ByteDance: Seedance 1.0 Pro Fast

Seedance 1.0 Pro Fast is the Fast profile identified by doubao-seedance-1-0-pro-fast-251015. It generates silent video from a written scene or an image-led starting composition. The exact selector distinguishes this profile from Seedance 1.0 Pro; the word Fast does not define a guaranteed completion time for an individual task. Choose the input mode according to what should establish the scene, then evaluate the returned clip against that brief.

A text-led shot begins with a concise sequence: the subject and setting, the action, the camera route and the final view. For a product reveal, describe the object on its surface, the light across its material and a short movement that ends on the feature being shown. For a location concept, move from a wide establishing view to a recognizable focal point. Avoid asking one short clip to perform several unrelated scenes whose subjects and settings must all be introduced from scratch.

An image-led shot begins with an image_url item in the content array and a text instruction for movement. Prepare the crop, background and subject arrangement in that image, then state which elements move and how the camera behaves. This lets a still composition become a short animated scene. It is useful for a portrait, a tabletop object or a prepared visual concept, while the actual result still needs inspection for identity, shape and motion throughout the clip.

Image inputs may use an accessible URL or Base64 data URL. The documented formats are JPEG, PNG, WEBP, BMP, TIFF and GIF; the image must be below 30 MB, with both dimensions strictly between 300 and 6000 pixels and width divided by height strictly between 0.4 and 2.5. Choose a clean source with visible subject detail. The image guide recommends adaptive framing, so the prepared source proportions should match the composition you want to animate.

The 1.0 profile documents two-to-twelve-second duration at 24 fps. Pro Fast has 1080p as its default resolution, and the retained output table lists 480p, 720p and 1080p tiers. The operation guide excludes 1080p for the reference-image scenario. Fixed aspect ratios include 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9. Pick a resolution and shape appropriate to the selected input mode, rather than assuming every combination is interchangeable.

Seed and watermark options help keep request settings explicit. The integer seed range is -1 through 2^32-1; -1 requests randomness. A fixed-camera instruction is conditional: the guide excludes it in the reference-image scenario and does not guarantee the visual effect where it is supported. Parameter commands embedded in the text override separate optional parameters. Keep those instructions consistent, especially when revising a prompt whose old duration or resolution command might otherwise remain in the scene description.

Usage is measured in video tokens: output width × output height × frame rate × output duration / 1024. The completed task reports this quantity as usage.completion_tokens. Retain the reported usage with the finished clip and its requested settings.

Submission creates an asynchronous task. Save its identifier, follow its status and retrieve the completed video from content.video_url after success. Generated files are documented as retained for seven days, so keep the completed artifact with its prompt and any input image. Review the complete action, camera movement and final framing; if one detail misses the brief, describe that specific correction while retaining the parts that worked.