The Character Bible: Locking Identity Before the First Frame
Every frame of a finished piece inherits its identity from one locked reference sheet — build it wrong, or swap it mid-production, and the drift shows up in every shot after.
Why identity has to be locked before anything else
Every finished piece — a fifteen-second reel, a four-part short, a full production — is built from dozens of separate generation calls. Different scenes, different angles, different moments, often generated hours or days apart, sometimes by different people. None of that matters to the viewer if one thing holds constant across every single call: the character looks like the same character.
That constancy does not happen by accident, and it does not happen by re-describing the character in every prompt. Text descriptions are lossy — "a woman in her thirties with dark hair and a leather jacket" produces a different specific face every time it's generated, because the model is filling in everything the words didn't specify. The only reliable way to hold identity constant across dozens of independent generations is to generate it once, lock it, and reference the same image file every time after.
That is what a character bible is. It is not a style guide or a mood board. It is the single reference image every downstream generation — every expression, every angle, every wardrobe change, every action pose — points back to. Get it right once, and the rest of the production inherits that correctness for free. Get it wrong, or worse, swap it out partway through, and every shot after the swap visibly disagrees with every shot before it.
This lesson covers the first two stages of a five-stage pre-production pipeline: Anchor, Master Bible, Action Grid, Grid-to-Video, and Finish. Stage 2, the action grid, gets a full lesson next — here it's previewed only, so you can see where the bible you're about to build gets used.
Stage 0: the anchor
The anchor is the raw material the entire bible is built from, and it has one hard requirement: it has to be genuine. Not a casual grab, not a low-resolution export, not a frame where the subject is mid-blink or lit from a bad angle. A real, clean source frame — a properly lit photo, or a frame pulled from real footage at a moment where the face is legible and the expression is neutral or close to it.
You need two anchors, and they don't have to come from the same source image. The face anchor establishes the features an image model needs to lock identity: bone structure, eye shape and spacing, skin tone, hairline. The wardrobe anchor establishes the garment, materials, and silhouette the character is defined by — a signature jacket, a specific color combination, a piece that reads as unmistakably theirs. If the cleanest available shot of the face happens to be in different clothing than the wardrobe you need for the production, that's fine — pull two separate anchors and let the master bible generation reconcile them into one sheet.
What you should never do is default to whatever snapshot happens to be sitting closest to hand. The anchor is the seed. A blurry, awkwardly lit, off-angle anchor produces a bible with the same defects baked into every one of its ten reference elements, and those defects compound into every shot the bible is used to generate. Ten minutes spent finding or capturing a genuinely clean anchor pair saves hours of downstream drift-chasing.
Stage 1: the master bible
With both anchors locked, the next step is a single generation call that produces one high-resolution reference sheet — the master bible. This is not ten separate generations stitched together; it's one image-model call that references both anchor frames simultaneously and produces a complete, internally consistent character reference in one shot.
A production-grade bible sheet contains ten distinct elements, and skipping any of them means that piece of the character's identity was never locked — it gets improvised fresh, and inconsistently, every time a downstream shot needs it.
The info block anchors the character's basic facts so nothing has to be re-derived from memory later. The color palette locks the exact tones — skin, hair, wardrobe — so a shot generated in week three doesn't drift half a shade warmer than one generated in week one. The scale and proportion sheet, shown from multiple angles side by side, is what keeps the character's build consistent whether they're framed in a wide establishing shot or a tight close-up.
The eight expressions and the handful of subtler micro-expressions matter more than they look like they should. A character who has only ever been generated smiling will fight the model when a later scene calls for controlled anger — the model has no reference for what that character's anger looks like, so it improvises, and the improvisation often doesn't match the locked face. Generating the emotional range up front means every later scene has a true reference to pull from instead of a guess.
The five head angles — front, three-quarter, profile, and rear coverage — exist for the same reason a live-action production shoots reference turnarounds: so that any camera angle a later scene calls for already has ground truth to match against. Posture, wardrobe callouts, the signature prop, and hand-gesture references round out the sheet with the physical details that make a character legible as a specific person rather than a generic type.
Review, pick, lock
Generating candidate bible sheets is fast and cheap relative to the value they unlock, which means the temptation is to grab the first output that looks roughly right and move straight into production. Resist that. The discipline here has three steps, always in this order: review, pick, lock.
Review means putting every candidate sheet next to the anchors and checking it honestly — does the face genuinely match the anchor, or has it drifted toward a generic version of "attractive person in their thirties"? Are all ten elements present, legible, and internally consistent with each other? Pick means choosing the single strongest candidate — not a composite, not "close enough," one sheet. Lock means saving that one file under an explicit, unambiguous name that marks it as the production's single reference — not "bible_v3_final_ish.png" sitting in a folder of six near-identical siblings, but a file every collaborator can find without guessing.
The bible is pre-production, not per-shot
The most common mistake teams make with character bibles is accounting for them wrong. A one-off ad hoc reference generated for a single shot feels cheap in the moment — one quick call, move on. But that cost repeats on every single shot that needs a reference, and worse, nothing guarantees any two of those ad hoc references agree with each other. The bible flips that math: one investment, paid once, at the start, that every subsequent shot in the entire production draws from for free.
This holds even for small productions. A three-shot piece with a character whose face subtly disagrees between shot one and shot three reads as broken to a viewer, even if they can't articulate why. A thirty-shot production built on one locked bible reads as a single coherent production, because it is one — every shot is generating from the same ground truth. Treat the bible generation as pre-production overhead you pay once, not as a cost you're trying to avoid per shot.
This is exactly the discipline behind real produced work — a crimson-vs-blue twin-rivalry action short built for @jeremyknox.ai held two distinct locked identities across a dozen separate scenes, generated over several sessions, precisely because both characters' bibles were locked once at the start and never regenerated.
What comes next
The locked bible you build in this lesson is the identity anchor for everything that follows in this track — the action grid in the next lesson references it directly, the continuity law lesson after that treats it as the constant every scene chains back to, and the capstone production assembles a finished piece built entirely on top of it. It's also the same discipline referenced throughout the companion track on agentic film studios: whatever pipeline generates the shots, the bible is what keeps the character the same character across all of them.
Get this stage right, and every later stage in the pipeline gets easier. Get it wrong, and no amount of careful prompting downstream can fully undo the drift.