Photographing real people (actors)
Current status: real people are not supported yet. Photos of real people are canceled at upload, and video generation is blocked for any character marked as a real person. Video models block images containing real faces as a matter of policy, and we are still preparing actor identity verification and rights management. This document is the shooting standard for when that is ready — shooting now is worthwhile, but you can't upload and use the photos yet.
To generate images and video with a contracted actor's face, you first need a set of reference photos. This document explains what to shoot and how.
There is an illustrated version of this guide in the app — diagrams of shooting angles, framing, and lighting setup, plus the full shot list. It will open from the Characters tab when real-person support launches. This document is a summary of it.
First: confirm rights
Don't use a face without the person's consent and usage rights. And even with consent, the service does not accept real people right now — we deliberately removed any path that passes on self-declaration alone. The official path is for the actor to go through verification themselves, and it will open once it is ready.
Even with rights secured, the generation models refuse. Generation services block images of real people under their portrait-rights policies, and switching models gives the same result. This is separate from your contract.
One or two photos won't do
There's a reason. A generation model doesn't take only the face from a reference image — it drags along the background, pose, gaze, and lighting too. With only one photo, that photo's background and angle follow every shot. The model needs several angles and expressions to tell "this person's face" apart from "the lighting that day."
Conditions to keep fixed throughout
| Item | Condition | If you don't |
|---|---|---|
| Background | Plain gray or white wall, at least 1.5m between the actor and the wall | Background patterns leak into the generated results |
| Lighting | One soft light source at 45° in front + a reflector on the opposite side | Hard shadows get learned as facial features |
| Lens and distance | 50–85mm (35mm equivalent), 2–3m | Wide-angle close-up distortion becomes part of the identity |
| Resolution | At least 2000px wide, JPEG/PNG | Facial detail can't be recovered |
| Hair and wardrobe | One consistent set matching the character's default look | The model can't tell which one is the reference |
Not allowed: covering the face with sunglasses, masks, or hats; strong filters or beauty retouching; watermarks; photos with other people in them.
What to shoot, and how many
The shot list maps 1:1 to the upload slots on the character sheet. Shoot in order, and if you run short on time, drop the ones at the bottom.
- Master sheet — must shoot (12 photos)
Full body front, 3/4, side, back, and 3/4 back; face close-ups front and 3/4; back-of-head hair; hands; shoes; color reference; relaxed pose. This becomes the face reference for every shot.
- Expressions and mouth shapes — must shoot (12 photos)
Neutral, smile, laugh, anger, sadness, surprise, embarrassment, thinking, determination + mouth shapes "ah," "oh," and mid-speech. Shoot all 12 with the same framing (chest up), same distance, and same lighting. The only thing that changes is the expression. With a photorealistic face, a slightly off expression instantly reads as a different person, so this is the hardest part to replace with generation.
- Camera and composition — if possible (12 photos) Records how the person reads at each shot size, keeping
identity consistent in close-ups and wide shots.
- Hands and props — if possible (12 photos) Hands are the part generation models ruin most often.
- Motion — if possible (12 photos) Body habits when walking, sitting, and gesturing. Still photos help too.
- Wardrobe variations — don't shoot Shoot only one default outfit (front, back, seated), and generate the rest
from Looks on the character card. Clothing has endless combinations.
- Scene compatibility — only 3 lighting extremes Window light, backlight, and a dark interior. Shooting 12 locations is
unrealistic; generation fills in the rest.
Steps once real-person support opens
You can't carry out the steps below right now. Use this order only after real-person support and identity verification are released.
- Upload the master photo to the main image slot on the character card and approve it.
- Add expression and hand photos to the identity anchor as attached reference images — the derived sheets (expressions, motion,
props) are filled by generating from the approved master (derived sheets have no direct upload slot).
- Generate wardrobe and locations from Looks on the character card — the approved master serves as the face reference.
- (Coming) When actor identity verification opens, a rights-confirmation step will be added — right now
real-person uploads are canceled, so you can't proceed to this step.
Uploaded images can't be remixed (they have no seed or prompt). Rather than regenerating slight variations of the same photo, it's better to capture every angle you need during the shoot.
Next: Planning assets — setting up characters, locations, and props covers how sheets and looks carry through to images and video.