Pictures that make better clips
References tell the AI what things look like. How to make and upload pictures for your clips, how to number them, and how to combine a character, a place and a prop in one scene.
Words are good at telling the AI what happens. Pictures are better at telling it what things look like. That's what references are for: you attach pictures to a clip, and the clip uses them for faces, outfits, places and props.
Where references live
Next to the Generate button there's a card called Reference. It opens your pictures, in two tabs:
- Images — pictures you upload or generate, used as references as they are.
- Character — character sheets, made from photos or a description so a character stays the same in every clip.
Everything you make or upload stays in your library (your profile → Images), so you can reuse it across shows.
Getting pictures
Upload one from your computer — PNG, JPEG or WebP, up to 10 MB.
Generate one in the Images tab: describe it in the box at the bottom and press Generate. A picture costs 0.10 credits and comes out tall (9:16), the same shape as the clips.
Change one of yours: add it as Image 1 in the box and describe what to change — "make it night", "put a party hat on him", "same room, but flooded".
Numbering: Image 1 to Image 9
You can attach up to nine pictures to a clip. They're numbered in the order you add them — Image 1, Image 2 and so on — and you use those names in the scene:
Image 1 walks into the kitchen in Image 2 and hides the cake from Image 3 behind his back.
Naming them in the scene matters. A picture that's attached but never mentioned is a picture the AI has to guess about.
A good combination
For most scenes you'll want some mix of:
- The character — a character sheet.
- The place — a picture of the room, the street, the planet.
- The prop — the cake, the trophy, the cursed sandwich.
That's usually enough. More pictures don't always mean a better clip; three clear ones beat nine busy ones.
Tips for pictures that work
- One subject per picture. A clean shot of the thing, not a collage.
- Similar light. A night-time place and a daylight character fight each other.
- Plain backgrounds for characters and props, so the AI knows what to take.
- The same pictures across a show. Reusing references is what makes episodes feel like one world.
Next: put your references to work with a well-written scene — how to write a scene the AI actually follows.
