One click, whole workflow
Recipes are here. Package a whole workflow and run it in one click
Explore recipesSeedance 2
byByteDance (Seed team)
Cinematic single takes where picture, speech and foley are born together, held unbroken for up to 30 seconds
How Seedance 2 works
Write the shot like a director's note: one subject, one move, the sound. Add references when identity or camera language must stay fixed, and let the model render picture and soundtrack in a single pass.
Write the shot
Describe subject, one camera move and the sound, then anchor longer beats with timestamps like 0-5s and 5-12s.
Attach your references
Tag character stills, a camera-move clip or a voice clip as @Image1, @Video1 and @Audio1, and say what each one controls.
Render picture and sound
Pick a tier, resolution and length up to 30 seconds, and get one continuous take with its soundtrack generated alongside.
What Seedance 2 is good at
Built for shots that need to feel filmed rather than assembled: long unbroken takes, locked characters, believable cloth and water, and sound that arrives with the image.
Made with Seedance 2
A gallery of single-take clips across runway, product design, architecture and sculpture, each with its sound design written into the prompt.
What people build with Seedance 2
Directors, brand teams, fashion creatives and social producers reach for it when a shot needs sound and picture born together and a camera that never cuts.
Seedance 2 specs
Plans and credits- Versions
- Seedance 2.5, Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini
- Inputs
- Prompt (text), Images (up to 30 images), Videos (up to 10 videos), Audio (up to 10), End Image (image)
- Output
- video
- Duration
- 4–30 seconds
- Resolution
- 480p, 720p, 1080p, 4K
- Aspect ratios
- 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
- Audio
- Generates audio
- Cost
- 291–64,275 credits per run
Seedance 2 questions
Seedance 2.5 generates up to 30 seconds with far more references; 2.0 caps at 15 seconds. On this node 2.5 accepts up to 30 images, 10 videos and 10 audio clips, while 2.0 takes up to 9 images and 3 videos and has no audio reference input. Details on 2.5 come partly from community reporting.
Use Fast or Mini to draft, then 2.5 or 2.0 Standard for finals. Fast is quicker and cheaper than Standard with less refined detail; Mini is the cheapest tier with the lowest fidelity, suited to high-volume ideation. Both are capped at 15 seconds. Draft at 480p or 720p, then re-run the winning prompt at 1080p or 4K.
Use text-to-video for concept-first shots, image-to-video to animate one approved still, and reference-to-video for multi-asset jobs. Image-to-video uses your still as the first frame and can land on an end image. Reference mode suits character sheets, a camera-move clip and a voice clip tagged as @Image1, @Video1 and @Audio1. Keep to roughly 4-5 assets with clear roles.
Yes. Audio is on by default and produces effects, ambience and lip-synced speech from what you write in the prompt. Put dialogue in quotes and name the language for non-English speech, since lip sync can degrade in some non-English languages. Turn audio off if you plan to score the clip in post.
It struggles with readable on-screen text in motion, small-object hand interactions and crowded scenes. Text can morph into pseudo-glyphs, fingers can warp and crowds go stiff. It's also a poor fit for hard editorial cuts inside one clip, and re-rolling at top resolution and 30 seconds gets expensive.
Seedance comes from ByteDance's Seed team, with the earliest release listed as 2026-02-12. Seedance 2.0 dates from February 2026; 2.5 was reported later in 2026, and its specifics come largely from secondary reporting.
Try Seedance 2 on Fuser
Cinematic single takes where picture, speech and foley are born together, held unbroken for up to 30 seconds
