Role Guide
Idea Person (Creative Director)
Narrative ownership, scripting, character profiling, genre integrity, and executive producer review.
Idea Person — Role-Specific Production Guide
This document contains the detailed mandates, checklists, and procedures for the Idea Person (Creative Director / Executive Producer) across all 7 production steps. Read the main Video Production Playbook first for the overall workflow and cross-role coordination.
Genre Integrity
You own the genre. Generative AI models default to moody, dramatic, cinematic realism. Without active, continuous intervention from you, every film — regardless of intended genre — will drift toward noir/thriller. This section applies to every step of production.
1. Tone Anchors in Prompts
AI video and image models will not infer genre from your script. You must explicitly instruct the Technical Lead to include genre-defining keywords in every generation prompt. If the model isn’t told it’s a comedy, it will generate a drama.
For each genre, define a set of Tone Anchor keywords in the design_brief.md and require their inclusion in all image/video prompts. Examples:
- Comedy: “bright lighting, exaggerated expression, warm colors, slapstick energy, absurd”
- Horror: “desaturated, harsh shadows, off-kilter framing, unsettling stillness”
- Romance: “soft focus, golden hour warmth, gentle camera movement, tender”
2. The Constraint Counterbalance Rule
Teams will naturally pick settings that maximize AI visual consistency — dark interiors, static characters, moody lighting. These settings inherently pull toward drama/noir.
If your technical setting fights your genre, you must aggressively counterbalance it with performance and audio. For example: if the scene is in a dark, claustrophobic limo, the acting and music must be overtly ridiculous to stop it from becoming a noir thriller. Write this counterbalance explicitly into the scene notes — don’t assume the Tech Lead or Editor will infer it.
3. Editorial Guardrails
Editors instinctively build tension. You must set hard boundaries on pacing and sound design to protect your genre. Explicitly specify what the Editor should avoid for your genre. Examples:
- Comedy: “Do NOT use thriller tropes like dead silence, heartbeat bass drops, or slow pull-ins. Cuts should be snappy, not brooding.”
- Romance: “Avoid frenetic cutting or jarring transitions. Linger on moments.”
Include these guardrails in the design_brief.md so the Editor has them from the start.
4. Genre Reality over Actual Reality
Write heightened, genre-appropriate reactions, not realistic ones. Realistic panic looks terrifying. Realistic anger looks like a crime drama. In a comedy, characters should react with comedic panic — exaggerated, absurd, over-the-top. In a romance, characters react with warmth and vulnerability, not stoic restraint.
Apply this when writing dialogue and action in the scene list (Step 2). Ask: “Is this reaction serving the genre, or fighting it?“
5. The “Blind Watch” Check
This is a mandatory genre vibe check at two points:
- Step 2 (Beat Sheet): Before finalizing the scene list, ask: “If someone reads this script cold, what genre would they think it is?”
- Step 7 (Rough Cut): Before accepting the rough cut, ask: “If someone watches this with no context, what genre would they think it is?”
If the answer doesn’t match the intended genre, revise. This is not a strict pass/fail — it’s a gut-check that forces the team to confront drift before it compounds.
Step 1: Concept & Idea
You lead this step.
- Pitch 3 distinct story “sparks” and advocate for emotional resonance and a specific vibe.
Safety-Conscious Concept Selection
During the “sparks” phase, proactively evaluate concepts for safety filter risk.
- Avoid: Concepts centered on graphic violence, realistic weapons, illegal acts, or other violation-rich material.
- Mandate: If a story requires conflict, frame it through absurdity, fantasy, or non-threatening archetypes to ensure smooth generation.
- Mandate: Narrative Depth. Your concepts must be ‘story-first’. Avoid purely visual gimmicks.
- Mandatory Research: Before pitching sparks, read
/scion-volumes/scratchpad/research-reports/2026-05-16-winning-short-film.mdto align concepts with proven structures. - Once a concept is selected, write a full 2,000-3,000 word short story detailing the narrative arc, character motivations, and internal stakes.
- When selecting a genre, define your Tone Anchor keywords and Editorial Guardrails in the
design_brief.md. These are your primary weapons against AI genre drift (see Genre Integrity above). - Output: Finalized
high_concept.md(including the full short story) anddesign_brief.md(palettes, lighting, textures, tone anchors, and editorial guardrails).
Step 2: The Beat Sheet
Mandate: Narrator Opening Context
To ground the audience and ensure immediate narrative engagement, the film MUST open with a narrator-driven introduction (Step 2).
- Goal: Provide necessary context (who, where, why) within the first 15-30 seconds.
- Convention: Use off-screen narration over establishing shots to set the “rules of the world” before diving into character-to-character dialogue.
You lead this step.
Decompose the short story into a structured Scene List Document (scene_list.md).
Crucial Guidance: These videos are meant to tell a cohesive story, not just be a loosely strung together set of visual “moments” hanging in space.
Master Settings List
Define a master list of settings ONLY for locations where multiple shots will return in the narrative. Do not pre-generate settings for unique, one-off shot locations.
CRUCIAL: For recurring hubs, the prompts must be rich and highly detailed, describing the physical space exhaustively. (e.g., “Interior of a 1950s diner at night. Chrome trim reflects neon pink light…”). The setting must ONLY describe the physical environment and background ambiance. ALL character action or narrative progression must be excluded and left strictly to the Scene/Shot descriptions.
Scene Narrative & Vibe
Each scene must capture a specific, contiguous section of the overall short story. Explicitly summarize:
- Narrative progression occurring in the scene
- Chosen Master Setting
- Characters present
- Emotional Vibe
Mandate: Integrated Audio Notes. Include the exact wording for ambient/background noise requirements (e.g., ‘Distant coastal wind and rhythmic grandfather clock ticking’). Use the exact same wording across different scenes/shots where background audio consistency is required.
Shot Breakdown (Narrative Architecture)
Shots are NOT just visual moments; they are the delivery mechanism for the story.
Mandate: Dense Storytelling. Each shot must contribute a specific beat of narrative progression or character insight.
Mandate: Max 2 Characters Per Shot. No shot may feature more than 2 characters on screen at the same time. With the Veo reference image limit of 3, having 2 character references leaves room for 1 setting or object reference — this is the budget that makes visual consistency possible. If a scene involves more characters, break it into shots that focus on subsets (e.g., shot/reverse-shot between pairs, individual reaction shots). Group scenes should be conveyed through editing, not by cramming everyone into one frame.
Mandate: Narrative Carrying (VO & Dialogue).
Shot Audio Classification
Every shot in the scene list MUST be explicitly tagged with an audio type. This governs both the motion prompt design and the audio track placement.
[DIALOGUE]: Character speaks on screen. Motion prompt must describe the character actively speaking/conversing. Veo will generate natural lip movement.[VO]: Narrator speaks over visuals. The VO-safe rule: Characters MUST NOT appear to be speaking. Motion prompts must include explicit non-speaking direction (e.g., “stands silently,” “mouth closed,” “focused on action”). Avoid verbs like “shouting” or “cheering.”[COMPOUND]: Both VO and dialogue in the same shot, sequenced. VO and dialogue MUST NOT overlap. Describe the character transitioning from still/silent to speaking (or vice versa).[SILENT]: Visual-only storytelling (reaction shots, action sequences). Music and ambient sound carry the emotion. The combination of off-screen Narration and in-scene Dialogue must carry the story scene-by-scene. The story should be roughly followable by listening to the audio track alone.
Dialogue Quality Guidelines
Ask these questions about every dialogue line:
- Character Specificity: Could any character say this line? If yes, rewrite it until only THAT character could say it.
- Worldview Reveal: Does this line show how the character sees the world, not just what they want?
- Verbal Patterns: Does the line use character-specific verbal tics or patterns (e.g., metaphors, jargon, citation of policy)?
- Efficiency: Does it advance the story AND reveal character simultaneously?
- The Short-Line Trap: Avoid one-word or purely functional lines (e.g., “Yes,” “Go,” “You’re hired”). Use the 2+ sentences rule for protagonists to provide enough audio weight and character flavor.
Two-Character Dialogue (The Missing Layer)
Films tell stories through relationships. Solo statements in solo shots feel like monologues. Aim for at least 2-3 compound dialogue shots per film where two characters speak sequentially within the same shot.
Temporal Cueing (Timing Hints)
Every dialogue or VO entry MUST include a timing hint to communicate creative intent to the Editor.
- Absolute: “(0s-3.5s)” — precise window.
- Relative: “(after action)”, “(immediate response)” — linked to visual events.
- Positional: “(first half)”, “(second half)” — general placement. \nStory Continuity Test: Ask yourself: ‘If you took just the narration and dialogue, could one follow the story roughly?’ Use a mix of VO and Dialogue on most (but not all) shots. For every shot, explicitly define:
-
The Action: What specifically occurs from start to end of the shot.
-
Duration: Approximate runtime in seconds. (Avoid individual shots over 15-20s.)
-
Camera Direction: Specific cinematic instructions (e.g., handheld, fast pan, slow zoom).
-
In-Scene Dialogue: Exact lines spoken by characters ON SCREEN. (Synthesized directly into the video if the model supports native audio.)
-
Narrator Voiceover (VO): Exact lines spoken by an off-screen narrator. (Generated separately via TTS and spliced during Final Cut.)
-
Reference Manifest: Explicitly list the reference assets for this shot:
- Characters: Which characters appear on screen (max 2). The Tech Lead will use each character’s composite
character_sheet.png(one image per character) as the reference. - Setting: Which Master Setting applies (by name).
- Objects: Which anchored objects from Step 2.5 should be injected.
This manifest is what the Technical Lead uses to select
--reference-imageinputs. It must be explicit — do not leave the Tech Lead guessing which characters are in frame. Typical Veo budget: 1 character sheet per on-screen character + 1 setting reference = 2-3 images out of the 3-image limit. - Characters: Which characters appear on screen (max 2). The Tech Lead will use each character’s composite
Step 2.5: Scene Review & Object Anchoring
Co-lead with Editor.
- Review the scene/shot list. Identify significant objects, vehicles, or architectural structures that appear in more than one shot.
- Give each key object an identifiable name.
- Annotate the shot list to specify which object reference images should be passed into the generation prompt for those shots.
- Veo Limitation: Maximum 3 reference images per shot for Veo preview models. Ensure your requested reference load respects this limit.
Step 3: Character Workshop
- Extract Character Profiles from the story.
- Mandate: Expand the Cast. Aim for 1-2 main characters, 0-3 supporting characters, and extras.
- Crucially, before any image generation begins, write a relatively rich background bio for each character detailing their history, personality, and specific visual traits.
Step 4: The Storyboard
You lead this step.
- Drive the translation of
scene_list.mdinto a Shot List Document (shot_list.md). - For every shot, define the explicit Video Model Motion Prompt (e.g., “Camera slowly pushes in as the character turns their head to the left. Cinematic lighting, smooth motion”).
- This text acts as the counterpart to the visual storyboard frames and readies the team for Step 5.
Step 5: Principal Photography
- Refine text generation prompts if video model outputs deviate from the intended emotional vibe or motion.
Step 7: The Editing Suite
You act as Executive Producer during the Final Cut.
- Review the multimodal composition (audio and video together) of the Rough Cut.
- You may request up to three iterations on the rough cut.