AI character sheet index for consistent story characters
An AI character sheet is a production reference that fixes identity, costume, props, expressions, and motion states before you generate story shots. Choose the sheet that matches the continuity problem you need to solve.
Vixel AI character sheet index
Choose the production reference that matches the continuity problem you need to solve. These systems and prompt structures are authored by Vixel.
Identity lock turnaround
Establish the one approved identity baseline that every later story shot inherits.
Required views: Face close-up, Front, Three-quarter, Profile, Back.
GPT Image 2 structure: One original character across matched front, three-quarter, profile, and back views; preserve named identity anchors, scale, neutral light, and a clean unbranded background.
Seedance handoff: Use the approved board plus one start frame; define a single pose change, one camera move, protected identity anchors, and the exact final gaze and screen direction.
Expression and performance ladder
Vary emotion and dialogue performance without turning the character into a different actor.
Required views: Neutral, Suspicion, Fear, Anger, Relief, Resolve.
GPT Image 2 structure: Six matched head-and-shoulder portraits of the same original character; hold face, hairline, age, wardrobe, lens, and light while only expression, gaze, and head tilt change.
Seedance handoff: Name the starting and ending emotion, then direct the eye, brow, breath, and jaw transition under a locked frame or one slow push.
Wardrobe and prop continuity board
Prevent costume layers, accessories, and story-critical props from mutating or changing owners.
Required views: Outfit front, Three-quarter, Back, Material details, Prop held, Prop stowed.
GPT Image 2 structure: One outfit and prop system across matched views; lock layer order, cut, fasteners, material, wear side, prop geometry, and held-versus-stowed states.
Seedance handoff: Describe a reach, grip, use, and settle sequence; protect contact hand, garment layers, prop geometry, and the final prop state.
Body-language acting range
Give one identity a readable range of power, pressure, openness, and vulnerability.
Required views: Neutral, Open, Guarded, Dominant, Vulnerable, Action-ready.
GPT Image 2 structure: Six full-body acting poses for the same character; preserve identity and costume while varying weight, spine, hands, feet, and silhouette under one camera setup.
Seedance handoff: Move from pose A to pose B with a readable weight transfer, a declared planted foot, no limb crossing, and a clean final silhouette.
Dialogue coverage and eyeline board
Lock the axis, eyelines, lens scale, and reaction coverage needed for an editable conversation.
Required views: 35 mm medium, 50 mm MCU, 85 mm close-up, Over-shoulder, Listening reaction.
GPT Image 2 structure: Matched dialogue coverage in one scene; preserve eyeline, screen side, costume, prop state, light direction, and room geometry across framing changes.
Seedance handoff: Direct one listen-register-respond beat, keep the eyeline and screen side stable, and finish on a cut-ready reaction.
Cast scale and relationship board
Keep multiple characters distinct while preserving height, spacing, screen side, and prop ownership.
Required views: Cast lineup, Paired three-quarter, Profile confrontation, Seated/standing scale, Prop scale.
GPT Image 2 structure: A lineup and paired views with distinct faces, silhouettes, height ratios, palette tokens, and owned props; never swap wardrobe or objects.
Seedance handoff: Give character A one action and character B one reaction; preserve identities, height ratio, screen sides, spacing, and prop ownership.
Action physics sequence
Approve contact, force direction, recoil, and recovery before spending video-generation credits.
Required views: Anticipation, Contact, Peak action, Recoil, Settle.
GPT Image 2 structure: Five chronological states of one action; keep identity, costume, prop, camera side, and screen direction while making contact and force physically legible.
Seedance handoff: Animate anticipation through contact and recovery; declare the contact point, force direction, one camera move, and exact settled state.
Branch-state continuity matrix
Separate what must stay canonical from the visible consequence each choice is allowed to change.
Required views: Pre-choice baseline, Route A state, Route B state, Ending state.
GPT Image 2 structure: The same character before a choice and across declared route consequences; protect canonical identity and base wardrobe, changing only the listed route deltas.
Seedance handoff: Reference one approved route state, animate only its consequence, preserve inherited invariants, and end at the state required by the next beat.
Seedance start-to-end handoff
Turn a character board into a short, executable motion contract with a usable next-shot anchor.
Required views: Approved first frame, Motion checkpoint, Approved end frame, Depth layout.
GPT Image 2 structure: Matched start, midpoint, and end frames for one continuous action; preserve identity, environment geometry, wardrobe, prop, camera side, and light direction.
Seedance handoff: Reference the approved identity board and start frame, direct one timed action arc and one camera move, then finish on the approved end composition.
Method
- Build an identity board: Approve front, profile, three-quarter, expression, wardrobe, and prop views before treating any single beauty image as the character source.
- Choose high-information invariants: Keep a concise canonical block for facial structure, hair shape, silhouette, wardrobe anchors, palette, and signature objects.
- Separate identity from performance: Put action, emotion, blocking, lens, and lighting in the shot brief so they can change without rewriting who the character is.
- Review drift before motion: Reject identity or wardrobe drift at the image stage; video generation amplifies ambiguous anchors instead of repairing them.
Common failure modes
Every image looks like a different actor
Reduce generic adjectives and preserve a compact, distinctive identity block plus approved visual references.
The character is consistent but emotionally flat
Keep identity invariants fixed while making expression, posture, gesture, and gaze explicitly shot-specific.
Wardrobe details mutate
Name only the wardrobe anchors that matter to recognition and continuity, including color, cut, material, and signature accessory.
Related generated recipe proofs
- Timed cast-and-shot handoff - Pair a clean cast identity strip with a timed shot strip: the cast strip locks face, silhouette, wardrobe, and prop ownership; the shot strip locks start pose, camera lane, action…
- Character-sheet motion bible - Build a character bible first: locked silhouette, face geometry, wardrobe tokens, expression ladder, and prop ownership, then generate branch frames from that bible.
- Character-sheet first motion reference pack - Create a character sheet first in GPT Image 2, then use it as the identity authority for every storyboard and Seedance clip.
- Compact visual-bible repair sheet - Before video, create a compact visual bible that distinguishes fixed identity tokens from route-variable tokens and final consequence targets.
- Character-sheet video continuity board - Build a compact identity sheet before motion: face geometry, hair silhouette, outfit tokens, expression set, and action-start poses become the checksum for Seedance rather than ho…
- Dual-reference cast storyboard handoff - Use two explicit references for video generation: a clean cast/identity sheet locks face, silhouette, wardrobe, and prop ownership; a storyboard board locks shot order, camera lan…
- Sixteen-panel picturebook continuity sheet - Use a sixteen-panel sheet as continuity proof: lock protagonist, prop, location, choice anchor, and consequence across a whole short-session route before generating individual key…
- Character-sheet storyboard fusion board - Pair a character sheet with a compact storyboard plan: identity, outfit, silhouette, and prop are locked in the character layer, while shot order, camera intent, and consequence a…
150 character references for Canvas
Collection by Kōda.
A 150-reference collection spanning stylized 2D, 3D, realistic, and cinematic characters, ready to become identity anchors for GPT Image 2 frames and Seedance motion.
Hosted by Vixel with Kōda's permission for an attributed asset library. Public presentation uses a name-only Kōda credit, while durable creator and source provenance remains attached to every private Canvas import. Creator Standard or Pro unlocks Vixel's hosted index and workflow—not ownership, exclusivity, or a claim that the characters are Vixel originals.
Preview the collection and Canvas workflow
Frequently asked questions
Which AI character sheet should I make first?
Start with an identity lock turnaround: face close-up plus front, three-quarter, profile, and back views under matched scale and neutral light. Add expression, wardrobe, action, cast, branch, or motion sheets only when that continuity risk becomes real.
How do I keep the same AI character across scenes?
Approve an identity board, keep one concise invariant identity block, attach stable references, and put pose, emotion, camera, and lighting in a separate shot block.
Should every shot reuse the exact same prompt?
No. Reuse the identity invariants, then write a new shot delta for action, emotion, camera, location, and lighting. Repeating the entire prompt can freeze performance without preventing drift.
Does Vixel host Kōda's 150 character sheets?
Yes. Vixel hosts the collection with Kōda's permission and a visible name-only credit. Public visitors can inspect six representative previews; Creator Standard or Pro unlocks the searchable 150-sheet index and zero-credit Canvas import. Durable creator and source provenance stays attached to the private Work, and Vixel does not claim ownership or exclusivity.