Generate up to 30 seconds, direct the result with as many as 50 multimodal references, and refine specific moments with frame-level precision.
Start from text or combine images, video, and audio in the reference workflow. The current backend is a preview route and may apply stricter input moderation.
Final usage is the selected resolution rate multiplied by output duration.
Similar Models
A multimodal video generation model for text prompts, first-frame animation, and reference-guided production clips.
Create 4 to 10-second videos from text, up to seven reference images, or one short reference video with synchronized sound.
Create fixed 2K videos from prompts or coordinated image, video, and audio references, with native sound and flexible 5-15 second delivery.
Model overview
Seedance 2.5 Preview on sjolt exposes separate text-to-video and reference-to-video routes for 4 to 30-second output. Start from text alone, or guide a generation with up to 30 images, 10 MP4 clips, and 10 audio references through the asynchronous task API.
Generate any whole-second duration from 4 through 30. Use the longer timeline to build a complete scene with an opening, development, camera progression, synchronized sound, and a deliberate ending instead of stitching together several short clips.
Combine up to 30 images, 10 MP4 clips, and 10 audio files in one reference-guided request. Assign each source a role with @ mentions to preserve characters and products, borrow camera movement, match visual style, and coordinate dialogue, music, or ambience.
Describe changes at specific moments and anchor them with reference frames, clips, and audio cues. Target a character, action, prop, background, or sound beat while helping the surrounding identity, motion, lighting, and audiovisual continuity remain coherent.
Model characteristics
Plan a complete narrative or visual arc with enough time for scene development, measured camera movement, transitions, and a resolved ending.
Bring together 30 images, 10 videos, and 10 audio files for identity, style, motion, camera, dialogue, and sound guidance.
Use time-ordered instructions and multimodal anchors to focus changes on a particular moment or visual element while maintaining continuity around it.
FAQ
Seedance 2.5 Preview is a ByteDance video generation model with separate routes for prompt-only generation and multimodal reference guidance.
Upload up to 30 PNG, JPG, JPEG, or WebP images, 10 MP4 videos, and 10 MP3, WAV, M4A, AAC, or OGG audio files.
The text-to-video route does not accept reference files. The reference-to-video route requires at least one image, video, or audio URL, while the prompt remains required on both routes.
Choose a whole-second duration from 4 to 30, 480p or 720p output, and 16:9, 9:16, 1:1, 3:4, 4:3, 21:9 framing.
State what must remain stable, assign each uploaded reference a role, then describe the action, camera movement, scene changes, lighting, and sound in chronological order.
Submit the text-only or reference-guided request to its dedicated endpoint. Poll the returned task ID, then read the generated MP4 URL from the successful task result.
