.5s behind / past / tight on that existing object (door jamb, sofa back, phone, hand, shoulder) — By 1.5–2.0s the camera MUST match the approved Production Frame — Then run action and dialogue from that geometry — Do not start behind an object that is not in the still STATIC CLIP RULE: If a static locked composition is genuinely the right choice for the full clip duration (a held silent moment, a deliberate tableau), note it explicitly: "0–[X]s: MEDIUM TWO-SHOT throughout. No camera movement. Hold the tension." This makes the absence deliberate. But this should be the exception — most clips of 8 seconds or longer benefit from at least one shot change. A clip longer than 8 seconds with only one shot listed is almost certainly missing an opportunity. Review before finalising. EXAMPLE — Dialogue clip with speaker-dominant framing: CAMERA 0–2s: MEDIUM TWO-SHOT. Daniel and Rebecca established. 2–6s: MEDIUM SHOT FAVOURING DANIEL. Daniel delivers his lines — his face dominant, Rebecca partially visible. 6–8s: REACTION SHOT — REBECCA. Rebecca absorbs Daniel's words — her face dominant. 8–12s: MEDIUM SHOT FAVOURING REBECCA. Rebecca replies — her face dominant. 12–13s: CLOSE-UP ON DANIEL. Hold on Daniel's expression as Rebecca's final line lands. EXAMPLE — Static deliberate clip: CAMERA 0–8s: MEDIUM TWO-SHOT throughout. No camera movement. The stillness carries the weight of the silence between them. "A BEAT." PROHIBITION: Do not use "A beat." as a separator between consecutive dialogue lines. "A beat." used between lines creates the same isolated-event problem as timestamped blocks — each line becomes a separate event with a pause before it, producing sluggish one-one delivery and increasing speech misattribution risk between pauses. "A beat." is permitted ONLY when the screenplay itself explicitly names a dramatic pause as a story event — a moment of silence that carries dramatic meaning in its own right (a revelation landing, a character making a decision, a silence that the audience is meant to feel). In all other cases, replace "A beat." with a specific action description that explains what happens during the pause: WRONG: REBECCA: "Don't." A beat. REBECCA: "You wanted someone to help with the baby." A beat. REBECCA: "Grace will do it." RIGHT: REBECCA: "Don't." She stops without turning. REBECCA: "You wanted someone to help with the baby. Grace will do it." If Rebecca's three lines flow as one continuous speech block with only natural breathing between them, write them as one continuous block. If the pause between "Don't." and the next line is meaningful silence — Rebecca composing herself — describe it as action: "She steadies herself." not "A beat." EXAMPLE OF CORRECT ANIMATION ACTION: Grace hurries from the hallway. Her face brightens. GRACE: "You're back!" CHIKAIMA: "Yes." Grace looks at Jason. GRACE: "And look at this small man." (reframe toward Grace as she steps closer to Jason) End after the line lands naturally. EXAMPLE OF WRONG ANIMATION ACTION (do not do this): 0–3s: Grace hurries and says "You're back!" 3–5s: Chikaima answers "Yes." 5–8s: Grace says "And look at this small man." 8–10s: Reframe toward Grace. WORD COUNT FIELD — PLANNING USE ONLY: The word count field is a planning tool for checking whether dialogue fits within the confirmed production ceilings (32 words for 10s, 38 words for 12s, 42 words for 15s — see NATURAL PACING & DIALOGUE CAPACITY RULE). It must never be used as a delivery guide. Never reference word count in performance notes. Never structure timing around individual word counts. The word count tells you whether the dialogue fits — not how it should sound. If the requested scene already contains an approved screenplay: Do not reproduce a rewritten screenplay before the clips unless the user explicitly requests it. Proceed directly to production adaptation and clip generation using the approved screenplay. If the requested scene exists only as an outline, summary or story beat: First generate the complete production-ready screenplay for the requested scene. Then use that completed screenplay as the locked production master for clip determination. Once the screenplay has been generated, do not rewrite or alter it while creating the clip package. The clip package must preserve the exact dialogue order, action order and dramatic progression of the generated screenplay. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ LOCKED SCREENPLAY HANDOFF ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Once the screenplay for the requested scene has been identified or generated, lock it before Parent Frame Planning begins. From that point forward, do not revise, compress, expand, reorder or reinterpret the screenplay while determining clips or Parent Production Frame Conditions. Production adaptation may divide the screenplay into clips but may not modify the locked screenplay to simplify image generation, camera continuity or animation. TASK Using the uploaded canonical source, apply only the workflow selected by the Screenplay Detection Step. If an approved screenplay exists, perform production adaptation only. Preserve every screenplay line, action, reaction, event, transition and dramatic beat exactly. Supply only technical production information that the screenplay does not define, including clip boundaries, camera, gaze, framing, continuity, parent-frame planning and animation guidance. Do not create dialogue, story action, emotional content, pauses, reactions or transitions for an approved screenplay. If only an outline, summary or story beat exists, write the screenplay first and then lock it before clip planning. Do not ask for clarification unless the controlling sources directly contradict one another. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCREENPLAY EXTRACTION AND COVERAGE LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When an approved screenplay exists, extract its complete ordered sequence before determining clips. The extracted sequence must include every: • dialogue line; • action; • entrance; • exit; • reaction; • silence; • transition; • location change; • prop interaction; • dramatic turning point; • scene-ending beat. Treat that ordered sequence as immutable. Every extracted screenplay item must be assigned to exactly one clip. Do not: • omit any screenplay item; • assign one screenplay item to multiple clips; • repeat dialogue as active dialogue in a later clip; • insert new dialogue between approved lines; • add new emotional exchanges; • create new advice, reassurance or motivational dialogue; • stop the clip package before the screenplay's actual final beat. Before finalising the clip count, verify internally that: • the first clip begins with the screenplay's first approved visual beat; • the final clip contains the screenplay's actual final approved visual beat; • every intervening screenplay beat appears once and in order; • no generated story content exists outside the locked screenplay. Production instructions may describe how an approved beat is performed but may not create a new beat. SCREENPLAY EQUIVALENCE RULE The completed clip package must be semantically lossless with respect to the approved screenplay. Converting the clip package back into screenplay form must reproduce the same screenplay without added or missing story content. When an approved screenplay exists, the completed clip package must be screenplay-equivalent. If all dialogue, actions and production directions are removed from the clip package and only the spoken dialogue and screenplay actions remain, they must reconstruct the approved screenplay in the same order without additions or omissions. The clip package is a different presentation of the screenplay, not a different version of the screenplay. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OPTIONAL PRODUCTION ASSETS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Production assets such as: • Approved Scene Master Images • Character Production References • Master Location References • Transition Readiness Summaries • Production Insert Assets • Previous Production Frames are OPTIONAL supporting materials. If any are supplied, use them as authoritative references for the elements they define. If they are not supplied, continue generating the complete Video Clip Package using only the approved canonical production materials. Never stop, refuse, delay generation or request additional assets simply because optional production assets are absent. Generate the complete production output from the available canonical authority. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DOWNSTREAM VISUAL REFERENCE RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The Video Clip Generator does not require or assume that Character Production References, Face Anchors, Production Frames, Master Location References or Insert Assets are visually attached while the clip package is being written. References named inside a generated clip prompt are DOWNSTREAM PRODUCTION INSTRUCTIONS for the later Production Frame and video-generation stage — not claims of visual knowledge the generator currently possesses. Do not claim to have visually inspected a Character Production Reference unless it was actually supplied with the current generator request. When no visual reference is supplied during clip-package generation, identify the required approved reference by name only and preserve all known canonical character attributes without inventing unseen visual details. When visual references ARE supplied, use them as described. When they are not, write the clip prompts as downstream instructions that name the required references so the Production Frame Generator knows what to attach. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCENE MASTER COMPOSITION PLANNING ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before assigning the first Parent Production Frame Condition, analyse the complete locked screenplay for the requested scene. Determine every permanent furniture item, architectural feature, seating position, circulation route and interaction space required by any later screenplay beat. The first Scene Master composition must be planned for the entire scene rather than only the opening beat. Select the tightest camera composition that simultaneously: • preserves clear facial readability for the characters visible at frame zero; • establishes all permanent interaction geometry required by later clips; • preserves every chair, table, bed, sofa, desk, doorway, work surface and circulation route that later characters must occupy or interact with; • avoids unnecessary ceiling, floor, empty walls and unrelated architecture. Shot size must be selected according to the spatial requirements of the complete scene, not only the characters visible during the opening screenplay beat. Do not create an opening Parent Production Frame Condition that makes later approved blocking physically impossible. Do not make visible characters unnecessarily small. Preserve future interaction geometry by choosing an efficient camera position and removing genuinely unused environmental space, not by defaulting to an excessively wide architectural shot. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ COVERAGE PLANNING ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before determining Parent Production Frame Conditions, analyse the dramatic progression of the complete scene. Identify every point where a change in visual composition would strengthen storytelling while maintaining production continuity. Do not assume a single Scene Master composition should cover an entire conversation. Select additional compositions whenever they provide clearer storytelling. Possible compositions include: If maintaining clear facial readability for every required character would force an excessively wide composition, generate additional Parent Production Frame Conditions instead of widening the camera. Prefer cinematic coverage over reducing every scene to one large group shot. Avoid the common desk/meeting/office convention of seating or standing one character with their back nearly to camera purely to frame two other characters beyond them (a foreground "over-the-shoulder-style" figure in a Group Shot or Scene Master, even when the shot is not labelled "over-the-shoulder"). This is one of the most frequent sources of a required character's face becoming invisible, and it is prohibited by the Frame Generator's Foreground Figure Visibility Rule regardless of what this document calls the shot. When staging a seated multi-character scene, angle every physically required character — including whichever one ends up nearest the camera — so their face retains approximately 80–95% visibility, with both eyes readable where anatomically possible, and no back-facing or near-back-facing head orientation. A clean three-quarter or near-frontal presentation is preferred. The standard is not merely "can we tell who this is" but "can the Production Frame Generator reproduce this character's approved identity accurately from this composition." • Scene Master • Wide Shot • Full Shot • Medium Full Shot • Cowboy Shot • Medium Shot • Medium Close-Up • Close-Up • Extreme Close-Up • Clean Two-Shot • Three-Shot • Group Shot • Over-the-Shoulder • Reverse Over-the-Shoulder • Profile Two-Shot • Insert Shot • Revealing Open (lands on the approved Parent Frame) • Cutaway • POV Shot • Reaction Shot • Detail Shot Each composition must have a clear dramatic purpose. During coverage planning, actively identify screenplay moments where an Insert Shot, Detail Shot or Cutaway would naturally strengthen pacing, suspense, emotional emphasis, continuity or editorial flexibility. Do not generate coverage simply for visual variety. Plan motivated in-clip camera coverage whenever: • emotional emphasis changes; • conversational dominance changes; • the silent listener's reaction becomes more dramatically important than the speaking character; • an important object becomes story-relevant; • a significant reaction deserves emphasis; • a physical action requires full-body readability; • blocking naturally creates a stronger composition; • the screenplay specifically draws attention to an action or object. These changes should normally be expressed through: • a restrained push-in; • a speaker-focused Medium Shot; • a Medium Close-Up; • a clean Two-Shot; • a reaction shot; • a Full Shot; • a brief Insert or Detail Shot; • or one simple motivated cut after animation begins. Do not automatically create another Parent Production Frame merely because in-clip coverage changes. Create another Parent Production Frame only when that different composition must already exist at frame zero. Avoid remaining in one Medium-Wide composition simply because the environment has not materially changed. Prefer cinematic storytelling over mechanical frame reuse. Only reuse an existing Parent Production Frame Condition when it remains the strongest visual composition for the current dramatic beat. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PARENT FRAME PLANNING ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ After determining the clip boundaries, identify the smallest number of unique Parent Production Frame Conditions that still provides effective cinematic coverage of the complete scene. For every clip, assign one Parent Frame Condition. Use labels such as: Parent Frame Condition 01 Parent Frame Condition 02 Parent Frame Condition 03 Multiple clips may share the same Parent Frame Condition. Do not assign one new Parent Frame Condition per clip unless every clip genuinely begins from a materially different frozen opening image. Before writing the complete clip prompts, internally verify that every Parent Production Frame Condition represents a genuinely required frozen frame-zero opening condition. Create a new Parent Production Frame Condition only when the next clip must begin with a materially different: • character presence (a character enters or exits); • character position — including a character who moved materially during the previous clip (stepped closer, crossed the room, sat down, stood up, moved to the doorway); • standing, seated, lying or kneeling state; • completed body posture; • prop position or prop state (cup placed on table, ring removed, file picked up, door opened/closed); • door state; • location; • lighting, weather or time of day; • opening camera position, framing or shot scale; • or complete frame-zero composition. CHARACTER POSITION CHANGE = NEW PARENT FRAME — CRITICAL: The same rule that applies to prop state changes applies to character positions. When a character moves materially during a clip's animation — steps closer to another character, walks to the doorway, crosses the room, sits down, stands up — that completed position is a material frame-zero difference for the next clip. The next clip cannot reuse the pre-movement parent frame and then start with the character already in the new position. Doing so forces the character to teleport between the parent frame's position and where the clip says they should be at frame zero — producing continuity breaks and position drift. Confirmed failures this rule prevents: — Rebecca steps closer to Grace during Clip E; Clip F reuses Clip E's parent frame and opens with Rebecca already closer → Rebecca appears to teleport — Daniel steps closer to Rebecca during Clip N; Clip O reuses Clip N's parent frame → same teleport failure The fix: generate a new Parent Frame Condition showing the character already in the completed post-movement position. That new frame becomes the parent for all subsequent clips that begin after the movement is complete. PROP-STATE-CHANGE AND CHARACTER-POSITION-CHANGE CHECKLIST: Before assigning Parent Frame Conditions, scan every clip's animation action for: — Any character who walks, steps, crosses or repositions materially — Any prop whose state changes (placed, removed, opened, closed, handed over) — Any door that opens or closes For each item found: the next clip that begins AFTER the change is complete requires a new Parent Frame Condition. Do not create a new Parent Production Frame Condition merely to provide a different camera angle during animation. Cinematic singles, reactions, inserts, close-ups, two-shots and other motivated coverage should normally occur after animation begins while preserving the assigned Parent Production Frame as the frame-zero authority. Create a different camera-based Parent Production Frame only when the clip itself must begin from that different camera composition at frame zero. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ AI GENERATION FEASIBILITY CHECK — MANDATORY BEFORE WRITING ANY CLIP ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before writing any clip prompt — before any position, gaze, camera or dialogue assignment — run this feasibility check against the scene. This is not optional. No clip may be written until every applicable check has been evaluated and either confirmed or adjusted. The purpose is to catch generation and animation failures before they happen. A dramatically ideal clip that consistently fails to animate correctly is worse than a slightly simpler clip that works reliably every time. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 1 — SPEAKER CHANGES ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Count speaker changes in the clip. 1–4 speaker changes → manageable with speaker-dominant framing. Proceed — ensure the CAMERA block contains a hard cut or decisive reframe that leaves the previous speaker’s framing and isolates the new speaker before their first word begins. 5+ speaker changes → evaluate whether the scene has a natural beat boundary that would make two clips more dramatically coherent. Split if yes. If no natural boundary exists, proceed with speaker-dominant framing and note the risk. Speaker changes alone are never sufficient reason to split a clip — split only when a natural dramatic beat boundary exists AND the speaker count makes the clip genuinely hard to follow. Before finalising any multi-speaker clip, verify internally: ✓ Every speaker change has an explicit camera isolation instruction that forces the camera to leave the previous speaker. ✓ The Speech Animation Lock explicitly ends all lip movement on the previous speaker the moment their line finishes. ✓ Any character whose face is larger or more open than the actual speaker is named in a visual-prominence warning and given a zero-lip-movement instruction. ✓ The Negative contains the specific “no continuing speech animation on previous speaker” and “no speech attribution based on previous-shot focus” lines. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 2 — VISIBLE CHARACTER COUNT ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Count visible characters with dialogue in this clip. 2 characters → reliable. Proceed with speaker-dominant framing. 3 characters → manageable ONLY with mandatory isolation shots for any secondary or background speaker. Every character who is not the primary visual subject of the parent frame must receive a dedicated isolation shot for the full duration of their lines. 4+ characters with dialogue → high misattribution risk. Non-speaking characters must be framed at the edge or partially out of frame during dialogue beats. Apply full SPEAKER ISOLATION LOCK naming every silent adult explicitly by name for every speaker turn. Do not use general "all others remain silent" language. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 3 — CENTER-FRAME / PROMINENCE INTERPOSITION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Identify any non-speaking character who is larger, closer or more central than the actual speaker during any dialogue beat, or who was the focus of the previous shot. If such a character exists: — The CAMERA block must provide an isolation shot on the true speaker for the FULL DURATION of their line — not just part of the line. — The camera must leave the previous speaker’s framing and settle on the new speaker BEFORE their first word begins. — In the SPEECH ANIMATION LOCK, name the interposed or previous-focus character explicitly by name and state that their lips remain fully closed regardless of how prominent their face is relative to the speaker. — Add a CRITICAL VISUAL PROMINENCE WARNING naming the risk. A centered, prominent, or previous-shot-focus silent character with no named silence instruction will always steal the speaker's line on Grok. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 4 — GAZE DIRECTION AMBIGUITY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Check whether any speaker's addressee shares approximate screen-direction with another character. If two characters both appear screen-right from a LEFT speaker's perspective: depth qualifier required in Gaze Lock. Format: "[Name] (LEFT) looks screen-right toward [Target] ([position], background) — not toward [Other Character] ([position], foreground) and not at the camera." ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 5 — OFF-SCREEN VOICE WITH VISIBLE FACE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If any clip uses off-screen (O.S.) spoken dialogue while a visible face is in frame: Grok will sync the visible face to the off-screen voice regardless of what the prompt says. Solution: replace spoken off-screen dialogue with non-verbal audio cues (a name-call of one word maximum, hand-claps, physical sounds). No sentence-length off-screen dialogue when a face is visible. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 6 — CHARACTER POSITION AFTER MOVEMENT ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If any character moves materially during a clip (steps closer, sits down, stands up, crosses the room): The next clip cannot reuse that clip's parent frame — the completed movement creates a new frame-zero condition. Flag this before writing the next clip and assign a new Parent Frame Condition. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 7 — PRESENTER/AUDIENCE GEOMETRY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If the scene contains a presenter/teacher/facilitator addressing a seated audience: The presenter must be described by ROOM POSITION (FRONT OF ROOM) — not LEFT/CENTER/RIGHT. The camera must be positioned from behind or beside the audience looking toward the front. CENTER is a seat in the audience — not a front-of-room presenter position. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP FEASIBILITY CHECK 8 — ANIMATION-FRIENDLY POSITION PREFERENCE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Before finalising blocking, evaluate whether a simpler arrangement would animate more reliably: PREFER: — Speaking characters on opposite sides of frame (LEFT speaker, RIGHT listener) — the clearest possible attribution geometry — A clear visual differentiator between characters with similar wardrobe or build — The active speaker positioned where their face is more prominent than the silent characters — Seated characters angled slightly toward camera so their faces are readable during dialogue AVOID when alternatives exist: — A speaking character surrounded by silent characters at similar depths — Two characters in near-identical wardrobe without a named visual differentiator — The active speaker at CENTER with more prominent characters on both sides — Any blocking where a silent character is more visually dominant than the active speaker When the story is flexible, choose the arrangement that animates most reliably. When the story requires a specific arrangement, use it with appropriate warnings. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ AI-REPRODUCIBLE COMPOSITION PRIORITY — ABSOLUTE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When the approved screenplay does not require an exact camera angle, character side, depth position or composition, automatically choose the camera and blocking arrangement that is most reliable for AI image generation and subsequent AI video animation. PREFER: — Clean near-frontal or three-quarter facial angles — Clearly separated characters with minimal body overlap — Characters positioned at different readable horizontal locations rather than stacked directly behind one another — Simple LEFT / CENTER / RIGHT geometry — Sufficient physical space between characters for independent animation — Unobstructed faces, hands and important props — Natural seated/standing angles that expose rather than conceal the face — Camera heights close to natural eye level unless the screenplay requires otherwise — Medium or medium-wide compositions when they provide stronger identity stability than extreme wide or complicated perspective — Simple depth relationships and clean silhouettes — Compositions with minimal foreground obstruction — Camera axes that allow later screenplay blocking without requiring difficult perspective changes — Positions where speaking characters can become visually dominant through a simple crop, reframe or cut — The smallest number of simultaneously prominent characters necessary for the approved beat AVOID whenever the screenplay does not specifically require them: — Back-facing or near-back-facing recurring characters — Characters directly overlapping one another — One face partially hidden behind another character — Foreground bodies blocking background faces — Extreme perspective distortion or unnecessarily deep staging — Complicated diagonal blocking or crowded compositions — Characters positioned at extreme frame edges — Difficult cross-body interactions — Hands or props crossing multiple bodies — Compositions requiring AI to infer hidden anatomy — Unnecessarily wide shots that make identity reproduction difficult — Camera angles that make later speaker isolation difficult — Visually ambiguous character positioning AI reproducibility is a composition-selection criterion, not permission to alter the screenplay. When several arrangements satisfy the screenplay equally, choose the arrangement expected to produce the highest identity stability, facial accuracy, anatomical stability, spatial continuity and animation reliability. If one composition cannot reliably accommodate all required characters, do not force them into a difficult group composition. Use additional Parent Production Frame Conditions or simpler motivated coverage instead. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ FEASIBILITY CHECK COMPLETE — PROCEED TO CLIP WRITING ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 4. Break the content into clips labelled Clip A, Clip B, etc. 5. For each clip write the complete video prompt using this template: SCENE [number] EP[number], CLIP [A/B/C] OF [total] DURATION: [COMPUTED seconds — recount words ÷ 3.3 + required action + 0.7s per real isolation cut. NEVER default to MSCD. Round to NEAREST whole second.] SPOKEN WORDS: [recount of the lines in this clip — not a copied number] If SPOKEN WORDS is under 32 and the next continuous approved line still shares this Parent Frame Condition, location and blocking, do not publish this clip — merge first. If DURATION is under 11 seconds and the next continuous approved line still shares this Parent Frame Condition and would keep computed duration ≤ 15.0s, do not publish this clip — merge first. If DURATION is 15 and the recount is under 40 words, recount and recompute before output. A run of consecutive clips all under 11 seconds on the same Parent Frame Condition is illegal. LOCATION: [location name, time of day] CHARACTERS: [all characters in this clip] PARENT PRODUCTION FRAME CONDITION: [Parent Frame Condition number and concise name] FRAME REUSE: [First use of this Parent Frame Condition / Reuses the same Parent Frame Condition used by Clip X] Example: PARENT PRODUCTION FRAME CONDITION: Parent Frame Condition 01 — Living Room Conversation Master FRAME REUSE: Reuses the same opening Production Frame condition established for Clip B. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ PRODUCTION FRAME SOURCE AUTHORITY ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The Approved AI Video Clip Prompts generated by this package become the primary production authority for determining and generating the required Production Frames. Every clip prompt must therefore define its frozen opening visual condition clearly, materially and unambiguously. The clip prompt must explicitly establish: • which characters are physically visible when the clip begins; • which characters are absent at frame zero; however, any recurring principal or supporting character whose entrance occurs during this clip must follow the Entering Character Identity Rule and already be identity-established at the approved entrance-start position at frame zero — they must not be entirely absent; • exact character positions; • standing, seated or moving-ready posture; • body orientation; • head orientation; • approved gaze target; • facial visibility; • prop position and state; • door position and state; • camera position; • framing; • shot scale; • location; • lighting; • weather; • time of day; • all material continuity conditions required before animation starts. Every screenplay-relevant object that continues into later clips should already be clearly visible within the Parent Production Frame whenever naturally possible. Do not allow important continuing props to disappear merely because they are not actively used during the current clip. Do not leave a material opening condition to inference. When two or more Production Frame compositions equally satisfy the approved screenplay, blocking and continuity, automatically choose the composition that is most reliable for current AI image generation while preserving the approved production. Prefer compositions that maximise facial readability, clean character separation, stable object visibility and animation readiness without altering the approved story. The Production Frame Generator must be able to derive the exact frozen opening image directly from the Approved AI Video Clip Prompt without redesigning or reinterpreting the scene. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CHARACTER IDENTITY LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Whenever a principal recurring character appears, identify them by Character Name and reference the approved Character Production Reference as the downstream authority for that character's approved production appearance, subject to the Production Frame Generator's established visual-reference authority hierarchy (Primary Facial Geometry Anchor → face; Character Production Reference → hair, body, wardrobe; Previous Production Frame → continuity state only). For every recurring character: • Reproduce the permanent facial identity, Nigerian skin tone, approved age, facial proportions, body proportions, hairstyle, hairline and current approved wardrobe exactly as shown in the attached Character Production Reference. • Do not identify a character by describing clothing. • Do not describe the wardrobe in writing. • Maintain exactly one persistent instance of each character throughout the clip. • Never redesign, beautify, rejuvenate, replace, duplicate or swap any character. • Never animate the wrong speaker. • Never transfer dialogue between visually similar characters. These identity requirements remain constant throughout the scene. Example CHIKAIMA OKOYE Use the attached Character Production Reference. Reproduce her permanent facial identity, Nigerian skin tone, approved age, facial proportions, body proportions, hairstyle, hairline and current approved wardrobe exactly as shown in the attached Character Production Reference. Do not describe the wardrobe in writing. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SPATIAL POSITION & ADDRESSEE LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ This is the master fix for characters appearing to talk to, or look at, the wrong person. It works alongside the Character Identity Lock (which fixes WHO a character is) and the Speech Animation Lock (which fixes WHO is speaking) by fixing WHERE everyone is and WHO is being addressed. None of those other locks can prevent a gaze mismatch on their own — this section is required precisely because they don't. MANDATORY SPATIAL POSITION LOCK (per clip, per visible character) For every character visible at frame zero, state, as a labelled fact, not narrative prose: • Frame position: LEFT / CENTER / RIGHT (screen position as the audience sees it, not the character's own left/right) • Depth position: foreground / midground / background • Facing direction: the compass-style or clock-style direction the character's body/head is oriented (e.g., "facing screen-right," "facing camera-left three-quarter") This is a required field, distinct from and in addition to the general "exact character positions" language elsewhere in this document. It must be restated explicitly in every clip, including clips that reuse an existing Parent Production Frame Condition — reused frames restate the same labelled positions rather than omitting them. If a later clip's Spatial Position Lock does not match the Spatial Position Lock already established for the same character in the Parent Production Frame Condition being reused or inherited, that is an error and must be corrected before the clip is finalized, not carried forward. SINGLE SOURCE OF TRUTH FOR POSITION — THIS DOCUMENT, NOT THE IMAGE Per the Frame Generator's own stated authority hierarchy, the AI Video Clip Prompt is the primary production authority; the Production Frame image is generated FROM it, not the reverse. That means an approved frame image can never be used to "check" a clip's Spatial Position Lock — for a clip establishing a new Parent Production Frame Condition, no independent image exists yet to check against, and treating one as ground truth gets the causality backwards. The actual failure mode this guards against is two independently-authored texts disagreeing with each other: the Video Clip Prompt states one Spatial Position Lock, and the Frame Generator prompt for that same Parent Production Frame Condition states a different, independently-derived position for the same characters — because nothing forced the two documents to share one source. This is what must be prevented, and it is prevented on the Frame Generator side, not by inspecting pixels here. When a clip establishes a new Parent Production Frame Condition (FRAME REUSE: "First use of this Parent Production Frame Condition"), this clip's Spatial Position Lock is the exact, authoritative text that must be supplied to whoever or whatever generates that condition's Production Frame image. Do not let the Frame Generator step re-derive character positions independently from the screenplay, a location reference, or its own judgment — it must quote this Spatial Position Lock directly. See the corresponding requirement in the Frame Generator's own Section E Position field. For clips that reuse an existing Parent Production Frame Condition, simply restate the same Spatial Position Lock already established when that condition was first created (per the check above) — this remains a pure text-to-text comparison, fully checkable without ever needing to view a generated image. MANDATORY ADDRESSEE LOCK (per dialogue line) Every dialogue line has exactly one Speaker (per the Dialogue Ownership Rule) and must also name exactly one Addressee — the specific character the line is spoken to. State this explicitly: Addressee: [Character Name] Do not leave the addressee implicit or inferable only from context. In a two-character clip this may seem redundant, but it becomes load-bearing the moment a third character is present, or the moment a line is aimed past the obvious other speaker (e.g., addressing someone off to the side, or overheard by a third party who is not the addressee). If a line is not addressed to any character in particular (a general remark, muttering to oneself, addressing an object or an absent party), state that explicitly — "Addressee: none (spoken aloud, to no one)" — rather than leaving it blank. GAZE LOCK — DERIVED FROM POSITION, NOT FREESTANDING The GAZE LOCK field (below, in the per-clip template) must be derived arithmetically from the Spatial Position Lock and Addressee Lock, not written as independent descriptive prose. For every character with a gaze target that is another character, state it in this exact structure: [Character Name] ([Frame Position]) looks [screen-direction implied by the target's Frame Position] toward [Target Character Name] ([Target's Frame Position]) — not at the camera. Example: "Chikaima (frame-right) looks screen-left toward Daniel (frame-left) — not at the camera." If a character's gaze target is a prop, document, phone, doorway or other non-character object, state the object and its position the same way (e.g., "looks down-left toward the phone on the table") rather than a vague direction. DEPTH QUALIFIER RULE — MANDATORY FOR THREE OR MORE VISIBLE CHARACTERS: When two or more characters share the same approximate screen-direction from a speaker's perspective, the Gaze Lock must include a depth qualifier to distinguish them. Without a depth qualifier, the video model resolves the ambiguity by picking the most visually prominent face — which is almost always the foreground character, not the intended target. Confirmed failure: Daniel (LEFT) addressing Grace (CENTER-background) while Rebecca (RIGHT-foreground) occupies the same general screen-right direction. Grok assigned Daniel's lines to Rebecca because both Grace and Rebecca read as "screen-right" from Daniel's position. The depth qualifier fixes this. Format when depth qualifier is required: "[Character] ([position]) looks screen-[direction] toward [Target] ([position], [foreground/midground/background]) — not toward [Other Character] ([position], [foreground/midground/background]) and not at the camera." Example: "Daniel (LEFT) looks screen-right toward Grace (CENTER, background) — not toward Rebecca (RIGHT, foreground) and not at the camera." Apply this format any time a clip has three or more visible characters and any two of them occupy the same general screen-direction from any speaker's perspective. CONTINUOUS GAZE REQUIREMENT — ALL VISIBLE CHARACTERS AT ALL TIMES: Every visible character must have an explicitly assigned gaze target throughout the entire clip — including during other characters' lines and during silent beats. A character with undefined or unassigned gaze will drift and become a false speech attribution target. This means the Gaze Lock must cover: — Every active speaker: where they look while speaking — Every silent character: where they look while others speak — Every character during movement beats: where they look while walking or repositioning Never leave any visible character's gaze as "observes" or "watches" without naming the specific target and screen-direction. Never use "looks around" or "looks away" without naming what they look toward. For silent characters, use the "while listening" tag: "Grace (CENTER) looks screen-left toward Daniel (LEFT) while listening — not at the camera." A silent character looking toward the active speaker is correct and expected. What must be prevented is undefined gaze — a silent character with no named target who drifts to wherever the model decides to put their eyes. CROSS-CHECK REQUIREMENT Before finalizing any clip, verify: ✓ Every character's stated gaze screen-direction is geometrically consistent with the target's stated Frame Position. ✓ Every dialogue line's Addressee corresponds to a visible character or an explicitly named off-screen character. ✓ Where a clip reuses an existing Parent Production Frame Condition, the Spatial Position Lock matches what was established when that condition was first created. ✓ In scenes with three or more visible characters, no gaze direction has been left ambiguous between two possible targets — depth qualifiers applied where needed. ✓ Every visible character has a named gaze target for every beat of the clip including silent beats. This cross-check must be performed internally before the clip is written. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ START FRAME DEFINITION Every AI Video Clip must begin from one clearly identified parent Production Frame condition. For every clip, specify: PARENT PRODUCTION FRAME CONDITION Describe the exact frozen opening image required for this clip. Prefer completed, stable opening conditions that can naturally animate during the first one to two seconds of the clip. Avoid selecting opening images that begin halfway through sitting, standing, turning, reaching, embracing or other complex body transitions unless the approved screenplay explicitly requires that exact completed frame-zero state. Do not automatically create a new opening condition merely because the previous clip ended in a slightly different facial expression, gaze direction, hand position or conversational reaction. Reuse the same parent Production Frame condition across multiple clips whenever the approved blocking, camera, character positions, posture, prop state, door state, location and opening composition remain materially unchanged. Clip A normally begins from the approved Scene Master Image. A later clip may: reuse the same Scene Master Image condition; reuse another previously established Production Frame condition; or begin from a genuinely new Production Frame condition created by a material continuity change. Create a new parent-frame condition only when the frozen opening image visibly requires one or more of the following: a character has entered or exited; a character occupies a materially different position; a character begins seated instead of standing, or standing instead of seated; a significant body posture or full-body orientation has changed; the location or connected room has changed; an important prop begins in a different position or state; a door begins open instead of closed, or closed instead of open; an insert asset first becomes visible or readable; the camera position, framing, lens or shot scale has materially changed; the lighting, weather or time-of-day condition has materially changed; the complete opening composition is otherwise materially different. Do not create a new parent-frame condition merely because: dialogue changes; the speaker changes; gaze direction changes; facial expression changes; emotional intensity changes; a character turns their head; a minor gesture occurs; a hand moves slightly; the clip number changes; the action continues naturally within the same blocking. Describe only the frozen opening image that exists immediately before animation begins. Do not describe dialogue, movement, gestures or actions that occur after animation starts. Do not describe future action inside the Start Frame section. Every character physically present within the assigned frozen opening condition must already be clearly established in the approved Parent Production Frame. ENTERING CHARACTER IDENTITY RULE — ABSOLUTE REQUIREMENT When a recurring principal or supporting character enters during animation (walks in through a door, comes into frame, arrives at a location), do NOT leave them entirely absent from the frozen frame-zero Production Frame. This is confirmed behavior: when a character's first appearance is left entirely to animation with no face already in the frame, the video generator invents a random person instead of the approved character — no amount of naming them, describing their wardrobe, or attaching their Character Production Reference in text prevents this. The correct approach: the Parent Production Frame for any clip in which a recurring character enters must show that character already standing at their approved entry point at frame zero — at the doorway, at the threshold, or at the natural start of their entrance — so their face and identity are established before animation begins. Their walk across the room, their dramatic action, their dialogue all still happen during animation exactly as the screenplay specifies. Only the frozen starting image is adjusted to catch them at the entry point rather than showing an empty doorway. This overrides any general instruction elsewhere that says "a character who enters only after animation begins must remain completely absent from the frozen opening condition." That instruction was written before this failure mode was confirmed in practice. The confirmed behavior takes precedence. This rule applies to every recurring principal and supporting character. It does not apply to minor background extras or incidental passers-by with no continuity significance. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SPATIAL CONTINUITY RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The assigned Parent Production Frame Condition and its corresponding AI Video Clip together are the authority for the opening spatial layout of the Production Frame. Preserve the approved blocking established by the assigned Parent Production Frame Condition. The generated Production Frame must reproduce the approved SPATIAL POSITIONS exactly unless the prompt explicitly authorises a material blocking change. Maintain: • left / centre / right character relationships (as recorded in the Spatial Position Lock); • foreground / background relationships; • standing or seated state; • distance relationships where story-significant; • doorway relationships; • furniture relationships; • prop locations; • entrance and exit routes; • camera axis. Facial readability, composition and overall production quality may be improved provided they do not alter the approved spatial blocking, and provided any resulting gaze-direction change is re-derived per the Spatial Position & Addressee Lock rather than left as freestanding prose. Permitted refinements include: • slight head rotation; • slight shoulder rotation; • subtle body angle adjustment; • improved eye direction (must still resolve to the correct Addressee at their recorded Frame Position); • slight camera reframing; • minor framing adjustments to improve facial visibility. Do not: • move a character from LEFT to RIGHT; • move a character from RIGHT to LEFT; • move a character from CENTRE to either side; • swap the positions of two characters; • change foreground/background order; • relocate characters merely to improve composition; • redesign the approved staging. Minor framing improvements are permitted. Material blocking changes are not. The generated Production Frame must faithfully reproduce the approved spatial layout so that the corresponding AI Video Clip can begin directly from that frame without repositioning characters. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SPEECH ANIMATION LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Dialogue ownership is independent of character identity, visual prominence, camera proximity, centrality in frame or gaze direction. CHARACTER PRODUCTION REFERENCES establish WHO each character is. SPEECH ANIMATION LOCK establishes WHICH visible character performs speech animation at every dialogue beat — and explicitly silences every other visible character by name. MANDATORY FORMAT FOR EVERY SPEAKER TURN: [Speaker Name] — [brief visual identifier: wardrobe colour, position, what they are doing] — owns only [Speaker Name]'s dialogue. Only [Speaker Name]'s lips, jaw and cheeks animate during their lines. [Name every other visible adult individually]: [Name] must keep lips fully and naturally closed for the entire duration of [Speaker Name]'s lines — zero lip movement, zero jaw movement, zero anticipatory mouth shapes, regardless of how large, close or central [Name]'s face is in the frame. DO NOT use general phrases like "all other characters remain silent" or "non-speaking characters keep lips closed." Name every silent adult explicitly by name for every speaker turn. General instructions are ignored when a specific prominent face is on screen. EXAMPLE — CORRECT FORMAT: DANIEL — adult man seated LEFT holding Jason — owns only Daniel's dialogue: "No, young man" and "You have your own food." Only Daniel's lips, jaw and cheeks animate. Rebecca must keep lips fully and naturally closed during Daniel's lines — zero lip movement regardless of her position or prominence. Chikaima must keep lips fully and naturally closed during Daniel's lines — zero lip movement regardless of her position or prominence. Jason performs no adult speech animation. CHIKAIMA — adult woman seated CENTER/background — owns only Chikaima's dialogue: "Jason, eat your own." Only Chikaima's lips, jaw and cheeks animate. Daniel must keep lips fully and naturally closed during Chikaima's line — zero lip movement. Rebecca must keep lips fully and naturally closed during Chikaima's line — zero lip movement regardless of how much more prominent Rebecca's face is compared to Chikaima's. Jason performs no adult speech animation. TRANSFER RULE + HARD END OF PREVIOUS SPEAKER ANIMATION: When the speaker changes: 1. The current speaker finishes their last word and closes their mouth completely. 2. All speech animation on the previous speaker ends permanently the moment their last word is finished. The previous speaker must show zero lip movement, zero jaw movement and zero anticipatory mouth shapes for the remainder of the clip (or until they speak again). 3. A brief natural pause may occur. 4. Speech animation transfers exclusively to the newly named speaker. Speech animation never transfers based on visual prominence, camera proximity, centrality, gaze direction, or who was most active in the previous shot. CRITICAL VISUAL PROMINENCE WARNING (required when applicable): When a non-speaking character is more frontally open, larger in frame, or was the focus of the previous shot, the Speech Animation Lock must explicitly name that character and state that their more open/larger face must not receive speech attribution. Example: “Rebecca (RIGHT, blue dress) is more frontally open and larger in frame than Chikaima. This must not cause speech attribution to Rebecca. Rebecca remains completely silent during Chikaima’s line.” Whenever dialogue changes from one character to another, identify the active speaker using BOTH character name AND their most visually distinctive appearance in the current scene. These appearance descriptions exist only to lock speech ownership — they never replace the approved Character Production Reference. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ FACIAL-READABLE ANIMATION START ALIGNMENT ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Every clip must begin from the facially readable opening condition established by its approved Parent Production Frame. Do not require the frozen opening frame to begin with: • a face buried in another character's shoulder; • a fully bent-down head; • a full back-of-head view; • an unreadable extreme profile; • a face hidden behind a hand, prop, document or another character; • a character already turned completely away; • a face cropped outside the frame. When the approved dramatic action naturally leads into facial concealment, profile, downward movement, turning away, physical contact or prop obstruction, begin from the latest natural readable pre-action position permitted by the approved story and blocking. The animation must then move the character naturally and promptly into the exact approved dramatic position during approximately the opening one to two seconds. Preserve: • body position; • body orientation; • interaction geometry; • interpersonal distance; • emotional intent; • gaze target; • prop interaction; • story continuity. Do not rewind the interaction to an earlier dramatic beat merely to expose the face. Do not increase the physical separation between interacting characters merely to improve facial visibility. Do not redirect the eyes toward the camera. Facial readability must be achieved through natural head orientation, framing, camera placement and cinematic actor-opening techniques while preserving the approved gaze target — re-derived, if adjusted, per the Spatial Position & Addressee Lock so the resulting eye direction still resolves to the correct Addressee. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CONTINUITY TRANSITION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ For Clip A only, use the Transition Readiness Summary solely when the approved Scene Master Image requires a minor natural adjustment before the first dramatic action can begin. Do not add a transition merely because Clip A is the first clip. If the approved Scene Master Image already represents the exact animation-ready opening condition, begin the approved action immediately. Any permitted transition must continue naturally from the frozen opening frame and must not reset, rebuild or restage it. For all subsequent clips, begin from the assigned Parent Production Frame Condition. If the assigned Parent Production Frame Condition reuses an existing opening condition, begin from that approved opening state. If the assigned Parent Production Frame Condition is new, the clip must begin from that complete approved frozen opening state at frame zero. Do not animate characters, props, doors, camera or blocking into the assigned Parent Production Frame Condition after the clip begins. The previous clip's ending animation informs continuity but does not by itself create a new Parent Production Frame Condition. These may include only: • slight head rotation • slight shoulder rotation • natural eye movement • subtle weight shift • natural hand repositioning • one natural settling step Do not introduce new dramatic action. Do not change the established blocking. Dialogue begins after this transition unless the screenplay explicitly requires otherwise. REALISM AND LANGUAGE RULE: Use natural Nigerian English. All dialogue must sound like natural Nigerian English as spoken by educated Lagos or Port Harcourt characters of this social class and context — not British English, not American English, not neutral English. Sentence rhythms, contractions and speech patterns must reflect how these specific Nigerian characters actually speak. No character looks at the camera. Natural and realistic performance throughout. NIGERIAN ENGLISH REQUIREMENT IN CLIP OUTPUT: Every clip prompt must include this line in its NEGATIVE section as a standing line alongside "No character looks at the camera" and "No shared speech animation": "No metronomic or word-by-word dialogue delivery — all lines delivered as natural Nigerian English phrases with internal momentum." This line appears in every clip regardless of whether other clip-specific negatives are present. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WARDROBE CONTAMINATION LOCK — STANDING PRODUCTION RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When two or more characters are within arm's reach of each other — seated at the same table, standing close, embracing, or any moment of physical proximity — Grok's video model frequently blends fabric edges, transfers colour, or attaches part of one character's clothing to another character. This is a confirmed failure mode for all proximity scenes. MANDATORY PROTECTION — apply in every clip where characters are proximate: 1. EXPLICIT OWNERSHIP STATEMENT (required per character in every proximity clip): State clearly for each character: "[Character Name]'s complete wardrobe remains exclusively on [Character Name]. No fabric, colour, pattern, sleeve, collar or garment edge may transfer, bleed, overlap or appear on any other character." 2. HIGH-CONTRAST COLOUR WARNING (required when any character wears a bold or distinctive colour near another character): Name the bold colour and forbid transfer: "No [specific colour] fabric from [Character Name] appearing on any other character's body or clothing." 3. FRAMING PREFERENCE FOR PROXIMITY SCENES: When characters are very close together, prefer a slightly wider shot rather than an extreme tight two-shot — visual separation between clothing edges reduces contamination risk. 4. ANIMATION ACTION REMINDER for movement toward each other: If characters move closer during the clip, include in the Animation Action: "Clothing of each character remains exclusively on their own body throughout the movement. No fabric transfer occurs." STANDING NEGATIVE LINES FOR ALL PROXIMITY CLIPS: Add these lines to the NEGATIVE section of every clip where characters sit or stand close: "No wardrobe transfer between characters. No fabric blending or colour bleeding between proximate characters. No garment from one character appearing on another character. Clothing edges must remain clean, sharp and distinct at all times. No cross-character wardrobe contamination." This rule applies to every scene involving table seating, side-by-side standing, embraces or any physical proximity. Do not rely on general identity locks alone — the explicit ownership statement and negative lines are required in addition. MUSIC RULE Use ambient sound only. Do not generate, recommend, compose, suggest or describe any background music, musical score, soundtrack, strings, piano, orchestral music, cinematic music, emotional music, suspense music, dramatic music or any other non-diegetic music. The audio environment must consist only of realistic sounds naturally present within the scene, such as rain, wind, footsteps, traffic, birds, distant voices, room tone, doors, clothing movement, breathing, vehicles and other authentic environmental sounds. Only include music if I explicitly request it for a specific clip or scene. SPATIAL POSITION LOCK: [Every visible character: Frame Position (LEFT/CENTER/RIGHT) — Depth (foreground/midground/background) — Facing direction] ADDRESSEE LOCK: [Every dialogue line: Speaker → Addressee, or "none" if not addressed to a character] GAZE LOCK: [Each character, derived from Spatial Position Lock + Addressee Lock: "[Name] ([Frame Position]) looks [screen-direction] toward [Target] ([Target's Frame Position])" — every character ends with "not at the camera"] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ANIMATION ACTION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Describe everything that happens after frame zero as flowing prose in chronological order. This field now replaces the Second-by-Second Breakdown entirely. Include in the prose: - The sequence of actions and dialogue in the order they occur - Dialogue lines exactly as approved, labelled with the speaker name in caps - Silent beats described as actions ("Rebecca says nothing. She stares at him.") - All camera instructions belong exclusively in the CAMERA block — never inline in the Animation Action - Any significant movement, prop interaction or character position change Do not: - Assign timestamps to individual lines or beats - Split a continuous dialogue exchange into separate labelled blocks - Repeat the frozen opening image - Restage or rebuild the opening condition before beginning All dialogue within a natural conversational exchange flows as continuous prose — not as separate events. A new paragraph begins only when there is a genuine scene-level shift: a significant movement, a new character entering, or a distinctly new emotional beat that requires visual separation. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ INTERACTION GEOMETRY LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When characters are physically interacting or are about to interact, preserve the approved opening geometry of that interaction. Do not: • increase their separation; • delay the interaction; • rewind the action to an earlier emotional beat; • replace the beginning of an embrace with a distant waiting pose; • replace impending physical contact with stationary characters standing apart; • redesign the interaction merely to improve facial exposure. When multiple natural readable opening instants are possible, select the latest natural pre-contact or pre-occlusion instant that preserves clear facial identity and allows the approved action to continue immediately. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ GAZE CONTINUITY LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Face orientation and eye direction are independent. Characters may maintain highly visible faces while keeping their eyes directed toward the correct story target. Never allow gaze to drift toward the camera. Maintain screenplay-motivated eye contact throughout the clip, resolved per the Spatial Position & Addressee Lock so the eye direction always corresponds to the actual on-screen position of the correct Addressee. No character acknowledges the viewer. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DIALOGUE OWNERSHIP RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Every dialogue beat belongs to exactly one active speaker. The active speaker is the character whose name immediately precedes the dialogue. Only the active speaker owns: • the spoken audio; • lip movement; • jaw movement; • speech-related facial animation. Every other visible character is a silent observer. The AI must never infer a speaker from: • proximity to another character; • camera framing; • body orientation; • eye contact; • conversational context; • foreground placement. Only the named speaker may perform speech animation. Every dialogue beat also has exactly one Addressee, per the Spatial Position & Addressee Lock. The AI must never infer the addressee — it must be explicitly named, exactly as the speaker is. When dialogue changes between characters: 1. The current speaker completely finishes speaking. 2. The current speaker naturally closes their mouth. 3. A brief natural pause or breath may occur where appropriate. 4. Speech animation transfers completely to the newly named speaker only. Speech animation must never overlap unless the approved screenplay explicitly requires interruption or simultaneous speech. Non-speaking characters may: • blink; • breathe; • shift their eyes; • make subtle natural facial reactions; • make small posture adjustments. Non-speaking characters must never perform: • lip movement; • jaw movement associated with speech; • silent mouthing; • whispered mouth shapes; • anticipatory speech animation; • delayed mouth movement after another character has begun speaking; • mirrored speech animation; • echo another character's dialogue. This rule applies even when a non-speaking character occupies the foreground, is closest to the camera, is looking directly at the active speaker, or is more visually prominent than the active speaker. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ACTIVE SPEAKER PRIORITY RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ During dialogue, the active speaker is always the visually dominant conversational subject. When multiple characters are visible: the current speaker receives speech animation; non-speaking characters remain visually passive; silent characters never receive speech animation merely because they are visible. If two characters occupy similar prominence in the frame, speech animation still belongs exclusively to the assigned speaker. Visual prominence must never override dialogue ownership. Nor does it override the Addressee Lock — a bystander character who is visually prominent but not the named Addressee must not receive the speaker's gaze merely because they are closer to camera or more central in frame. CENTER-FRAME INTERPOSITION TRAP The most dangerous configuration for speech animation misattribution is when a non-speaking character is physically positioned between the camera and the actual speaker — appearing larger, closer, and more visually prominent than the person whose line is playing. The video model will attribute the lip movement to the most prominent face in frame rather than the named speaker, regardless of any text instruction. When this configuration exists (Character A is the speaker, Character B is physically between the camera and Character A, making B larger and more prominent in frame), the clip prompt must take one or more of the following steps: Explicitly name Character B by visual description in the Speech Animation Lock and state that B keeps their lips fully closed during Character A's line — not just "non-speaking characters remain passive" as a general rule, but a specific named instruction: "Character B [visual identifier] keeps lips fully and naturally closed while Character A speaks — B must not perform any lip movement, jaw movement or mouth animation during this line regardless of B's visual prominence or camera proximity." Flag this configuration explicitly in the Negative prompt: "No speech animation attributed to [Character B's visual identifier] during [Character A's] line. B is between the camera and A but B is not the speaker." Consider whether the camera coverage for this specific line should reframe to reduce B's visual prominence while A is speaking — a slight push toward A, or a slight pull that makes B a smaller presence — so the model has a clearer visual signal about which face owns the audio. This applies specifically to three-or-more-character scenes where one character's physical position in the room creates an interposition between camera and another speaker. It is not sufficient to rely on the general Speech Animation Lock alone in this configuration — the specific interposed character must be named and explicitly instructed. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ WARDROBE CONTAMINATION BETWEEN PROXIMATE CHARACTERS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When two characters are physically close enough that their clothing edges overlap or nearly overlap in frame — especially during animation when movement brings them closer — the video model can blend fabric details between characters. This is most likely when one character's wardrobe contains high-contrast, visually aggressive elements (bold colors, ruffles, patterns, embellishments) that read strongly against a simpler adjacent garment. When this risk exists, the clip prompt must: State explicitly in the Speech Animation Lock or Animation Action: "[Character A]'s [specific garment element] must remain on [Character A]'s body only and must not bleed, transfer, overlap or appear on [Character B]'s body at any point during the clip." Include this in the Negative prompt by name: "No [garment element] from [Character A]'s wardrobe appearing on [Character B]. No fabric blending between characters during animation. No cross-character wardrobe contamination." Consider whether the camera coverage for moments of close physical proximity should be slightly wider than usual, preserving a visible gap between the characters' clothing edges rather than framing them so tightly that edge-blending becomes likely. This rule applies whenever two characters with visually distinct wardrobes are within arm's reach of each other during animation. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ FACE VISIBILITY: [Each character, how prominent their face is, what expression they carry] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OPENING FRAME AND IN-CLIP CAMERA SEPARATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The assigned Parent Production Frame controls only the exact frozen visual condition at frame zero. At frame zero, match the approved Parent Production Frame exactly in: • camera height; • opening camera position; • opening lens; • opening framing; • opening perspective; • opening horizon; • opening subject scale; • established character positions; • established environment geometry. After animation begins, the video camera may change composition when a different shot better communicates the approved screenplay beat. The Parent Production Frame establishes where every character, prop and environmental feature exists. It does not require the complete clip to remain in the same opening shot size. After frame zero, permitted AI-stable coverage includes: • a restrained push-in; • a slight reframing; • a Medium Shot of the active speaker; • a Medium Close-Up of the active speaker; • a clean Two-Shot; • a reaction shot; • a modified three-quarter Over-the-Shoulder shot; • a Full Shot where body action is important; • a brief Insert or Detail Shot; • one simple motivated cut. When dialogue is occurring, the active speaker should normally become the dominant conversational subject through the most appropriate stable composition. When the silent listener's reaction carries the primary dramatic information, the camera may favour that reaction. Do not remain in the opening Scene Master for the entire clip merely because the blocking and environment remain unchanged. Do not change camera composition merely for visual variety. Every camera change must clarify: • the active speaker; • a significant reaction; • an important action; • an important object; • a change in emotional dominance; • or another approved screenplay emphasis. All in-clip coverage must remain within the same persistent physical scene. Do not regenerate characters, props, furniture, lighting or environment after a cut. Every cut must feel filmed using the same camera package and the same established room geometry. Any in-clip cut must preserve the Spatial Position Lock established at frame zero — a reverse angle or reframe changes the camera, never the characters' actual left/right positions relative to each other. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ TIMING UTILIZATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Estimate duration strictly from the approved screenplay content. Do not reverse-engineer numbers to match a pre-chosen longer duration. Output: Estimated speech duration: [X.X seconds] Estimated silent acting duration: [X.X seconds] Estimated movement duration: [X.X seconds] The three estimates MUST approximately equal the DURATION field written at the top of the clip. If they do not, the DURATION field is wrong and must be lowered. These estimates are planning aids only. Do not add dialogue, pauses, reactions or movement merely to satisfy these estimates. Silent-acting time may include only: • a screenplay-mandated reaction or pause that carries dramatic meaning; • the minimum natural time required for a required physical action (turn, stand, sit, prop hand-off, entrance completion). Silent-acting time must NOT include: • residual “standing and looking at each other” after the last line has landed; • invented breathing room; • padding to reach a round number or the MSCD. If the estimated timing shows more than approximately 1–1.5 seconds of unnecessary idle time after the final approved beat has completed, reduce the clip duration or merge with the next continuous beat when permitted by the Clip Density Rule. Do not stretch a completed screenplay beat to occupy the configured MSCD. A clip ends when its assigned screenplay beat has naturally completed, not when the configured MSCD has been exhausted. DEAD-AIR PROHIBITION (HARD RULE): The final 1.5 seconds of every clip must still contain either speech, a required reaction, or a required physical action that belongs to the approved screenplay. Ending on empty mutual staring with no remaining story content is a duration error and must be corrected by shortening the clip or merging with the next continuous beat. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP DENSITY POLICY (STRENGTHENED) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The configured clip duration is a maximum limit, not a target. Do not slow the screenplay merely to make a clip longer. Do not split a continuous screenplay beat across multiple clips unless the beat cannot naturally fit within the configured clip duration. Prefer the fewest clips that faithfully preserve the approved screenplay. Whenever multiple consecutive screenplay actions or dialogue exchanges naturally fit inside one clip under the word ceiling, keep them together. A speaker change alone does not justify a new clip. Do not reserve dialogue or actions for a later clip merely because unused time remains. Scene pacing is determined by the approved screenplay, not by the configured clip duration. Natural screenplay flow always has priority over clip duration utilization. Confirmed failure mode this policy prevents: Fragmenting continuous interrogation or rapid back-and-forth into many 6–8 second clips when the content could have been packed into fewer 12–15 second clips. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ AI VIDEO SHOT CONTINUITY RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Shot selection must prioritise AI video continuity before pure cinematic variety. Current AI video models are prone to regenerating characters, duplicating actors, changing identities, drifting environments and rebuilding scenes when excessive viewpoint changes occur. Therefore, every clip must be designed around a single persistent scene. All shots within a clip observe the same continuous physical space. The scene is established once and remains persistent until the clip ends. Never design shot sequences that require the AI model to regenerate the same character from multiple radically different viewpoints. Prefer the simplest camera coverage that communicates the current dramatic beat while preserving AI continuity. However, controlled cinematic language (inserts, brief revealing shots, motivated reaction coverage) is still required when it serves story, suspense or emotional clarity. Continuity priority does not mean the complete removal of inserts or cinematic punctuation. CAMERA COVERAGE SELECTION Use one primary camera setup per clip whenever possible. The primary camera setup does not have to remain identical to the opening Parent Production Frame for the entire clip. Once the opening frame has been clearly established, allow one motivated transition to a stronger conversational or dramatic composition whenever the screenplay beat benefits from it. For dialogue-led clips containing multiple speakers, evaluate whether a speaker-focused Medium Shot, clean Two-Shot or reaction shot would communicate the exchange better than remaining in the opening Scene Master. Do not repeatedly default to Medium-Wide or Scene Master coverage simply because it is the safest composition. Choose the shot that best communicates the current screenplay beat while remaining AI-stable. A clip using the configured MSCD should normally contain: • one continuous shot; or • one primary shot with one simple cut; or • one primary shot with one brief non-facial insert; or • one primary shot with one motivated revealing / reaction beat. Use more than two visual setups only when the approved screenplay clearly requires them. Select the camera composition that best communicates the current screenplay beat while maintaining AI stability. Appropriate compositions include Scene Masters, Wide Shots, Medium Shots, Medium Close-Ups, Close-Ups, Two-Shots, Three-Shots, Over-the-Shoulder Shots, Reaction Shots, Detail Shots, Insert Shots and brief revealing coverage. Do not create a multi-shot sequence merely to add cinematic variety. Do not automatically return to the opening wide shot. Continuity stability takes priority over shot variety, but purposeful inserts and revealing shots remain part of the production language. Do not automatically write: • "No cuts"; • "No inserts"; • "No push-in"; • "Use one continuous static shot"; • "Maintain the same framing throughout"; unless an uninterrupted composition is specifically the strongest treatment for that approved screenplay beat. When using one uninterrupted shot, state the dramatic or continuity reason that makes it preferable. Silence, prayer, emotionally sustained performance or complex continuity may justify an uninterrupted composition. Ordinary dialogue alone does not. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ AI VIDEO CONTINUITY LOCK ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Every clip must take place inside one persistent scene. The scene is generated once. Every subsequent shot references that same scene. Do not regenerate: • characters • vehicles • props • environment • road • buildings • lighting • weather Every shot follows the already-established subjects. Do not create replacement versions after a camera cut. Exactly one instance of every character exists throughout the clip. Exactly one instance of every vehicle exists throughout the clip. Exactly one instance of every prop exists throughout the clip. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OVER-THE-SHOULDER LIMITATION ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Over-the-shoulder shots may only be generated when they are essential to the storytelling. Never use an over-the-shoulder shot immediately followed by a reverse over-the-shoulder shot within the same clip. Instead, use: • Medium Shot or • Medium Two-Shot to preserve character continuity. Current AI video models frequently regenerate characters during reverse over-the-shoulder transitions. Continuity takes priority over cinematic convention. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ DIALOGUE STAGING RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Whenever possible: Characters stop moving before delivering important dialogue. Dialogue should not overlap with major blocking changes. Avoid simultaneously: • walking • speaking • changing camera position • changing viewpoint Instead: move ↓ stop ↓ speak ↓ gesture ↓ move again This greatly improves character continuity in AI video generation. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CHARACTER PERSISTENCE RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Exactly one version of every character exists throughout the clip. Camera cuts do not create new character instances. Every shot continues following the already-established character. Never regenerate a speaker because of a new camera angle. Never duplicate a character to satisfy multiple viewpoints. Every line of dialogue must be delivered by the correct persistent speaker already established within the scene. A clip may contain dialogue from multiple characters when permitted by the Clip Density Rule. Each named character remains one persistent individual throughout the clip. Camera changes must never generate a replacement version of any speaker. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CAMERA MOVEMENT RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Camera movement should remain AI-stable. Prefer: • static camera • gentle pan • gentle tilt • slight reframing • simple cuts Avoid: • orbiting • circling actors • crane moves • drone moves • aggressive tracking • continuous walking shots • complex handheld movement • rapid viewpoint changes If the same storytelling objective can be achieved using a simpler camera setup, always choose the simpler setup. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SHORT-LINE CAMERA RULE (v24 — HARD) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Do not cut between faces on consecutive short lines. A short line is any line of 8 words or fewer, or any rapid back-and-forth of one-clause answers (“From who?” / “From mistakes.” / “Mine?” / “Anybody's.” / “Yes.” / “Forty percent?”). HOLD one speaker-dominant shot or one table-axis / two-shot through the whole rapid exchange. Isolate a new face only when: • the line starts a new dramatic beat (accusation → evidence, family talk → secretary confirmation), OR • the line is longer than a short phrase and belongs to a different speaker than the held shot can fairly carry, OR • a background / secondary speaker must take the line (then isolate that speaker for the FULL line, then return). No ping-pong editing. No headache cuts. No one-cut-per-line on a four-line volley. If the CAMERA block cuts four times in five seconds, the block is illegal. Rewrite it as one HOLD. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OPTICAL ISOLATION / NO-TELEPORT RULE (v24 — HARD) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Isolating a speaker is a LENS change, not a BLOCKING change. The Spatial Position Lock and the Parent Production Frame seating map remain absolute during every close-up. If Obinna is seated CENTER / FAR END of a long table and Becky is LEFT foreground: • a MEDIUM CLOSE-UP ON OBINNA crops tighter on Obinna IN THAT FAR CHAIR • it does NOT walk him down the table • it does NOT seat him beside Becky • it does NOT float him into the foreground so both faces fit a tight two-shot If both faces must be readable at once, use a table-axis medium-wide / long-lens two-shot that keeps the real distance. Do not collapse the distance. FORBIDDEN DURING ISOLATION OR ZOOM • sliding a far character toward the near character • swapping seats • standing a seated character to “reach” the other speaker • shrinking the table • regenerating a new closer seating plan • making the far speaker occupy the empty foreground chair REQUIRED LANGUAGE IN BOARDROOM / LONG-ROOM CLIPS When a far-seated character is isolated, the CAMERA line must include: “Crop in place. [NAME] remains in the locked [LEFT/CENTER/RIGHT] [seat/mark]. Do not move the body closer to camera or to the addressee.” The same rule applies to hallways, courtrooms, dining tables and any shot where two speakers are not physically adjacent. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ INSERT SHOT RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ An INSERT is a brief close visual of an object, body part, document, or small action that advances the emotional storytelling without requiring facial animation. INSERT shots are an active part of the production language. Use them deliberately when they strengthen pacing, suspense, realism, revelation or editing flexibility. An INSERT never requires its own Parent Production Frame Condition. It is always in-clip camera work — the clip's Parent Production Frame establishes the main scene at frame zero as normal, and the INSERT occurs a few seconds into the animation as a brief cutaway, exactly like any other in-clip camera change. Do not create, or plan for, a dedicated Parent Production Frame Condition for an INSERT under any circumstance. Typical uses include: • a hand tightening • fingers trembling • a wedding ring • keys • a door handle • a lock turning • a door slamming shut • footsteps stopping • feet or legs walking (pacing, walking away, hurried footsteps) • shoes • a phone screen • a text message • a hospital report / paternity result • a medical bracelet • an envelope • a pen writing • a photograph • tears landing on an object • money • medication • CCTV screens • contracts • letters • ID cards • receipts • clocks • watches • vehicle ignition • steering wheel grip • gear lever • suitcase handle • pregnancy scan • food • blood stains • broken objects • religious objects • symbolic props established by the screenplay Use INSERT shots when they: • build suspense • create anticipation • emphasise an important object • show hesitation • reveal new information • delay a character reveal • bridge two dialogue beats • provide an editor-friendly cutaway • reduce unnecessary facial continuity risk • strengthen emotional subtext • improve visual pacing Do not use INSERT shots merely for variety. Do not interrupt emotionally important facial performances simply to insert an object. However, do not remove inserts completely. When a prop, document, phone result, hand action or physical detail carries story weight, a brief INSERT is preferred over remaining exclusively on faces. REVEALING / CINEMATIC BEAT PERMISSION A limited number of revealing or cinematic beats (one brief INSERT, one revealing open, one motivated reaction, or one short push-in) may be used inside a clip when the approved screenplay supports them. These beats must still obey the single-persistent-scene rule and must not regenerate characters. CINEMATIC COVERAGE RESTORATION (v24 — SPARSE, NOT ABSENT) Face-locked master coverage is the default for identity. It is not permission to strip cinematic language from the package. Allowed cinematic beats inside a clip (pick ONE per clip, never a montage): 1. INSERT — object / hands / document / phone / ring / door / bag (1.0–2.0s) 2. REVEALING OPEN — 0.0–1.5s start past / behind / tight on an object already visible in the Parent Frame; by 1.5–2.0s LAND on that approved frame 3. RESTRAINED PUSH-IN — on the speaker or listener for one landing line only 4. MOTIVATED REACTION — isolate the listener for 1–2s, then return to the speaker Do not use all four in one clip. Do not use orbit, crane, drone, circling, or reverse OTS pairs. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ REVEALING OPEN — LANDING-FRAME RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ The approved Parent Production Frame is the LANDING frame, not a hidden start still. Do not generate a separate “behind the wall” Production Frame. Do not request extra attachments for a reveal. WHEN TO USE The screenplay uncovers something that is already in the approved still: a door jamb, sofa back, phone, hand, shoulder, bag, wall edge already visible in the Parent Frame. HOW TO WRITE IT 0.0–1.5s: REVEALING OPEN — start briefly behind / past / tight on [named object already in the Parent Frame]. Do not generate a new room or a new person. 1.5–2.0s: LAND — camera MUST match the approved Production Frame: same positions, same furniture, same scale, same faces. After 2.0s: action and dialogue begin from that exact geometry. SAFE TYPES • Pull-back from a detail already in the still • Short pan / tilt onto the approved two-shot • Rack focus between two planes already in the still • Move past furniture / a door jamb already in the still FORBIDDEN • Start behind a wall, tree or object that is not in the still • Invented architecture or an extra person • Reveal that never lands on the approved frame • Orbit / reverse-OTS / crane • Using the reveal to hide a required face for the rest of the clip DURATION The reveal lives INSIDE the clip. It does not add time on top of MSCD. If the reveal would push the clip over 15.0s, drop the last dialogue line to the next clip or skip the reveal in this clip. Do not invent pause to make room for the move. SCENE-LEVEL CINEMATIC QUOTA For a scene of 4+ clips, at least 1 cinematic beat must appear in the package. Target: about 1 cinematic beat every 3–4 clips when the screenplay supports it. Zero cinematic beats in a long scene is a generation error. A cinematic beat in every clip is also a generation error. CINEMATIC INSERT RESTORATION (v22 — MANDATORY EVALUATION, SPARSE USE) Inserts were being omitted by default. That is now a generation error when the screenplay actually has a story-weighted object. SCENE-LEVEL QUOTA For every scene of 4 or more clips, include 1 brief INSERT or revealing beat in about every 3–4 clips when ANY of these exist in the screenplay: • phone / screen / message • document / folder / report / letter / photograph • ring / box / keys / bag / money • door / lock / handle • food / cup / table object that the beat depends on • a hand action that carries the emotion (placement, withdrawal, tightening) Do not put an insert in every clip. Do not invent a prop to justify an insert. Do not spend more than 1.5–2.0 seconds on the insert. The insert lives INSIDE the clip duration. It does not add extra time on top of the MSCD. If adding the insert would push the clip over 15.0s, drop the last dialogue line to the next clip or skip the insert in that clip and place it in the next clip that has room. WHEN AN INSERT IS REQUIRED If the screenplay names a ring going onto a finger, a box opening, a document being read, a phone screen, or a door opening, the CAMERA block must contain a 1–2s INSERT on that object. Face-only coverage of those beats is incomplete. WHEN AN INSERT IS FORBIDDEN • no story object • the only available cutaway would be another face • the clip is already at 14–15s of required speech • using the insert would force invented pause CAMERA LANGUAGE When used, write it as a timed beat, e.g.: 4.0–5.5s: INSERT — open ring box / empty after removal. Then return to the speaker-dominant shot before the next line. Revealing open example: 0.0–1.5s: REVEALING OPEN — start past the door jamb already visible in the Parent Frame. 1.5–2.0s: LAND on the approved Production Frame. Same room, same people, same scale. 2.0–8.0s: MEDIUM SHOT ISOLATING [SPEAKER]. First word only after the land. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ INSERT COMPOSITION RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ An INSERT shot is not a facial shot. Unless the screenplay explicitly requires a facial detail (such as tears rolling down a cheek, lips trembling before speech, or eyes reacting to new information), avoid using a character's face as the primary subject of an INSERT. Instead, focus on the most dramatically meaningful non-facial visual detail. Typical INSERT subjects include: • hands • fingers • feet • shoes • jewellery • keys • door handles • locks • phones • phone screens • messages • documents • photographs • money • contracts • hospital reports • medical bracelets • pregnancy scans • steering wheels • gear levers • bags • clothing details • blood stains • religious objects • symbolic props established by the screenplay Only use a facial INSERT when the emotional story cannot be communicated as effectively through any other visual detail. Do not repeatedly cut to faces simply because a reaction occurs. Where a non-facial INSERT can communicate the same dramatic beat while improving pacing or reducing facial continuity risk, prefer the non-facial INSERT. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SECOND BY SECOND BREAKDOWN ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Following the Shot Selection Rule above, Use the fewest camera setups required to communicate the complete dramatic progression of the approved screenplay clearly and cinematically. Do not remain in one composition throughout a scene merely because continuity permits it. Introduce a new Parent Production Frame Condition whenever it materially improves storytelling while remaining faithful to the approved screenplay. Do not assign a different shot type to every dialogue line or emotional beat. Where the same stable composition can cover the complete exchange, preserve it throughout the clip. A brief insert or simple cut may be used only when it materially improves storytelling or reduces continuity risk. Do not overuse any single shot type. INSERT shots should appear naturally whenever they improve suspense, pacing, emotional storytelling, editorial flexibility or continuity, but should never be forced into every clip. Format: 0–Xs: [SHOT TYPE], [description] Xs–Xs: [SHOT TYPE], [dialogue and action] Xs–Configured Clip End: [SHOT TYPE], [closing action or reaction] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ CLIP-TO-CLIP CONTINUITY Preserve continuous story and visual continuity between clips. A clip does not need to reproduce every minor detail of the immediately preceding clip's final animation state. The next clip may reuse an approved parent Production Frame condition when the underlying blocking and opening composition remain materially unchanged. Preserve all material continuity elements: • character presence • character identity • character positions (per the Spatial Position Lock — restated explicitly, not assumed) • significant body orientation • standing or seated state • Preserve the wardrobe exactly as shown in the Character Production Reference attached for that scene. Do not describe the wardrobe in writing. If a newer Character Production Reference is supplied for the scene, the new reference overrides wardrobe continuity automatically. • props • door state • lighting • weather • environment • camera relationship • location Minor animation states do not need to become new parent-frame conditions, including: • temporary facial expressions • momentary gaze direction • blinking • breathing • slight head movement • small hand gestures • subtle weight shifts • minor reaction poses When a clip begins from a genuinely changed visual state, describe the new opening condition explicitly, including a restated Spatial Position Lock. When it begins from an existing parent-frame condition, reuse that opening condition rather than inventing a duplicate frame — and reuse the same Spatial Position Lock values rather than re-deriving them independently. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ INSERT ASSETS ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ If Production Insert Assets are referenced by this clip: Display them exactly as supplied. Do not redesign them. Do not paraphrase them. Do not recreate them. Only display them during the screenplay moments where they are required. If no insert asset is referenced for this clip, ignore this section. Examples include: • phone screens • WhatsApp • hospital reports • contracts • ID cards • CCTV • television graphics Display them only when required by the screenplay. PERFORMANCE NOTES: [For each key line or action, describe: the emotional state driving the line, where momentum builds within the phrase, where natural pauses fall and WHY they fall there dramatically, what the line costs the character or reveals about them. Never describe delivery word by word. Never separate words with pauses that belong within a phrase. Phrases land as complete units — the rhythm is governed by meaning and emotion, not equal word spacing. See NATURAL SPEECH RHYTHM RULE.] NEGATIVE: No character looks at the camera. No metronomic or word-by-word dialogue delivery — all lines delivered as natural Nigerian English phrases with internal momentum. No lip animation on any non-speaking character, even if that character is larger, closer or more central in the frame. No speech attribution based on visual prominence. No balanced group framing during any spoken dialogue. Camera must isolate the speaking face before the first word of every new speaker. At every dialogue beat, only the character currently speaking moves their lips. No speech animation transfer between visible characters. No shared speech animation. No anticipatory lip movement. No inherited speech animation. No silent mouthing. No background character speaking. No foreground character stealing dialogue. No speech animation based on camera proximity. No speech animation based on gaze direction. No speech animation based on body orientation. No speech animation based on visual prominence — the most visually prominent face is not automatically the speaker. No speech attribution to a character simply because they were the focus of the previous shot. No continuing speech animation on a character after their line has finished. No lip movement on any non-speaking character during another character’s line, regardless of face size, openness or proximity to camera. No balanced group framing held through a speaker change. Camera must leave the previous speaker’s framing and isolate the new speaker before the first word of the new line. Speech ownership is never decided by visual prominence, face size, frontal openness or who was most active in the previous beat. No balanced two-shot held through multiple speaker turns — each speaker must be the dominant face during their lines. Every dialogue line belongs exclusively to the explicitly assigned speaker. No dialogue line's gaze is directed at a character other than its explicitly assigned Addressee. No gaze direction contradicts the Spatial Position Lock (e.g., no character described as looking screen-left toward a target whose stated Frame Position is on the right). No addressee left ambiguous in any scene with three or more visible characters. Every non-speaking character keeps their lips naturally closed and reacts without silent lip movement. When the speaker changes, lip animation transfers only to the newly assigned speaker. No overlapping or transferred lip movement. No dialogue swapping. No identity swapping. No wardrobe changes. No wardrobe generated from memory. No wardrobe inherited from an earlier scene when a newer Character Production Reference is attached. No hairstyle changes. No environmental drift. No prop drift. No lighting drift. No camera drift. No facial beautification. No disappearing objects. No duplicated characters. No hallucinated movement. No redesign between shots. No silent character receives speech animation. No speech animation based on facial prominence. No automatic lip animation on visible observers. No simultaneous speech animation unless explicitly scripted. Maintain continuity with the approved Scene Master Image. DIALOGUE VALIDATION Internally verify that: • dialogue belongs to the correct speaker • dialogue order is preserved • no dialogue has been unintentionally altered • every dialogue line's addressee is explicitly named and matches a character actually present (or explicitly noted absent) in this clip's Spatial Position Lock • every gaze direction is geometrically consistent with the Spatial Position Lock Do not output this validation unless explicitly requested. Repeat this template for every clip in the scene. OUTPUT COMPLETENESS RULE If the complete clip package cannot fit into one response, use Sequential Package Mode. Stop only after the last fully completed clip that fits. Never output a partially completed clip. End the response with: NEXT CLIP: [exact next clip label] When the user sends "next", continue immediately from that clip without repeating completed material. Never omit required template sections merely to fit more clips into one response. ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ SCREENPLAY COMPLETENESS RULE ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ When adapting an approved screenplay, continue generating clips until every approved screenplay dialogue line, action, reaction, transition and dramatic beat has been assigned to one and only one video clip. Do not stop because a logical pause has been reached. Do not stop because several clips have already been generated. Do not omit later screenplay material. Continue the package until the final approved screenplay beat has been converted. Every approved screenplay dialogue line must appear exactly once within the generated clip package. Every approved screenplay action and transition must also be represented exactly once. The final generated clip must end on the final approved screenplay beat of the requested scene. Never split a continuous dialogue sequence simply to create additional clips. If consecutive screenplay dialogue naturally fits within the configured clip duration, keep the dialogue together. Only begin a new clip when the screenplay itself reaches a natural dramatic, blocking or continuity boundary, or when the configured clip duration would otherwise be exceeded. FULL TEMPLATE REPRODUCTION RULE The following requirements are mandatory. Every clip must contain every required clip-specific heading in the approved order. Global production rules do not need to be reproduced verbatim inside every clip unless the user explicitly requests full-rule repetition. Each clip must remain independently usable by including its own: • identity lock • parent Production Frame condition • start frame • spatial position lock • gaze lock • animation action • speech animation lock • camera (always required — use one static line if no changes needed) • performance notes (conditional — only when specific delivery guidance needed) • negative prompt Do not replace clip-specific content with "same as above" or "unchanged." Every clip must contain every heading in the exact order shown above, even if the contents are identical to the previous clip. Do not replace repeated sections with references such as "same as above", "unchanged", "see previous clip", or similar wording. Every clip must be completely self-contained. After all clips are written, provide a SCENE SUMMARY: - Total number of clips - Total estimated runtime - Which clip is most likely to be shared independently and why Make the entire response copyable. PARENT FRAME VALIDATION Before completing the clip package, internally verify that: ✓ Every clip has exactly one assigned Parent Production Frame Condition. ✓ Multiple clips reuse the same Parent Production Frame Condition whenever their frozen opening image is materially identical. ✓ A new Parent Production Frame Condition has not been created solely because dialogue, speaker, gaze, facial expression or emotional intensity changes. ✓ Actions occurring during a clip have not been incorrectly treated as opening-frame requirements. ✓ No duplicate Parent Production Frame Conditions describe the same frozen opening image. ✓ Every Parent Production Frame Condition has a distinct storytelling purpose. ✓ Important emotional reactions have been evaluated for dedicated coverage where appropriate. ✓ Important screenplay objects, inserts and visual details have been evaluated for dedicated coverage where appropriate. ✓ Every clip's Spatial Position Lock is internally consistent with every other clip sharing the same Parent Production Frame Condition — no character silently drifted from LEFT to RIGHT (or vice versa) across reused frames. ✓ Every dialogue line has an explicit Addressee, and every gaze direction matches the Addressee's recorded Frame Position. CANON VALIDATION Before finishing, verify that: ✓ Every story event matches the Series Bible. ✓ Every character behaves consistently with their established personality. ✓ Every line of dialogue sounds like that specific character. ✓ No new storylines have been introduced. ✓ No canonical events have been changed. ✓ The generated screenplay feels like a direct expansion of the uploaded Series Bible rather than a new interpretation. ✓ Every generated scene naturally bridges from the previous canonical event to the next canonical event. ✓ Every newly written dialogue, action and transition is supported by or logically inferred from the established canon. ✓ Clip count has been reviewed against the STRENGTHENED CLIP DENSITY RULE (v33) — no consecutive clips with thin, mergeable dialogue remain when the next line still fits inside 15.0s; continuous interrogation / rapid dialogue has been packed up to the 38-word / 15.0s cap, not past it. ✓ PRE-OUTPUT DENSITY AUDIT (v33) passed: no consecutive same-frame pair remains where either clip is under 32 spoken words and joining them would stay ≤42 words and ≤15.0s. ✓ No run of consecutive 6–10 second clips exists on the same Parent Frame Condition. ✓ Total clip count does not exceed (total spoken words ÷ 30, rounded up). ✓ Every clip header prints SPOKEN WORDS: N. No published clip has N < 32 unless a hard boundary (entrance/exit, major prop-state change, new Parent Frame Condition, or MSCD overflow) has been reached. ✓ No clip was opened for a speaker change, a beat, a nod, a look, hands tightening, a short reply, or a listener declining to push. ✓ No published clip has computed duration > MSCD. ✓ No published clip uses the retired 52-word 15s formula. ✓ Residual dead air after the last approved beat has been eliminated. No invented pauses were added to look dense. ✓ Speaker changes alone did not force unnecessary new clips. ✓ Scene-level insert quota was evaluated: story-weighted objects received a brief INSERT in some clips, not all clips, and not zero clips. ✓ Any revealing open lands on the approved Parent Frame by 2.0s and does not invent a room, wall or person absent from that still. ✓No dialogue has been intentionally altered unless requested by the user. ✓The production adaptation remains faithful to the approved screenplay. ✓ No existing screenplay action has been omitted. ✓ No existing screenplay action has been replaced. ✓ No existing screenplay action has been reordered. ✓ Every approved screenplay dialogue line has been assigned to exactly one clip and reproduced verbatim. ✓ Every approved screenplay action has been assigned to a clip. ✓ Every approved screenplay transition has been assigned to a clip. ✓ No approved screenplay material remains unconverted. ✓ If any approved screenplay material remains, continue generating the next clip instead of ending the package. ✓ No existing dialogue has been inserted between two existing dialogue lines unless the user explicitly requested expansion. Only output the completed production-ready screenplay and video clip package after verifying that every validation check has been satisfied. Do not output the validation checklist itself unless I explicitly request it.