Character: the same person in every shot

Save a person, animal or cartoon character once, and they come out with the same face in every picture and clip.

A character is a person, an animal or a cartoon figure you want to keep the same from shot to shot. You save a few pictures of them under a name, like @quell. Write that name in a prompt and the AI copies their face, hair and build from the pictures — not just someone who fits a description.

A detective in a fedora and grey suit leaning in a hallway doorway — a different man from Quell, with a longer face and different features. — The same scene, asked for twice. Left: the detective written out in words — stubble, fedora, grey suit, red tie. The AI made a different man. Right: the prompt says @quell, so it copied his face from his pictures.

Detective Quell himself, the man from his saved pictures, leaning in the same hallway doorway. — The same scene, asked for twice. Left: the detective written out in words — stubble, fedora, grey suit, red tie. The AI made a different man. Right: the prompt says @quell, so it copied his face from his pictures.

Pictures: 1 to 20. Rosa, Quell and Pembrook have 3 each
Description: Required — models that only read words get it instead of the pictures
Best with: A model that looks at pictures (see below)
Cost: Free to make. Using one costs only the model’s normal price

Which pictures to give it

Here are Rosa’s three pictures from the Gullrock Coffee story. They follow a few simple rules:

Rosa facing the camera with a warm smile, in her navy henley and rust apron, in front of a plain grey wall.

Rosa turned three-quarters to the left, glancing back at the camera, plain grey wall.

1Only the one person

No one else in the picture, so the AI knows whose face to copy.

2A plain background

A simple wall keeps the focus on the person. The character leaves out the background anyway — your prompt decides where they are.

3A few angles

Front, three-quarter and side, so the face holds up when they turn. Make the clearest front picture the cover.

4Same face, same clothes

Every picture shows the same person, dressed the same way. Mixed pictures give the AI a mixed idea of who they are.

No pictures of your character yet? Make them

This is how Rosa’s three pictures were made, from nothing:

1
Make one picture of them with any picture model. Describe the person and put them in front of a plain wall, for example “Photo, head-and-shoulders portrait of a coffee roaster in her early 50s facing the camera with a warm half-smile…”.
2
Open that picture in a model that looks at pictures (Rosa’s used GPT Image 2.5 Sunburst Edit) and ask for a new angle: “The same woman, same face, hair, earrings and clothes exactly, now turned three-quarters to the left and glancing back toward the camera. Plain warm-grey studio wall behind her.”
3
Do it once more for the side view: “…shown in side profile facing right.”
4
Save the three pictures as one character (next section).

TipWant your character in a look — a cartoon, a painting, felt? Redraw these pictures in that style before saving them. See Style: one look for everything.

Make the character

1
In the Library, press Make entity or style from pictures, tick the pictures and press Make entity. (Or press the + beside Entities in the left menu and add the pictures in the window.)
2
Keep the kind on Character and type a name. It becomes the @name you type in prompts.
3
Press ✨ Describe to have the description written from the pictures, then read it. A character must have one.
4
Press Create @….

The New character window: the name rosa with its handle @rosa, Rosa’s three pictures with the first as cover, her description, the Gullrock Coffee collection and Create @rosa.

Use the character

Type @ in any prompt box and pick them. Then write what they do, what they wear if it should change, and where they are — for example “@quell holds up an old pocket watch by its chain and studies it closely”.

TipDon’t describe the character’s face in the prompt — it fights their pictures. The pictures already carry the face.

For the face to really match, make the picture with a model that looks at pictures. The tag under the prompt shows which you have: “@quell sends 3 pictures” means the AI sees him; “sends its description” means it only reads about him (see Why the model you pick matters).

Cartoon characters: words get closer

A simple cartoon design is easier to describe than a real face. With Pembrook the fox, the written description alone came very close — but compare the eyes and the shape of the face. Only the character gets him exactly.

A felt cartoon fox in teal glasses and a mustard hoodie waving in a playroom — close to Pembrook, but with bigger, rounder eyes and a different face shape. — Left: Pembrook written out in words. Close, but not him. Right: @pembrook.

Pembrook himself, from his saved felt pictures, waving in the same playroom. — Left: Pembrook written out in words. Close, but not him. Right: @pembrook.

Characters in video

The way all three example films were made: make each shot as a picture with the character, check it, then animate that picture in the Video lane. The clip keeps the face from the picture.

Or use Kling O3 · Entities, which builds a clip straight from your entities. A character with two or more pictures goes in as a Kling “element” — its front picture plus its other angles — which helps the face hold up as they move. The tag under the prompt says so:

A prompt on Kling O3 · Entities with the tag “@rosa goes in as a Kling element (3 pictures)”. — On Kling O3 · Entities, Rosa’s three pictures go in together as one Kling element.

See Entities in video for both ways, with the Gullrock film.

If it doesn’t look like them

?A different face

Check the tag under the prompt. If it says “sends its description”, the model only reads words — switch to a model that looks at pictures.

✂Picture limit reached

With many entities in one prompt, the last ones may only send their description. The tag says so. Use fewer entities in that shot.

🎨The look didn’t take

A character’s own pictures outweigh “in the style of…”. Redraw the pictures in the style and save those instead.

✕The face is off in every shot

Improve the pictures: one person, plain background, clear front view as the cover. See Change, delete or check an entity to swap them.

More in Entities

Entities: start here

Save a person, product, place or look once, then use it by name — so it comes out the same in every picture and clip.

Prop: the same product in every shot

Save a product or object once — its shape, colours and label — and it comes out the same wherever you put it.

Style: one look for everything

Give every picture and clip the same look — a painting, ink, felt and clay — with a ready-made style or one made from your own pictures.

Palette: keep to your brand colours

Save 3 to 6 brand colours under one name, and every picture and clip you add it to uses them as its main colours.

← Back to the Help Center