Making a Still Image Move: Keyframe Lessons from Rabbits, Cakes and Chairs

Image-to-video begins with a picture you already approve. The character, product placement and lighting direction can be established before movement enters the process. The next task is to decide what happens after that moment, rather than describing every visible detail again in text.

The frame establishes the start; the prompt establishes change

The official interface accepts first and last frame inputs. A small action can be enough: a rabbit lowers its head toward a cake, one ear twitches, and crumbs fall. That contains a subject, a change and a visible consequence. “Make this image lively” gives you much less to judge.

Look for the rabbit scene’s reaction chain

The official example description connects a rabbit, a cake and a pastry chef in a simple sequence. When evaluating a similar scene, check whether eating remains continuous, whether the cake stays in place and whether the chef’s arrival produces a readable reaction. Attractive fur is only one part of the shot.

Small adjustments can carry comedy

The chair example turns measuring and moving furniture into an understated event. It is a useful reminder to inspect contact: does a hand reach the chair, does the chair actually move, and do the people remain in the same room? If a group scene is too difficult to read, begin with one person and one chair.

Use a last frame when the endpoint matters

A final product position or an intended pose can be expressed through a last frame. The transition still needs to make sense. If the two frames differ radically in scale, background and light, the model must solve several changes at once. Prepare frames with a compatible canvas and a consistent main subject before asking for a complicated transition.

Leave physical room for motion

Check whether the subject is cut off, whether there is space in the intended movement direction and whether important objects are fully visible. Fixing these issues while preparing the still is often more direct than asking generation to repair the composition later. A fixed output ratio may also involve center cropping.

Start with an interpretable test

Try one frame, one main action and a slow camera move. Once the identity and space are stable enough for your purpose, consider a last frame or additional cuts. Compare the input with the beginning and end of the output instead of judging only uninterrupted playback. Keyframes communicate conditions; they do not guarantee identical details in every intermediate frame.

For a small experiment, use the same image with a stationary camera and then a gentle push-in. Keep the version with more convincing motion before adding other changes. Gradually increasing difficulty helps separate an input-image problem from an overcomplicated movement instruction.

Watch the related sample

The chair comedy

Open the studio and try your own scene