Likeness-locked keyframes anchor your video model, so the character stays the same person from the first frame to the last.
Generate an original character, keep only the frames that stay on-model (often about one in five) and train Phota on those. After that, it generates the character reliably.
@lena Or train from real photos of talent, exactly like any other Phota model.
@tiana Give your image-to-video model a start and end frame of the same person, and let it fill in the motion between them.
@tiana 
Phota generates the likeness-locked start and end frames, and your image-to-video model fills in the motion between them.
Anchor each shot with keyframes from the same likeness, so the clips cut together into one piece.
@tiana 
A Phota keyframe anchors every scene with the same character, so the clips cut together into one continuous piece.
Real talent or an original character. Train a likeness model from a handful of images.
First and last frames, or an anchor for every shot, composed and lit to match the character.
Drop the keyframes into any image-to-video model, and it fills in the motion between them.
Re-anchor, extend, and cut shots together. The character stays the same across takes.
Video models drift over frames, so the person at the end of a shot isn't the person at the start.
Likeness-locked keyframes pin the shot at both ends, so identity doesn't drift.
Per-model character setups don't transfer between video tools.
Keyframes are just images, so they work with whatever image-to-video model you run.
Each clip re-invents the character, so sequences don't cut together.
Every keyframe comes from the same persistent likeness to match the shots.
Video models have no persistent identity. They generate motion frame by frame, and the face drifts along the way. Anchoring the shot with likeness-locked keyframes holds generation to the same person at the points that matter.
Any image-to-video model that accepts a start frame, an end frame, or reference frames. Keyframes are standard images, so they work with whatever video pipeline you already use.
Anchors reduce drift a lot because generation is held at both ends of the shot. For deeper control, native video personalization is on our roadmap.
Yes. Synthetic characters train into likeness models just like real talent, so an original character anchors video just as consistently.
Yes. Train a model for each person, then put them together in the same keyframe. Everyone keeps their own likeness, so a shot with two or three people holds the same way a solo shot does.
