How do you create a character bible for AI video to lock character appearance across scenes?

A character bible for AI video is a locked set of pre-production assets: multi-angle character sheets (front, side, back, face close-up) generated at 4K, four options per character with the best one locked, stored in your agent's context and attached to every generation prompt. One production held two characters consistent across a 70-second film this way — no LoRA fine-tuning required.

Build the bible before you generate a single second of video — locking character references first is the step that prevents consistency problems through the rest of the film. invideo is an agentic video creation tool with all the current image and video models available, so the whole workflow below runs in one place.

1. Cast the character with hero portraits. Generate face portraits first and judge them like casting headshots. Recraft renders skin-level imperfections — pores, lines, stubble — that keep faces photorealistic, while GPT-Image-2 and Nano Banana handle stylized or illustrated characters; the invideo agent routes to whichever fits your film. To choose a look faster, run the identical character prompt on two image models in parallel and pick the aesthetic you prefer.

2. Build the turnaround sheet. Convert the approved portrait into a 360-degree character sheet: four angles (front, side, profile, back) plus a face close-up and a mid-angle close-up, generated at 4K. Include close-up panels deliberately — small details like scars and accessories drift first if the model only ever sees wide views. Remove any objects from the character's hands before generating the turnaround, or the prop will appear inconsistently across angles.

3. Generate options, then lock. Produce four variations of each character sheet, select the best, and lock it before any video generation begins. Budget for iteration: one documented production averaged 5 generations to lock a single character at roughly $9.78 per character, and needed only 11 images total to cover reference sheets for 4 characters and 1 prop.

4. Save the bible to persistent context. Upload the locked sheets to the invideo agent with an explicit instruction to store them — the working prompt language from one production was a direct command to deeply understand the references and save them to context for all further generations. Then attach the character references to every single generation prompt, with no exceptions; consistency comes from the discipline of never generating a shot without the bible attached.

5. Make a sheet per appearance beat. If your character's costume or accessories evolve across the story, create a distinct character sheet for each beat rather than stretching one sheet across the whole film — one production needed a separate sheet for every sequence because the character picked up a new trinket in each location.

6. Fix drift at the source, not the shot. When a continuity error shows up in a generated shot, don't re-roll the shot — ask the invideo agent to inspect the character sheet itself. In one documented case the invideo agent identified the exact panel in the character grid containing the error, corrected it, and stored the updated sheet so every subsequent shot inherited the fix while the rest of the film stayed intact.

When you move to video generation, the bible pays off most with models that accept character references directly: Seedance 2.0 reference-to-video ingests character sheets alongside location references, which carries identity across clips in a way that frame-based extend methods can't — and since every roster model runs inside invideo, the invideo agent attaches your locked sheets to whichever model each shot needs. The proof this works without fine-tuning: a 70-second short film held 2 characters visually identical across every scene using only character sheets and agent context, and a 2-day, $950 animated episode locked its full cast the same way.

Watch some of these to see what works for you:

the AI always needs to see what the character is exactly, right? Or else it'll kind of hallucinate and imagine something that's under the cap. So, we don't want to do that. We always want the character to be seen as we see it on the character sheet.

— invideo's creative team