Seedance 2.0 Fast Reference-to-Video

Seedance 2.0 video generation from reference images. Upload up to three images, write a prompt that describes what moves and how, and the model builds a video from them.

Your references can be characters, environments, and objects. Describe each one in the prompt by label and the model composes them into a single scene. Camera movement, physics, lighting, and motion are all prompt-driven. Output is a 16:9 MP4 at 480p, up to 4 seconds.

How do you use reference images with Seedance 2.0 to generate video?

Label each reference image in your prompt (IMAGE 1, IMAGE 2, IMAGE 3) and describe what each one does in the scene. Seedance 2.0 composites them together and follows your motion and camera instructions. Outputs H265 MP4 at 480p, 16:9.

Reference images (up to 3) Each image gets its own role in the scene. Label them in the prompt: "IMAGE 1 is the character, IMAGE 2 is the creature, IMAGE 3 is the environment." Want the model to understand which element to move and which to treat as background? Be explicit. Unlabeled references get interpreted however the model decides.

Prompt This is where the work happens. Describe the scene, the motion for each element, the camera, and the lighting separately. Want a slow push-in that ends on a close-up? Write it. Want a character to raise their arm while rain falls around them? Describe both independently. The model reads the whole prompt before generating, so order your instructions from most to least important. Break it into sections if the scene is complex.

Camera instructions Put your camera direction in the prompt under its own label. "Slow push-in, transitioning to a smooth orbit" will produce different results than leaving it out entirely. Seedance 2.0 responds well to specific camera instructions. Vague prompts produce vague motion.

Duration Default is 4 seconds. Long enough for a camera move or a subject action, short enough to keep generations fast. Test at 4 seconds first. Extend once the motion and composition look right.

Resolution Set to 480p. Good for previewing and iteration. Check the composition, motion, and timing at this resolution before committing to longer or higher-quality renders.

Seed Randomize by default. Fix the seed to hold the base generation stable while you test prompt edits. Change the seed when a generation is close but not quite right and you want a fresh variation.

What is Seedance 2.0 Reference to Video good for?

Short cinematic clips where you are compositing a character, an object, and an environment from separate source images. Fantasy scenes, product reveals, concept art brought to motion. Any situation where you have the visual assets but need them in motion.

The strongest use case is multi-element scene composition. If you have a character reference, an environment reference, and a prop or creature reference, Seedance 2.0 can combine all three into one scene following your prompt direction. That is hard to pull off reliably in a single-image-to-video workflow.

It also works well for previsualization. Write a detailed camera and motion brief in the prompt, reference your concept art, and get a moving version to review before production commits to anything.

Where it is less suited: if you only have one image and a simple action, a standard image-to-video workflow will be faster and easier to prompt. The reference system is most valuable when you need the model to track multiple distinct visual sources at once.

FAQ

How do I reference multiple images in a Seedance 2.0 prompt? Label each image in the prompt before describing it: IMAGE 1, IMAGE 2, IMAGE 3. Then describe what each one does in the scene. "The character from IMAGE 1 walks forward. The environment from IMAGE 3 is the backdrop." The more clearly you separate the roles, the more the model honors them.

What resolution and format does Seedance 2.0 output? This workflow outputs H265 MP4 at 480p, 16:9. Good for quick iteration and composition checks. Use it to lock down the motion and framing before you invest in longer or higher-resolution generations.

Can Seedance 2.0 follow specific camera movement instructions? Yes. Camera direction goes in the prompt under its own section. Slow push-ins, orbits, parallax, handheld shake. Describe the movement explicitly and Seedance 2.0 will attempt to follow it. Generic prompts produce generic motion.

How long can a Seedance 2.0 video clip be? Up to 4 seconds with this workflow. That is enough time for a camera move, a subject action, or a scene reveal. Test at 4 seconds before going longer.

How do I run Seedance 2.0 Reference to Video online? You can run it on Floyo. No installation, no setup. Open the workflow in your browser, upload your reference images, write your prompt, and hit run.