Image to Video

Turning a storyboard frame or a product still into a shot

Game studios and short-film makers often start from a key frame they have already painted. Give image-to-video one frame, not a sheet of panels: the image you upload becomes the literal first frame, so a four-panel sheet starts the clip as a collage and the model animates the collage. A multi-panel storyboard belongs in reference mode, where it is read as a plot reference instead of a picture.

For product clips, say what must not change (shape, colour, logo, printed text) and describe only the motion around it: a slow turn, light sweeping across the surface, a hand setting it down. The model treats everything in the frame as something it may animate, so anything you do not lock can drift.

• Use a multi-panel storyboard in reference to video →
• Product clip prompt template →

One image or two, and what changes

Image-to-video accepts up to 2 images, and the count changes the job. One image is read as the first frame: the model animates forward from it and decides where the shot ends. Two images are read as first and last frame: the model has to arrive at your second image, which makes the motion far more predictable but also constrains what it can do in between.

Two-image jobs fail most often when the pair is too far apart — a close-up and a wide shot of different scenes give the model no plausible path between them. Keep the subject and the framing related, and let the prompt describe the transition you want.

Photos of real people are refused

This is the single most common failure in image-to-video, and it surprises people because the image is often their own photo. Every Seedance model refuses to animate a recognizable real face. Holding the rights to the image does not change the outcome, and neither does switching models — the check sits upstream of all four.

The compliant path is to create a character from that image under Assets first, then reference the character. The character is a generated likeness rather than the photograph itself, which is what makes it acceptable — and it has a second benefit: the same character can be reused across clips, so a series stays consistent.

What makes a good source image

The model animates what is already in the frame; it does not invent what the crop cut off. An image where the subject is fully visible with room around it gives the motion somewhere to go. A tight crop leaves the model nothing to work with and usually produces drift at the edges.

Motion blur, heavy compression artifacts and text overlays all degrade the result, because the model treats them as part of the scene and tries to animate them too.