Seedance 2.5 image-to-video: turn a still photo into a moving shot

Text-to-video invents everything: the room, the light, the product, the face. That is a feature for mood pieces and a bug the moment the shot needs to show your actual product or a character who must look the same tomorrow. Image-to-video removes the biggest variable. The composition, the styling and the identity come from your image; the model's job shrinks to motion and camera. Fewer surprises, fewer rerolls, and results that match the rest of your brand's footage.

Not every photo is a good first frame. The ones that work share a few traits:

• Sharp and well lit. The model amplifies what is in the frame. A soft, noisy photo gives you a soft, noisy clip.
• One clear subject. Busy scenes force the model to guess what the motion is about. Give the frame a hero.
• Some breathing room. A little space around the subject gives motion and camera moves somewhere to go. A subject cropped tight to the edges animates like a stuck postcard.
• Honest resolution. Around HD or better on the short side is a good baseline; tiny images get upscaled into mush.
• No baked-in text you want to keep. Overlaid text, corner logos and watermarks tend to wobble as the frame moves. Keep images clean; add text in the edit.

In image-to-video the prompt's job changes: you are no longer describing the scene, you are directing what happens to it. The structure that works:

• What moves. One subject motion: "she turns toward the window and smiles", "steam rises from the cup", "the liquid pours in".
• How the camera moves. One move: a slow push-in, a slow orbit, or none at all. Saying "the camera holds steady" is a legitimate and underrated instruction.
• What stays fixed. If something must not change, say so: "her face and the label stay unchanged", "the background stays still".

Example, start to finish: "The woman turns her head toward the window and smiles softly, hair moving slightly. The camera holds steady. Her face, the cup and the background stay unchanged. Warm morning light." A complete image-to-video brief in under 40 words.

Products are the best-case scenario for this mode. Shoot or generate one clean hero image, then ask for motion around it:

• The orbit. "The camera slowly orbits around the bottle, light gliding across the label, background stays unchanged." The safest premium-looking move in the toolkit.
• The push-in. "Slow push-in on the product as condensation rolls down the glass." Works especially well for food and drink.
• The reveal. "The camera pulls back to show the full box on the table as the light warms up."

One warning: do not ask the model to animate text or invent label details it cannot see. Keep the label readable with slow moves, and if the label must be pixel-perfect, composite the real product photo over the clip in the edit.

For recurring characters, image-to-video is how you stop rerolling. Generate one reference image you like - or shoot one - then use it as the first frame for every shot in the scene. The same face walks in, pours coffee, looks out the window: three generations, three different actions, one identical person. That is also how our studio handles it: the storyboard keeps a character reference attached to every shot slot, so consistency is a setting rather than a hope.

• The face changes from my photo. Usually too much requested motion, a low-quality source, or a prompt that re-describes the person differently than they appear. Gentler motion, better source, and describe only the motion, not the identity.
• The image morphs or melts. You asked for a big transformation. Image-to-video is best at small, believable motion; if you need a dramatic change, do it across two shots with a cut between them.
• Text and logos warp. Any camera move plus burned-in text equals wobble. Slow the move down, or clean the plate and add text in post.
• The clip is technically fine but dead. The photo was static and so is the prompt. Add one small subject motion - steam, hair, background life - and the frame wakes up.
• Hands go strange. Same as ever. Reframe away from precise hand moments or fix them in the edit.

The clip inherits the shape of the source photo, so crop first and generate second. Making vertical content? Crop the source to 9:16 before uploading - a vertical crop of a finished 16:9 clip will disappoint you, because the motion already happened in the parts that got cut. Same logic for squares. Decide the destination, shape the photo, then generate. And if the result becomes one shot of a bigger video, our studio's storyboard and browser timeline take it from there: order the shots, caption them and export the cut without leaving the browser.