Images
How do I keep a character consistent across images?
Image models start fresh on every generation, so the same prompt can produce a different face each time. Consistency comes from references, a fixed description and editing.
Start with a reference image
Make or choose one clear image of the character — front-facing, neutral background, good light. This becomes your reference. Some people make a character sheet first: the same character from the front, side and back in one image, which gives models more to work from.
Write one description and reuse it
Write a short, fixed description: age, hair, eye colour, build, signature clothing, any distinctive marks. Paste the exact same words into every prompt and change only the scene and pose. Changing the wording of the description changes the character.
Edit instead of regenerating
To put the character in a new scene, upload the reference to an editing model and describe the change: same person, now sitting in a café, wearing a raincoat. Editing keeps more identity than generating from text each time.
Carry it into video
For video, start from an approved still of the character rather than from text. Image-to-video keeps the face and clothing much closer to your design. Keep shots short and generate each scene from its own approved still.
Real people
Do not create images of real people without their consent, or images meant to mislead others about who someone is. Provider filters often block recognisable public figures, and the Terms forbid impersonation.
FAQ
- Which model is best for consistency?
- Models that edit from references, such as Flux Kontext and Nano Banana, keep identity best. Try your character on two of them.
- Why does the same prompt give a different face?
- Each generation starts fresh. A reference image and a fixed description are what keep the face stable.
- Can I use my own photo?
- Yes, if it is you or you have the person's permission.