Seedance is ByteDance's multimodal video model, available on Higgsfield and built for realistic motion. It takes text, images, video clips, and audio as inputs at once and generates video with native sound in a single pass, with lip sync in multiple languages. Start with image-to-video at 720p before scaling up, and add audio only on your final pass to save credits while iterating.
What is Seedance 2.0?
Seedance is ByteDance's video model family for realistic motion. It accepts text, images, video clips, and audio at the same time and generates video with native sound in a single pass. For character consistency within a clip, it uses the reference system; for the same character across separate generations, save the character as an Element and reference it in the prompt.
Variant | Best for | Resolution |
|---|---|---|
Seedance 2.0 | Maximum quality, full multimodal input | 480p to 4K |
Seedance 2.0 Fast | High-volume iteration | 480p to 720p |
Seedance 2.0 Mini | Maximum speed, short-form content | 480p to 720p |
How do I generate my first clip?
Open higgsfield.ai → Video → Seedance 2.0.
Step 1: start with image-to-video at 720p. Upload a sharp, front-facing, well-lit reference image and add a short prompt describing the action and camera move. Keep the first generation short: image-to-video is more predictable than text-only and uses fewer credits while you learn how the model responds.
Step 2: write a structured prompt. Cover four things in order: what the subject is doing, setting and lighting, camera move, mood or atmosphere. Vague prompts produce generic motion, specific prompts produce intentional shots.
Step 3: review the full clip before iterating. Most quality problems appear at the 5 to 8 second mark: check whether the motion holds throughout and whether the camera move executed.
Step 4: change one thing per iteration. If something is off, change the prompt or the reference, not both at once.
Step 5: scale to higher resolution. Once your prompt produces the right composition at 720p, run it at 1080p or 4K with the identical prompt, and add native audio at this step.
The generation bar covers the key settings: clip duration, aspect ratio (including Auto, which infers the ratio from your input), resolution up to 4K, and bitrate. The credit cost updates on the Generate button as you change them.
How do I write a good prompt?
Use this structure: subject and action first, setting and lighting second, camera move third, mood or style last.
Camera terms Seedance 2.0 reads directly: dolly in, truck left, arc shot, push in, pull back wide, handheld follow, crane up, orbital move.
Example: "A character walks through a crowded train station at rush hour, checking a phone, camera tracking close at shoulder height, warm overhead fluorescent light, tense and rushed. SFX: station ambient noise, announcements in the background."
How does the reference system work?
Reference type | What it does |
|---|---|
@character | Holds face geometry, skin tone, and style consistent within the clip |
@style | Applies lighting, palette, and mood from an image or film still |
@motion | Reads camera behavior and motion pattern from a video clip and replicates it |
@audio | Syncs visuals to audio, generates lip sync, matches ambient sound |
You can combine up to 9 images, 3 video clips, and 3 audio files in one generation. Start with one or two references.
How do I keep a character consistent across multiple generations?
@character references hold a face consistent within a single generation. For the same character across multiple separate generations, save it as an Element: click @ Elements under the prompt field (or type @ in the prompt) and select the character, with no re-uploading per clip. If the character is a real person, train a Soul ID first and create the Element from its portraits. See How do I create and use a Soul ID character?
Common mistakes and fixes
Output ignores the camera instruction. Replace vague terms with specific ones: dolly in, pull back wide, arc right, crane up.
Character looks different halfway through. Add more specific visual anchors: hair color, clothing details, distinguishing features. For cross-generation consistency, reference the character as an Element.
Motion looks jerky at the cut. Generate the final frame of one clip and the opening frame of the next as references, then use first-and-last-frame to generate the transition between them.
Credits running out faster than expected. Audio and high resolution drive credit consumption: prototype without audio at 720p, lock your prompt first, then run the final version with audio. The exact cost is always on the Generate button.
Can I use Seedance in Canvas, MCP, or Supercomputer?
Yes. In Canvas, connect a reference to the Seedance node and describe the reference's role at the start of the prompt. All generations through Canvas, MCP, CLI, and Supercomputer always deduct credits regardless of any Unlimited access on the web. See How do I use Canvas for multi-node generation? and What is Higgsfield MCP, and how is it different from the web app?