Seedance 2.0 Fast is a fast variant of ByteDance's Seedance 2.0 series, suitable for short-film concepts, character shot tests, and advertising asset iteration. doubao-seedance-2-0-fast-260128 supports text, image, and audio/video references, and can generate videos with sound, delivering short clips in 480p or 720p while balancing character references and creative control.
Submit requests to the public API at api.acedata.cloud using the documented parameters, then use the results in your application.
Input parameters and result formats vary by service. Use the public API for this model and follow its guide for generation, task retrieval and editing operations.
The above are the available specifications and control methods for this model on this platform; API parameters do not represent all native model capabilities.
Use reference_image to provide a character or subject appearance, then describe new scenes, clothing, actions, and camera movement in text. It is suitable for creations where “the same person enters different stories,” without treating the photo background as a fixed scene; clear, front-facing, unobstructed character assets make it easier to establish references.
In addition to images, you can add reference_video and reference_audio to guide actions, camera movement, and sound rhythm. Enable generate_audio to generate video with sound; prompts should separately explain what happens visually, how the camera moves, and what you want to hear, avoiding conflicting intentions across multiple assets.
Use first_frame when motion needs to begin from a specified image, and add last_frame when you need to control the ending image; use reference mode when you want to redesign the composition. You can also reduce camera movement changes with a fixed camera and request the final frame to check how the shot concludes or prepare the starting image for the next segment.
Upload authorized character photos and a single-shot description, such as a person walking in a park and turning to smile, then select a portrait aspect ratio to generate a short video. Keep the reference character unchanged while modifying the scene and actions separately to deliver a set of candidate storyboards for selecting character performance and narrative direction.
Use a product image as the first frame, specify a camera push-in, lighting changes, or environmental motion, validate the concept in 480p first, then generate the selected version in 720p. If the background needs to be rearranged, use the subject reference mode instead; the deliverable is short shots for editing selection, not a complete advertising project.
Write scenes such as rainy streets, beaches, or indoor activities as prompts, and use legal audio and video materials to guide the rhythm while enabling audio generation. Output audiovisual drafts in landscape or widescreen aspect ratios for directors, animation teams, or content editors to discuss environmental atmosphere, motion direction, and sound design.
When the delivery target is 480p or 720p short shots and character and audio-video references are needed, 2.0 Fast is suitable for creative prototyping and multi-option selection. If the task explicitly requires 1080p or 4k, choose the 2.0 standard version; Mini is a lightweight option in the same series. Fast's positioning does not mean every task has a fixed completion time.
2.0 Fast is suitable for generating new 4–15-second shots and reference videos should not be understood as a full video editing mode. If you need up to 30 seconds, creation based only on audio references, or explicit editing and extension of existing videos, choose Seedance 2.5. The selection should be based on the required operation, not just the model generation.
In content, assign specific roles to images, videos, and audio, and explain the purpose of each asset in the text; organize first-and-last-frame constraints and subject references separately according to the task.
Specify model=doubao-seedance-2-0-fast-260128 for /seedance/videos; first test with duration=5, resolution=720p, and a clear aspect ratio. When audio is needed, explicitly set generate_audio=true; it is disabled by default.
For asynchronous processing, first obtain the task_id, query /seedance/tasks or receive a callback; after completion, check the subject, actions, ending, and audio, then save the finished video and returned final frame as needed.
The character reference image defines the clothing, the video reference defines a slow turn, and the character completes the turn in a bright studio while the camera remains steady.
Use a 720p short shot to verify the subject and action, then arrange versions with and without sound; do not use standard edition 4k settings for Fast.
reference_image is used to guide the appearance of a character or subject, while the scene, clothing, and action can be described anew; first_frame makes the video begin moving from the given image. Choose the first frame if you want to preserve the original image composition, and choose a reference image if you want the same character to enter a new scene. Do not mix the two modes.
Yes, enable it by setting generate_audio=true; audio is not generated by default. Prompts can describe ambient sounds or action sounds; reference_audio is used as a reference for sound and rhythm, and does not mean the final video will necessarily reproduce the input audio word for word or segment by segment.
This version uses 480p, 720p, and a 4–15 second range, and automatic duration can also be selected. For 30-second tasks, choose Seedance 2.5; for 4k tasks, choose the 2.0 standard edition. Automatic duration lets the model determine the length; it does not remove this model's duration limit.
Fast can use a video as reference_video to help guide the action and camera movement of a new shot, but do not treat it as an edit or extend operation. When you need to explicitly modify an existing video or extend a clip, choose 2.5 and organize assets and parameters according to the corresponding task mode.
Submit the exact model value and content array to /seedance/videos. Text items can contain up to 1000 characters, and image addresses use image_url objects. You can set async=true and then query task_id, or configure callback_url to receive completion notifications, then retrieve the video from data.video_url.
Choose a development approach or application suited to this model and follow its setup guide.
Python SDK Integration Guide
SDK installation, authentication and examplesTypeScript SDK Integration Guide
SDK installation, authentication and examples
