One paragraph rules the face
Consistent characters start as text: age, look, three signature details, wardrobe, five words of personality. Generate from exactly that paragraph every time and the identity holds. Change it and the face drifts.
References anchor identity
Upload your best previous render as the reference. The model treats it as ground truth for the face while you vary scene, outfit and framing. Judge identity on close-ups, not full-body shots.
A character is an asset
Download the sheet, then feed it forward: portrait sets, image-to-video motion tests, face-swap onto scenes. The next-step buttons carry your result into each tool with one click.
FAQ
How do I keep the same face across images?
Reuse the identical description word-for-word, attach your best result as the reference image, and change only one thing per generation.
Can I use the character in video?
Yes - send any render to image-to-video. Models like Seedance animate the still while preserving the look.
Anime or realistic?
Both - pick a style preset above the prompt. Or write the style into your canonical paragraph so it travels with the character.
Is this a character generator?
Yes. One canonical description plus a reference image gives a character you can reuse across scenes.