Image-to-video is the fastest way to get a believable shot: start from a still you already like, and let a model add motion. Done well, it looks cinematic. Done carelessly, it warps. Here's how to get the clean result every time.
Start with a strong frame
The video inherits everything from your source image — lighting, composition, subject. Generate or upload an image node on the canvas, get the framing right, then connect it to a video node. A clean, well-lit still gives the model far less to guess at.
Describe motion, not the scene
Your prompt should describe how things move, not re-describe what's in the frame. "Slow push-in, gentle parallax, drifting particles" gives the model a job; repeating the subject just invites it to redraw — and redrawing is where warping comes from.
Use two frames for control
For a predictable result, give the model a start frame and an end frame. It interpolates between the two, so the motion is a controlled morph rather than a hallucination. This is ideal for blooms, reveals, and simple camera moves.
- Keep motion slow and contained for the cleanest output
- Lower the motion strength if the subject distorts
- Keep clips short — 3 to 5 seconds loops best
One still, one prompt, one connected node. That's a shot.
