ByteDance: Seedance 2.0 Fast

Seedance 2.0 Fast is the separately named Fast profile, selected with doubao-seedance-2-0-fast-260128. It supports video generation, editing and extension using text, images, video and audio references. The profile is useful when a scene brief needs more than a description of appearance: existing media can also communicate performance, camera language and sound. The Fast name distinguishes the request identity; it is not a promise that an individual asynchronous task finishes within a fixed time.

Start with the element that should define the new shot. Text alone can describe a new scene. An image can establish a character, product or composition; a video can provide motion or an existing sequence to revise or continue. The documented content combinations include images with audio, video with audio and images, video and audio together, with text optional alongside visual media. These combinations do not describe a standalone audio-only request, so pair sound references with the visual source they should guide.

Keep reference roles explicit in the content array and in the brief. image_url, video_url and audio_url items can carry reference_image, reference_video and reference_audio roles. Describe which image supplies the subject, which video supplies the camera or action and which audio supplies the soundtrack. For a product shot, an image might define the packaging while a video illustrates a movement past the object. Specify the desired ending so the short sequence has a useful closing view rather than stopping in the middle of an action.

For editing, identify the source footage, the change and the continuity to retain. A background or lighting change should leave the intended subject and action clearly specified. For extension, explain what happens next and how the new movement connects to the existing scene. References guide the generation; inspect the actual join and complete clip for appearance, motion direction and timing rather than assuming the continuation preserves every frame-level detail.

The documented 2.0 Fast range is four to fifteen whole seconds, with -1 available for a model-selected duration within that range. The default duration is five seconds. Output resolution is 480p or 720p, with 720p as the default and 1080p excluded. Choose among fixed 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9 shapes, or adaptive framing. Adaptive framing follows the relevant prompt or reference context; a first-frame shot follows its opening image, while multimodal framing can prioritize video over images. Read the returned task fields for the actual duration and ratio.

Use generate_audio to choose whether the generated video includes synchronized sound. The documented default is true; false requests silence. With audio enabled, the profile generates voices, effects or background music from the prompt and visual content, and the guide recommends enclosing dialogue in double quotation marks. State the speaker and intended relationship between sound and action. Review speech with a fluent listener and check sound alignment across the clip, including any edited or extended segment.

The 2.0 Fast guide excludes the older frames and camera_fixed controls and does not support flex offline processing. Describe camera behavior in the scene and use duration for the clip length. Seed and watermark options remain documented. The same seed is not a guarantee of an identical output, so retain the completed artifact when comparing revisions instead of relying only on the request settings.

Video-token usage is calculated as (input-video duration + output-video duration) × output width × output height × output frame rate / 1024. Requests containing input video and those without input video have distinct billing categories. The task query reports usage.completion_tokens and usage.total_tokens; retain these reported quantities with the completed clip.

Submission returns a task identifier before the completed video is available. Follow processing status, then retrieve content.video_url after success and keep the reported duration, resolution, ratio, frame rate and usage with the result. The expiry interval is measured from task creation, defaults to 172800 seconds and ranges from 3600 to 259200 seconds. Save the final clip with its references and prompt, then check subject identity, object contact, camera continuity and sound timing through the full sequence.