What it does
A character you build in UGC Studio is not locked inside UGC Studio. In Image Studio and Video Studio you can attach them to any shot, with any model that takes reference images, and the result is the same person every time.
Attaching a character brings three things with it:
- Their face. The identity pack images (Front, Three-quarter, Full body) load as reference images.
- Their identity. The character's locked description (age, build, face, hair, eyes) is added to your prompt, plus a "do not change the face" lock on models that accept one.
- Their default outfit. Used only when your prompt does not say what they are wearing. See How the outfit works.
In Video Studio the character's voice and Say it like pronunciations travel too, for spoken lines you write in "quotes".
There is no extra charge for attaching a character. You pay the normal price of the model you pick.
Attach a character
- Pick a model that takes reference images.
- Under Reference images, click UGC character and choose the character. Their pack images fill the reference slots (a model that takes one image gets the Front view).
- A card appears right under the references: the character's name, the Wardrobe for this shot field, and a way to remove them (× in Image Studio, Remove in Video Studio).
- Write your scene in Prompt as usual. Refer to the character by name ("Amy sits on the crates...").
The button only appears once you have at least one character with images in UGC Studio.
Clear All on the reference images also detaches the character in Image Studio, because the face and the description belong together.
How the outfit works
Give a character a Default wardrobe in UGC Studio (on the character's profile, "what they wear in every clip unless a Scene says otherwise"). In the studios that outfit fills Wardrobe for this shot.
- Your prompt does not mention clothing ("Amy reads her call sheet in the prep area"): the default outfit is used.
- Your prompt dresses them ("two technicians fit Amy into a white spacesuit"): the default outfit is left out completely. Nothing from it, not even a logo or brand name, reaches the model.
- You want something else for one shot: edit Wardrobe for this shot. The change is for that shot only. The saved character does not change, and detaching or switching characters puts the default back.
- You want no outfit at all: empty the field. The model then goes by what the character wears in their pack images.
The automatic outfit check
When you click Generate with a character attached, the studio first checks your scene: does it say what this character is wearing? It understands ordinary wording ("suited up", "in her wetsuit", "dressed for the gala") and ignores clothing worn by other people in the scene and accessories on their own (glasses, a hat, a watch).
After each generation the card tells you what it decided:
- Last generation: Default outfit used - your prompt did not dress them.
- Last generation: Default outfit left out - your prompt already says what they are wearing.
- Last generation: Outfit check unavailable - the check could not run, so the default outfit was sent the usual way. Your generation still went through.
Why the check exists. Image and video models do not follow "only if" instructions. Every word they receive ends up in the picture. In our own tests a character's default "ICONIC! sweatshirt" line, written as a fallback, still printed "ICONIC!" on a spacesuit and put the sweatshirt under the suit, even though the prompt asked for a spacesuit. The only reliable fix is to not send the outfit when the scene already covers it, and that is what the check does. Since it shipped, the same prompts come back clean.
Getting consistent results across a series
- Keep the same costume consistent by attaching a finished shot of it as an extra reference image, alongside the character's pack. Patches, colors and details carry over. Check the model's reference limit first (GPT Image 2.5 takes 5, so the 3 pack images plus 2 of your own).
- Grok Imagine with references always returns the shape of the reference image. With references attached the aspect ratio picker shows Auto for that reason; remove the references to choose a ratio. Other models, such as GPT Image 2.5, keep the ratio you pick.
- Name the character in the scene. It tells the model which person in a crowded shot is yours.
- Image-to-video from a finished frame keeps whatever the frame shows, outfit included. Use text-to-video with the character attached when you want the outfit rules above to apply.
Related
- The identity pack - the face images every shot follows.
- Reference images in Video Studio - what each video model does with references.