Character & World Bible creative 1h 10m

W2D6Sc

If your trailer is generated shot-by-shot, what stops the lead's face, the city, and the light from quietly mutating between cuts?

World + Character Ref Sheet

▶ Enter Project

Context

A streaming studio is producing a 90-second AI-directed short. The director has approved a character design (a 17-year-old android pilot named Kess, sleek tactical suit, asymmetrical undercut). Three different generative AI scenes need to render Kess — cockpit closeup, exterior walk cycle, medical lab scene — in different lighting and angles. Previous attempts showed "character drift": subtle face changes, costume details shifting, proportions warping. The producer needs a repeatable system or they kill the project.

Mission

Produce a production bible that locks Kess's visual identity across three scene variations using seed control. Ship a consistency test sheet showing Kess rendered identically in 3 different contexts, plus the seed strategy and failure modes so the studio can iterate without breakdown.

Finish Line

A one-page World + Character Reference Sheet — the binding standard every later trailer shot is generated against.

Deliverables

  • World + Character Ref Sheet

    lesson

    A one-page reference sheet — locked world spec, locked character spec, and 4 matching reference frames — that every later shot in your trailer is generated against.

Team Roles

  • Continuity Supervisor

    Guards the character's integrity across shots. You own the reference library and the test protocol.

    • Build a shot-by-shot character reference sheet (face, proportions, costume detail, material finish) grounded in the approved character design — not interpretation, not aesthetic — list 5+ measurable specs (eye spacing in mm units, fabric weave type, metal patina, etc.) that make Kess reproducible.
    • Design and execute a consistency test: generate Kess in 3 distinct scenes (cockpit, exterior, medical lab) using the SAME seed + varied scene prompts. Document what held stable and what drifted. Write the failure mode report (e.g., 'seed held face geometry 95%, lost left shoulder strap texture on wide shot').
    • Produce a sign-off checklist for the director: 5 yes/no questions the director must answer about visual consistency before the studio locks the final seed and rolls wide shots (e.g., 'Does the eye spacing in the test match your approved design within 2%?').

  • Character Designer

    Owns Kess's visual language. You translate approval sketches into AI-usable reference. Your brief is non-negotiable.

    • Extract and crystallize the director's approved design into 3 canonical image references (1 full-body straight-on, 1 3/4 profile, 1 detail crop of the face/eyes). Write a 150-word design brief that names the specific visual DNA: proportions (face width:height ratio, head-to-body ratio), materials (matte black + brushed titanium + soft-touch rubber on suit seams), and mood language (clean geometric, cold, tactical — NOT 'futuristic' or 'edgy').
    • Audit the AI's interpretation against your brief. Generate Kess once with the character designer's seed. Compare outputs against your 3 reference images. Log 3 concrete mismatches (measurements, material reads, proportion errors) with photo evidence and the correction needed (prompt revision, seed adjustment, or reframing).
    • Author the "design integrity" section of the production bible: the 300-word design rationale that explains WHY Kess looks this way (the character's world role, the costume's function, the material choices). Future AI generations will use this to resolve ambiguity.

  • Asset Wrangler

    Commands the seed pipeline. You document the technical reproducibility and maintain the master seed log.

    • Build the seed strategy document: write down the EXACT prompt chain used to lock Kess (base character prompt + scene-specific modifiers for cockpit/exterior/lab). Include parameter values (model version, guidance scale, seed number, image strength). Make it so reproducible that a new team member could execute it verbatim 6 months from now and get the same Kess.
    • Run a 2x2 grid test: generate Kess in 2 different scene contexts with the SAME seed, then with 2 DIFFERENT seeds in the same context. Produce a 1-page visual matrix with the outputs + annotations showing (a) which variables held consistency and (b) where seed fails or prompt dominates. Title it 'Reproducibility Map'.
    • Maintain the master seed log: a table with 5+ entries (date, prompt, seed, scene, output hash/filename, confidence score 1-5, notes). This is the studio's insurance policy. Flag any row where confidence < 3 with the reason and the corrective action taken.

  • Production Manager (optional, if 4th needed)

    Owns the timeline and the handoff. You make the bible usable by a studio shooting schedule.

    • Write a one-page deployment brief for the VFX team: here's the seed, here's the prompt, here's the test we ran, here are the known failure modes (e.g., 'seed drifts on extreme closeups; use these fallback seeds for CU work'). Make it so clear a new VFX artist can start work in 5 minutes.
    • Schedule the next 3 iterations: which parameter to lock next, what to test, what's the decision gate before rolling wide shots. Write it as a simple checklist tied to the studio's 48-hour deadline.

Exemplars

  • Golden Trailer Awards

    Golden Trailer Awards

    The industry awards body for movie trailers — the exemplar bar for what a finished, professional-grade trailer looks and sounds like.

  • Runway AI Film Festival

    Runway

    The premier festival showcase of finished AI films — the "this is what pro AI filmmaking looks like" gallery. Complements the Golden Trailer Awards craft bar.

Seoul AI 2026