How to Build a Vertical Drama Character Bible for AI Generation

Character consistency across 60 episodes is solvable only if you lock reference assets early and don't deviate. The character bible is the document that makes not deviating systematic rather than aspirational.

The character bible for live-action production is primarily a casting and performance guide: what the character looks like, how they speak, what their backstory is, and how they develop across the arc. The casting director uses it to find actors who embody the character. The actor uses it to develop a consistent performance. Both are human-mediated processes where judgment and interpretation fill the gaps the document leaves.

The character bible for AI generation is different in a specific way: the generation operator does not use judgment to fill gaps. They translate specifications into prompt parameters and reference pack configurations. A specification that is absent from the character bible is a decision the operator makes from their own interpretation — which is the deviation that produces character consistency failure at episode thirty.

This post covers the complete character bible structure for AI generation: every section it must contain, what each section specifies, and why each specification prevents the specific character consistency failure it addresses.

Section 1: The Psychological Configuration

The psychological configuration is the character bible's most important section for AI generation because it is the specification that every close-up performance prompt derives from. The generation operator who does not have a psychological configuration produces visual outputs that are aesthetically correct but emotionally generic. The operator who has a precise psychological configuration produces outputs whose emotional register is consistent with the character's specific psychology across 70 episodes.

The psychological configuration contains three elements:

The suppressed interior. What the character is actually experiencing beneath their controlled exterior. Not their backstory. Their current psychological state in the series' present moment. This is the emotional content that the involuntary tell reveals against the character's intention.

The performed exterior. What the character presents to other characters in the story. This is what other characters in the story see. The generation operator needs both layers to produce the gap between them that the involuntary tell reveals.

The involuntary tell. The specific physical behaviour that occurs when the suppressed interior breaks through the performed exterior against the character's intention. This is specified in physical, not emotional, terms because physical specification is translatable into generation prompt parameters.

Example — controlled alpha psychological configuration:

Suppressed interior: involuntary recognition of the protagonist as an equal, experienced as threat rather than attraction. He reads her correctly before he understands what he's reading.

Performed exterior: institutional authority, precision of movement, controlled vocal register, zero visible affect. Eyes forward, jaw level, body language economy. No unnecessary motion.

Involuntary tell: jaw sets fractionally at the moment his attention is captured against his intention. Not dramatic. Microscopic. Visible only in close-up. The only physical deviation from his standard movement economy. Generate this tell at every scene where the protagonist enters his attention field against his active resistance.

Section 2: The Tag Stack

The tag stack is the three-label shorthand system that allows writers, generation operators, and creative directors to reference the character's configuration without reading the full psychological specification at every use.

The tag stack format: Role Tag / Psychological Tag / Relational Tag.

The role tag identifies the character's function in the story's power dynamic. The psychological tag identifies the character's primary suppressed interior state. The relational tag identifies the character's specific posture toward the protagonist.

Example tag stacks:

Controlled Alpha: Institutional Authority / Involuntary Recognition / Resistant Investment

Protagonist: Strategic Concealment / Earned Capability / Active Refusal

Antagonist: Performed Legitimacy / Entitlement Architecture / Displaced Threat

The tag stack appears in every scene-level direction brief. The generation operator reads the tag stack as the first instruction before reading the scene's specific prompt. The tag stack confirms the character's baseline psychological state before the scene's specific emotional register is applied.

Section 3: The Visual Specification

The visual specification is the section that live-action character bibles contain in abbreviated form and AI generation character bibles must contain in precise technical form.

A live-action bible might say: "early 40s, dark hair, sharp features, always in expensive suits." The casting director translates this into actor selection criteria. The selected actor's face becomes the visual reference.

An AI generation bible must say the same thing in generation-translatable terms:

Facial structure specification:

Jaw: defined, angular, slight prominence at mandible angle. Jaw breadth at widest point: approximately 85% of cheekbone breadth. Not square, not soft. Precise geometry.

Eyes: dark brown to near-black iris, slight hooding of upper lid, horizontal eye shape rather than almond. Eye spacing: slightly wider than standard template. Upper lid line level, not arched.

Nose: straight bridge, no deviation. Tip: medium definition. Nostril width: proportionate, not broad.

Skin tone: medium warm — Fitzpatrick type III to IV, warm undertone. Even texture. No visible pores. Very slight shadow in jawline hollows.

Hair: dark brown to black, worn back from face, no visible scalp, medium-dense texture, no styling texture visible.

The level of specificity in this section is what determines whether the generation operator can reproduce the same character in session eight that was established in session one. Vague visual descriptions ("dark, sharp features") produce visual approximations that drift between sessions. Specific technical descriptions produce reproducible parameters.

Section 4: The Arc Position Wardrobe States

The arc position wardrobe states specify how the character's visual appearance evolves across the series' three arc blocks as their internal state changes. This is the section that makes wardrobe change systematic rather than arbitrary.

A character whose wardrobe state is specified across three arc positions can be generated with the correct wardrobe state for any episode based on its arc position. The generation operator does not make wardrobe decisions. They load the reference images for the episode's arc position and generate.

Example — controlled alpha wardrobe arc:

Arc Position 1 (Episodes 1 to 25 — institutional authority phase):
Full charcoal or navy suit, no texture variation. White shirt, no pattern. Dark solid tie, conservative width. Jacket always fastened in public spaces. Zero wardrobe informality.

Arc Position 2 (Episodes 26 to 50 — first involuntary investment phase):
Suit maintained but jacket occasionally removed in private spaces. First appearance of subtle texture in jacket fabric. Collar button occasionally unfastened. These are not chosen departures — they are signs that maintenance of the exterior is requiring more effort.

Arc Position 3 (Episodes 51 to 70 — acknowledged investment phase):
First appearance of shirt without jacket in scenes with the protagonist only. No casual register otherwise. The wardrobe departure is specific to her presence and invisible to institutional contexts.

The wardrobe arc makes the character's visual evolution a specification rather than an interpretation. The generation operator who loads the correct arc position reference images generates the correct wardrobe state for every episode without making any wardrobe decisions themselves.

Section 5: The Session Reproduction Specification

The session reproduction specification is the character bible section that is absent from almost every production company's character bible and whose absence is the most common cause of cross-session character drift.

The session reproduction specification documents the exact reference configuration that was used to generate the approved reference images for each arc position. It contains: the reference image source and resolution, the Soul ID model ID or equivalent character model identifier, the reference conditioning strength applied in generation, and the approved reference frame for each arc position that subsequent sessions use as their consistency check.

Example:

Arc Position 1 Approved Reference Frame: [filename].jpg, generated session 1, day 3. Soul ID model ID: [identifier]. Reference conditioning strength: 0.85. IP-Adapter weight: 0.80.

Session-open consistency check: generate one test output using the above configuration before any production generation begins. Compare jaw structure, eye colour, and skin tone against the approved reference frame. Proceed only if all three match within visual tolerance. If drift is detected, review Soul ID model configuration before proceeding.

The session reproduction specification converts the character consistency check from a subjective evaluation into a documented procedure. The operator who follows it produces consistent character identity across sessions. The operator without it makes judgment calls that accumulate into visible drift by episode forty-five.

Section 6: The Direction Brief Shorthand

The direction brief shorthand section translates the psychological configuration into the specific prompt language the generation operator uses in every scene where the character appears.

The shorthand contains: the standard performance prompt for the character's neutral state, the performance prompt modification for each emotional register the character transitions through in the arc, and the specific physical instructions for the involuntary tell at each activation threshold.

Example:

Neutral state prompt: "Controlled stillness. Eyes forward, level, no visible affect. Jaw level. Economy of movement. Authority posture without aggression."

Involuntary recognition modification: "Add to neutral state: fractional jaw set, duration one to two frames, immediately released. Eyes track toward protagonist entry point, one-frame deviation from forward orientation, then return. No other deviations."

Acknowledged vulnerability modification: "Jacket removed. First button unfastened. Posture maintained but one degree of economy released — not visible to other characters but visible to camera in close-up only."

The direction brief shorthand is the character bible's operational layer. Every generation operator who uses it produces the same character in the same emotional register regardless of which sessions they are working in.

How to Build the Character Bible Before Production Begins

The character bible is built during the brief development session and the first week of pre-production. It is not built during generation. A character bible built during generation is being built from the outputs rather than used to guide them — which reverses the correct production order.

The build sequence:

Brief development session: psychological configuration and tag stack are developed from the brief's character configuration specifications.

Week 1 pre-production: visual specification is developed from the psychological configuration. The generation operator builds the initial character reference images from the visual specification. The approved reference images are used to build the Soul ID character model. The approved reference frame is documented in the session reproduction specification.

Week 1 gate: commissioning party reviews the approved reference images on a consumer phone against the brief's character configuration specifications. Arc position wardrobe state images are approved for each arc block. The character bible's session reproduction specification is completed with the approved reference frame identifiers.

Week 2 onwards: every generation session begins with the session-open consistency check against the approved reference frame. Every direction brief uses the direction brief shorthand from the character bible.

Axis AI Studios Perspective

The character bible described in this post is the document that makes character consistency systematic rather than dependent on operator memory or judgment. At Axis AI Studios, the character bible is built during week one of every production before any scripts are commissioned. The session reproduction specification and direction brief shorthand are the sections most commonly absent from character bibles we see from other production companies — and their absence is the cause of the cross-session drift that requires correction before delivery.

For generators and operators who want to understand how to work from a character bible at production standard, the character bible is the primary reference document for every generation session. The session-open consistency check is the first task of every session, before any production generation begins.

For businesses commissioning AI-native vertical drama who want to confirm their production partner builds and uses a character bible at this standard, the due diligence checklist covers the character consistency infrastructure question directly.

Reach out at business@axisaistudios.com for businesses wanting to commission or for operators wanting to join the team.

FAQ

How Many Characters Does a Character Bible Cover?

Every character who appears in more than three episodes requires a character bible entry at the full specification level. Supporting characters who appear in five or fewer episodes require a visual specification and direction brief shorthand but not the full psychological configuration or arc position wardrobe arc. Background characters do not require character bible entries.

Does the Character Bible Change Between the First Series and a Sequel?

The visual specification, session reproduction specification, and direction brief shorthand carry forward from the first series to the sequel without modification unless the character's arc position wardrobe state changes in the sequel's premise. The psychological configuration may require updates if the sequel's premise positions the character in a significantly different internal state from where the first series ended. The tag stack is updated to reflect the sequel's new relational posture.

Can a Character Bible Be Built Retrospectively for a Production Already in Progress?

Yes, but the session reproduction specification must be built from the approved reference frames of the outputs already generated rather than from a deliberately constructed initial reference. The retrospective character bible is less precise than a pre-production character bible because the approved reference frames from early sessions may not be as precisely controlled as reference frames built specifically to be the production's visual standard. For future productions in the same franchise, the retrospective bible is used as the starting point and refined into a properly structured pre-production document.

Further Reading

For the Soul ID character model training that the session reproduction specification section of this post documents, the guide to using LoRA training for character consistency in vertical drama covers the complete training process and deployment configuration.

For the character asset library that the character bible feeds into across multiple series in a franchise, the guide to building an AI character asset library covers version control, arc position variant management, and franchise consistency infrastructure.

For the generation operator role that uses the character bible as their primary production reference, the guide to the generation operator's role in AI-native vertical drama covers the session-open consistency check procedure and batch generation documentation discipline.