Start and end frames, and talking avatars

A start frame is a picture the video begins on. It’s the surest way to control exactly what’s in a shot: make or upload the picture you want, then let the model move it.

Add a frame

Frames are for video. On Create, use Frames under the prompt: Start frame and End frame are both optional.

• On Home, choose + beside the prompt.
• On a phone, tap the Start frame chip.

Pick a photo from your Library, or Upload a photo. A frame has to be a picture, not a video. Images you’ve generated are listed too, so a good way to start is to make the image you want first, then use it as the start frame.

How frames work

• The take starts on the start frame, and ends on the end frame if you add one.
• With a start frame, the take usually follows the picture’s shape, so use a picture in the shape you want.
• Not every model can end on a frame. Auto only picks one that can. If you pick a model yourself and it doesn’t take an end frame, the take starts on your start frame and ignores the end one.
• Your prompt still matters: say what happens between the frames.

Reference photos

Image models can take reference photos as well as a prompt. When you tag an element with @ in an image prompt, its photos go to the model as references. See Keep a character the same across shots.

Talking avatars

Pick Kling Avatar 2.0 or OmniHuman 1.5 as the model and the frames become Photo of the person and What they say:

1. Add a photo of the person, ideally looking at the camera.
2. Pick a voice track from your Library, or Upload audio. To make one, use the Voiceover tool, or Audio → Voice in an edit.
3. Use the prompt to say how they move, for example “small head movements, friendly, talking to the camera”.

The take runs as long as the voice track, up to 60 seconds, and is priced by the second. For a longer script, split it in an edit and make it in parts.