Can I use a photo of a real person?

Yes. Upload photos and label them as a character, and Scenes AI uses yours instead of generating a sheet. The same works for locations, for a specific scene, or as general style guidance.

How many characters can one video have?

There is no fixed cap. Each character gets its own sheet, and each sheet costs one image generation, so the practical limit is what the story needs and what you want to spend.

What if the character sheet comes out wrong?

A vision reviewer checks each sheet and can trigger one regeneration. Beyond that you see the whole cast at an approval gate before any scene is generated, and you can send it back with changes.

Does consistency survive into the video clips?

Yes. Every clip is image-to-video — the approved still comes first and the video engine animates it — and the engines also accept the reference images directly. The likeness is carried in both ways.

Character consistency

AI characters that stay the same in every scene

Scenes AI generates a four-view character sheet — front, side, back and a face close-up on a single canvas — before it makes any scene, then passes that sheet back as a reference image into every shot the character appears in. That is why the same person shows up in scene one and scene twenty, and in next week's episode, instead of being reinvented each time.

What this gets you

The formats that need it most are the ones people most want to make.

A mini drama whose cast is the same in episode one and episode five.
A drama where two characters hold their faces and their voices through a whole conversation.
A channel with a recurring host that viewers start to recognise.
An ad campaign fronted by the same spokesperson across every video in it.
A series where episode five still looks like episode one.

Why generated characters usually drift

An image model has no memory between calls. Ask it for 'a woman in a red jacket' twenty times and you get twenty different women, because nothing connects one request to the next. Prompt wording alone cannot fix this — 'brown hair, green eyes' is satisfied by thousands of different faces. The only reliable handle is a picture, given back to the model as a reference.

What a character sheet is

Before any scene is generated, each character gets one image that is only ever used as a reference.

Four views on one canvas: full-body front, side and back, plus a face close-up.
No text and no labels on the sheet — those confuse the model when it is used as a reference.
A vision reviewer checks the sheet before it is accepted, confirming all four views are recognisably the same person.
If the sheet fails review it is regenerated once, rather than being quietly accepted.
You approve the finished cast at a gate before a single scene is generated from them.

How the sheet is used

The sheet is not decoration — it is an input to every scene it belongs in.

Each scene's image is generated with the relevant character sheets attached as references.
Scene prompts restate one or two distinctive identifying details, so the reference has something to anchor to.
Video engines accept references too, so the likeness carries into the motion and not just the still — though how many, and whether they can be combined with first and last frames, varies by engine.
Reference-image capacity is therefore the engine spec that matters most for consistency.

Consistency you can hear

Characters carry a voice description as well as a face — one canonical phrase covering gender, age, timbre, pitch, pace and accent. That exact phrase is repeated in every scene where the character speaks, so a character does not change voice halfway through a video. It is verified as a literal text match rather than left to chance.

Using a real person or place

You can upload your own photos instead of generating a sheet. Label them as a character, a location, a particular scene, or general style guidance, and Scenes AI uses yours as the reference — so a real product, a real room or a real person can appear throughout the video.

Not just people — props and brand marks too

Every reusable asset in a project is a named reference, and characters are only one kind.

Character — a person, held to one face, one wardrobe and one voice wherever they appear.
Environment — a place with its fixed anchors and lighting, so a room stays the same room.
Prop — an object: your product, or the letter the plot turns on, identical in every frame.
Logo — a brand mark, treated as content rather than annotation when it has to appear.

Why a prop reference matters commercially

In an ad or a UGC video the product is the subject, and a product that changes shape between shot two and shot six fails the video regardless of how good everything else looks. Registering it as a named prop — ideally from your own photograph — means every scene that shows it draws on the same reference. To be clear about what this is not: a logo reference is a single named asset the model is given for a project. There is no saved palette, no typography system and no reusable brand profile carried between projects.

Locations get the same treatment

A room that rearranges itself between shots breaks continuity just as badly as a changing face.

Each location gets a two-panel plate: a wide establishing shot, and the same space with the camera turned 180 degrees to face the other way.
Three to five fixed anchors are named per location, grouped by which end of the space they are in rather than by left and right — which reverses when the camera turns.
Light sources are pinned so the time of day does not drift between scenes.
The reviewer specifically looks for the mirrored-room failure, where the second panel is a flip of the first rather than a genuine reverse angle.

FAQ

Character consistency questions

Can't find what you're looking for? Visit help and support or email hello@usescenes.com and a human will reply.

No. You describe them once during planning. The sheet is generated from that description, and from then on the sheet is what the model sees. In later episodes the cast carries over automatically.