Skip to main content
The source video supplies the scene and its motion. A reference image supplies appearance. The prompt describes the change. Drag points supply positions that guide motion.

​
Describe the edit

Name the target, the visible change, and the parts of the scene to preserve. Start with one edit. Too vague
More specific
These examples are starting points. Adjust them to the actual source and reference image.

​
Replace a character or object

Upload a clear image of the desired subject with set_reference_image. Describe which source subject to replace.
When several subjects are visible, use position or appearance to identify one. A phrase such as “the person in the blue jacket” is clearer than “them.” A reference replacement applies from the next block while the stream continues. Expect one short pause as it applies. The edit moves to the new reference over the next few blocks. To change only the editing instruction, send the complete replacement prompt with set_prompt.

​
Replace clothing

Use the reference image for the desired outfit. Name the garment to change and the identity features to preserve.

​
Guide motion with drag points

A still image can serve as the source when your app publishes it as repeated video frames. For drag-only animation, set prompt to "". The empty string starts generation without a text instruction. For a guided action, name the subject and its movement:
Start a drag on the intended subject. Use a smooth path and hold the endpoint briefly. Send the positions from the last four source-frame ticks, oldest first, with set_pointers.
Drag points guide generated motion. They do not guarantee object selection or exact pixel placement. Send four empty lists to end a drag.

​
Change prompts during a run

A new prompt replaces the previous instruction from the next generated block. Restate the whole desired edit.
If a prompt describes continuous motion, releasing a drag does not remove that instruction. For precise application controls, update both the prompt and the drag window when the user releases the pointer.

​
See also