keeping a character on the same face throughout

Casting solves one thing: keeping the same character on the same face from beginning to end.

Create an actor, lock their face, and everything after that — makeup, wardrobe, cover shots, character cards — is anchored to that face. Any shoot segment that brings this character in uses it automatically — along with the voice you set for them.

Fastest route: three steps to your first actor

1. Create the actor — sidebar "Create → Casting" → "New actor". Two ways in:

• Generate from text — fill in the actor's name, the identity lock, and a look description, and let the AI generate one.
• Upload — bring a photo you already have.

2. Confirm the locked image — whatever is generated becomes the identity lock image. Not happy? Press "↻ Regenerate" for another take.

3. Change the makeup and wardrobe — pick templates on the workbench, add your notes, press "✦ Generate". Results land in "Casting results" on the right.

"Identity lock" is the field that matters most

When you create an actor there is an "identity lock" field. Write down the features of this person that never change, for example:

oval face, monolids, black shoulder-length bob, small mole on the right cheek

That string travels with the actor. The more specific it is, the less the face drifts when you change wardrobe or makeup.

The "look description" is what the AI generates the first image from — for example "a 27-year-old urban woman, 167cm, slim, front-facing half-body shot in natural daylight".

The cast board

Casting opens on the cast board — all of your actors.

You can search by actor name or feature at the top, and filter by project.

A row of small badges under each card shows at a glance what this one is still missing:

The badges are shortcuts, not just status — tap a missing one and it jumps to where you fill it in. ("+ Voice" only shows on paid plans.)

Press "Enter the casting room →" to start work.

Using the casting room

The screen is split: workbench on the left, casting results on the right. (On mobile these are two tabs at the top.)

Along the top is a row of actor portraits — hop to another actor with one tap, no need to go back to the cast board. "← Cast board" on the left takes you back.

Top left is the identity lock image — every generation is anchored to it. Below it are this actor's badges and their assignments (which projects, which group); at the far right of the assignment row is "Edit identity".

The workbench's five tabs

Makeup / Hair / Wardrobe / Accessories / Cover

Each tab has three blocks under it:

Templates — ready-made options, and you can pick several. Makeup has natural, smoky evening, Korean, Japanese, retro red lip and more; hair has long, short, buns, big waves, high ponytail, updos; cover has VOGUE style, GQ style, album cover, film poster and so on.

Look name — optional, but worth filling in. The interface says why: naming it is what lets you tell looks apart in the picker and the character library. Make five outfits all called "Untitled" and you will not know which is which later. Try "outlaw outfit" or "rain-battle version".

Notes — optional. Add the details you want, for example "red velvet gown, cool studio lighting, title in the top right".

Reference images — optional. You can attach a hair reference, a makeup reference and a wardrobe reference separately so the look follows the image you specify.

Then pick the model, ratio and size, and press "✦ Generate · N credits".

Three kinds of character card

One button on the workbench outputs a character card; the arrow beside it switches the layout:

The third deserves a note. On a wide shot the model can only guess the face from a tiny area, and the face goes soft. The actor sheet supplies a large head close-up as the source for the face, so that does not happen. It is deliberately flat-lit on grey, with real pores and no cinematic grade — it is a spec photo, not a portfolio shot.

Casting results

The right-hand column holds all of this actor's looks.

You can filter at the top: All / Hair & makeup / Wardrobe / Cover / Own.

What you can do with each look:

• Reuse — load this one's notes and reference images back into the workbench, tweak and regenerate.
• Rename — double-click the title.
• Delete.

You can also "Upload a look" to bring in images you already have — no credits.

Managing the identity

"Edit identity" at the far right of the assignment row opens a drawer with five sections:

Main image — the current identity lock image. "⬆ Upload a replacement" swaps in your own.

Identity — change the actor's name and the identity lock.

Voice — bind a reference audio clip as this actor's voice. See "Voice: the file and the text description" below. (Paid-plan feature.)

Acting — this actor's acting profile: how they perform, what their voice is like. Once saved, every segment that brings this character in works it into the scene at shoot time. (Paid-plan feature.)

It has three blocks:

• One line of personality — for example "a sniper who is arrogant outside, brittle inside"; press "Write it with AI" and the AI drafts the full profile.
• Behaviour — behaviour only, and every action needs its trigger. For example "cracks his knuckles during small talk, to fake ease". Two or three habitual actions, plus one or two signature moves you would recognise him by (how he walks, sits, reacts the instant someone calls his name). Do not write emotion labels, build or looks — those belong in other fields.
• Voice identity — see the next section.

⚠️ The character budget is shared: with several people in one scene each gets less (about 180 characters each with four in the scene). Precise beats long.

Advanced — deleting the actor, and so on.

Setting the voice

Under the "Voice identity" field are three sets of chips; tapping them composes the description:

Accent is especially worth setting. Left unspecified, the AI drifts toward mainland Mandarin; if you want a Taiwanese accent, tap that chip.

You can also skip the chips and type it yourself (for example "low and hoarse, slow, quieter the more serious it gets"). Tap no chips at all and it is left to the AI; tap any and the field is composed from the chips, and AI generation will not overwrite it.

Voice: the file and the text description

There are two ways to give a character a voice:

⚠️ It is one or the other, never both. For any one character in any one segment only a single track is used: if a file is bound, the file goes and that character's text description does not; the text description is used when nothing is bound, or when the file gets squeezed out by the budget. (The rest of the acting profile — the habitual and signature behaviour — is unaffected and always goes.)

So "will writing both make it more accurate?" — no. Write the text description anyway, though: it is the fallback in three situations — no slot left, first/last frame turned on for that segment, and the day you unbind the file.

A bound file is not a 100% match either. It points the model at a timbre and a way of speaking; it does not clone the voice. If a bound file still is not close enough, unbinding it and writing the text description instead often works better than trying another recording — text is an explicit instruction (Taiwanese accent, quieter the more serious it gets) and is more controllable than a sample.

Binding and unbinding both cost no credits. Binding is a paid-plan feature; unbinding is not (you can still clear one after downgrading).

For the full rules — how the budget is counted, and the cases where nothing gets carried at all (first/last frame, Studio quick shots, the knock-on effect of attaching one more reference image) — see the Character Voice Handbook.

Editing again, and versions

The drawer has "✎ Edit again", which changes the main image by description — "make the hair black and short", "soften the expression".

Each edit adds a version, and the version row below shows them all; pick one and press "Set as main image" to switch. The original does not disappear.

Adding to the Library

"Add to the Library" in the drawer sends this actor and their looks to the Library, so they can be picked elsewhere (as shoot reference images, for instance).

Where things are

Beyond the basics

• Three kinds of reference image — hair, makeup and wardrobe each take their own, which is more precise than cramming everything into text.
• Templates stack — pick "Korean" and "big waves" together, then fill in the detail with notes.
• Groups and project filters — once you have a lot of actors, assign them to projects, organise them into groups, and filter right on the cast board.
• Upload your own looks — not everything has to be AI-generated. Photos you shot or found can go into the casting results, no credits.

Credits and plans

• The generate button says what this one costs, and so does the character-card button.
• Uploading a look costs nothing. Binding and unbinding a voice cost nothing either.
• Some models and sizes need a higher plan; options you cannot pick are marked "requires an upgrade".
• Acting profiles (including the voice description) and voice binding are both paid-plan features.

Common questions

The face drifted after a wardrobe change.
First check whether the "identity lock" is specific enough — that string is the main thing holding the look together. You can also attach reference images on the workbench so the look follows the image you specify.

I want a different main image but I don't want to lose this one.
Use "Edit identity" → Edit again, or upload a replacement. The old one stays in the version row and you can switch back any time.

What exactly does "Reuse" reuse?
That look's notes and reference images, loaded back into the workbench. You still pick the templates yourself.

Where can I use an actor?
After "Edit identity" → "Add to the Library", they can be picked as shoot reference images, in the art department, and so on. Any shoot segment that brings this character in uses them automatically.

"Regenerate" is greyed out on an actor I uploaded.
Because an uploaded actor has no prompt to re-run. Add a look description under "Edit identity" and it works — you only need to do this once.

The character speaks with a mainland accent — how do I change it?
"Edit identity" → Acting → Voice identity, and tap the "Taiwanese" chip. Left unspecified, the AI drifts that way.

I wrote a long acting profile but not all of it seems to be used.
The character budget is shared across everyone in the scene. The more people, the less each gets (about 180 characters each with four). Two or three actions that make the character recognisable is enough — you do not need to fill it.

I bound a voice but the finished shot does not have it.
Check the "Voice N/M" chip above the generate box on the shoot page — it lists who was carried and who fell back to text. Three usual causes: the segment has first/last frame turned on (that carries none at all), the slots were taken by other characters, or it was generated in the Studio (the Studio does not use voice bindings). Details in the Character Voice Handbook.

The voice is carried, but it does not sound close enough.
A bound file is a timbre reference, not a clone — it is never a 100% match. Unbinding it and writing the text voice description instead is often the better move, because while a file is bound the text description is not sent alongside it; unbind and the text takes over.

I changed the face — do I need to bind the voice again?
No. The binding is on the actor, not on the image, so swapping the main image does not break it.

Once they are cast, put them in a story with “The Short Film Shoot Handbook” or “The Short Drama Shoot Handbook”. For more on voices, see “The Character Voice Handbook”.