AI video character passport: document what must stay recognizable.

Open the passport with authority and identity boundaries

Give the character a stable internal identifier, then record who owns the design, which performer or likeness permissions apply, where each reference came from, and what uses are allowed. Separate fictional design traits from attributes of a real person. List approved face, hair, wardrobe, voice, movement, and prop references with file hashes or durable asset IDs; mark missing angles or expressions instead of guessing them. The AI video character passport should also name forbidden transformations, disclosure obligations, expiration dates, and the reviewer who can approve a new source. A reference image is evidence of appearance, not blanket consent, an identity guarantee, or proof that a generation system will preserve it.

• Stable character and asset IDs
• Rights, consent, and use scope
• Approved sources and missing views
• Prohibited changes and escalation owner

Translate the design into observable continuity checks

Divide the record into invariants, controlled variables, and scene state. Invariants might cover facial structure, distinguishing marks, silhouette, palette, voice identity, or a signature prop. Controlled variables can include wardrobe, age treatment, lighting, emotion, damage, and styling when the story authorizes them. Scene state tracks what changes through time: dirt, wetness, injuries, carried objects, location, knowledge, eyeline, and emotional beat. Write every item so a reviewer can point to visible or audible evidence. Avoid vague labels such as cinematic or consistent. Add reference frames for front, profile, full-body, action, and expression only when those assets are authorized and actually available.

Version the passport beside attempts and edit decisions

Freeze a passport version before a shot batch, attach it to the exact brief and source files, and log the visible product context used for each attempt. Review candidates in sequence for facial drift, body proportion, costume construction, voice change, prop mutation, spatial continuity, and story-state errors. Preserve rejected outputs with reason codes when policy permits; a chosen highlight alone cannot demonstrate continuity. When the story intentionally changes the character, update the state ledger without rewriting prior requirements. Close each production round with the approved version, unresolved exceptions, reviewer disposition, and retest trigger. This method creates an auditable handoff; it does not establish a SEELE passport feature or promise repeatable identity preservation.