How do I turn a still image into a video with AI?

Give an agent connected to the FLORA MCP the image and a short description of what should move. The agent wires the image into an image-to-video model such as Seedance 2.5 on your FLORA canvas, runs it and returns an MP4. Our 5-second 720p clip of a potter at the wheel cost $2.457 and took 3 minutes 27 seconds.

At a glance

What you need

• An agent with the FLORA MCP installed. Install steps are in the FLORA MCP docs.
• A FLORA workspace with video generation.
• One still image: a photo, a render or a frame you generated in FLORA. Match its aspect ratio to the video you want.

Best for

• Social clips and loops from a photo you already approved
• Adding motion to a product or campaign image without a shoot
• Turning storyboard frames into shots for an edit
• Testing a camera move before committing to a longer clip

Step by step

1. Step 1: Pick or make the still

   The clip inherits everything in the still: the subject, the light, the crop. Generate it at the aspect ratio you want the video in, usually 16:9 or 9:16, so the model does not have to invent extra picture at the edges.

2. Step 2: Describe the motion and the camera

   Say what moves, how fast, and what the camera does. One action and one camera move hold together best in a 5-second clip. Ask for a quote before it runs.

   Animate this image with Seedance 2.5 through FLORA, 5 seconds, 720p, 16:9. The potter's wheel spins steadily and the wet clay cylinder rises slowly between her hands. Camera holds still with a very slow push in. Light stays constant. Quote the cost first.

3. Step 3: Let it render

   The agent adds a video node wired to your image on the FLORA canvas and starts it. Video takes minutes, so let the agent poll in the background. It returns the MP4 link when the run completes.

4. Step 4: Check the clip and change one thing at a time

   Watch for hands, faces and edges that warp, and for camera moves you did not ask for. If the motion is wrong, change only the motion words. If the framing is wrong, fix the still and run again.

What happened when we turned a still into a video in FLORA?

On October 16, 2026, we started from a still we generated with GPT Image 2.5 Flare for $0.055: a ceramicist's hands shaping a tall cylinder of wet grey clay on a wheel, in a small studio with soft overcast window light and shelves of unglazed pots behind her. It was 1280 x 720, already 16:9.

The agent called flora_add_to_canvas to add a video node with model i2v-gengateway-seedance-2-5-i2v, duration 5, aspect_ratio 16:9 and resolution 720p, wired by an edge to the still. It started the node with flora_run_canvas_nodes and polled flora_list_generations. The run completed in 3 minutes 27 seconds and was charged $2.457.

The clip is 5.06 seconds, 1280 x 720 at 24 fps, with an audio track. The wheel spins, the clay cylinder grows taller and smoother as her hands draw it up, and slip streaks along the clay. Her hands, the apron and the shelves hold their shape. The light matches the still from start to end.

What went wrong: we asked for "a very slow push in" and got a strong one. By the last second the frame has moved in so far that her face is out of the top of frame and the clay fills the middle. The fix is to ask for a locked-off camera, or to say "push in no more than 5 percent" when the subject must stay whole.

Model

Seedance 2.5 (image-to-video)

Tools called

flora_add_to_canvas, flora_run_canvas_nodes, flora_list_generations

Cost

$2.457 for the clip, $0.055 for the still

Time

3 min 27 s for the clip

Short clip or long clip: pros and cons

Which video model should I use?

We used Seedance 2.5, which handles realistic motion from a photo well. Kling and other image-to-video models sit behind the same FLORA MCP, so you can run the same still through two models and compare. Ask the agent to quote each first, since prices differ by model, length and resolution.

How do I stop the image from warping?

Keep the motion small and physical, and name what must not change. Hands, faces, logos and text are where drift shows first.

• Ask for one action, not three.
• Say "the product stays identical" or "her face does not change".
• Use a locked-off camera when the subject must stay whole.
• Generate the still at the video's aspect ratio so nothing is invented at the edges.

Can I animate several stills at once?

Yes. Ask the agent to add one video node per still and start them together. They render in parallel, so five clips take about as long as one. Quote the batch first.

Frequently asked questions