Use your character in Seedance 2.0
Seedance 2.0 has two paths that map cleanly onto the pack: image-to-video with an optional end frame, and reference-to-video where up to nine images are named in the prompt as @Image1, @Image2 and so on. The tagged references make it the easiest provider for telling the model which file is the face and which is the outfit, and clips run up to 15 seconds with synchronized audio.
Capabilities
- First frame (image-to-video)Yes
- Last frameYes
- Reference images9
- Native extendNo
- Max clip15 s at 480p or 720p on the fal model page (1080p and 4K appear on some fal schema pages; verify)
- AudioYes
Which files to use
hero.pngimage_url, the starting frame (image-to-video), or @Image1 in reference-to-video.closeup.png@Image2 in reference-to-video: face identity. Name it in the prompt as the face reference.three_quarter.pngend_image_url for a shot ending angled to camera, or @Image3 as an extra view.profile.png@Image4 reference, or end_image_url for a turn-to-profile shot.back.pngend_image_url for a walk-away shot, or @Image5 if the action shows the back.contact_sheet.pngOptional single reference when slots are scarce; tagged references work better than a tiled sheet.character.jsonBible text in every prompt; the seed field can be reused but only gives approximate repeatability.
Steps
- Sign in to fal.ai and open bytedance/seedance-2.0/image-to-video for a shot that starts on the sheet pose, or bytedance/seedance-2.0/reference-to-video for a shot that does not.
- Image-to-video: upload hero.png as the start image and, if the shot should end on a known pose, one of three_quarter.png, profile.png or back.png as the end image.
- Reference-to-video: upload hero.png, closeup.png, three_quarter.png and profile.png in order; they become @Image1 to @Image4 and the prompt must say what each one is for.
- Paste the bible text from character.json, then the action, camera and audio intent. Set duration (auto or 4 to 15 seconds), aspect ratio and resolution, and leave audio generation on unless you will add your own.
- Run the job; the playground shows progress and the result URL. Download the MP4 when it finishes.
- For a longer scene, grab the last frame of the result and use it as the next start image with the same bible text; there is no extend endpoint on fal for this model.
Prompt template
@Image1 is the character: {{bible}} @Image2 is the same character, use it for the face. Keep face, hair, skin tone, outfit and proportions identical across every frame. {{action}} Natural motion, single subject, consistent lighting, synchronized ambient audio. (In image-to-video mode drop the @Image tags; the start frame already defines the character.)
Tips
- Always say what each tagged image is for. Seedance does not infer roles from the files alone, so "@Image2 is the face" matters.
- Use the end image for turn-arounds: hero.png to back.png produces a clean rotation that preserves outfit detail from both views.
- Keep aspect ratio on auto for image-to-video so the clip follows the 2:3 still; set 16:9 only when you have cropped the start image yourself.
- The seed gives approximate rather than exact repeatability; store the seed from character.json anyway so retries stay close.
- The fast variants (bytedance/seedance-2.0/fast/...) take the same inputs and are good for blocking out motion before a full-quality render.
- Audio needs at least one image or video reference in reference mode, so do not strip the images to speed up an audio-only test.
Caveats
- Resolution options conflict across fal pages: the model page lists 480p and 720p (default 720p) while some schema and comparison pages list 1080p and 4K. Check the live schema for your account.
- The reference images field is image_urls in reference-to-video; prompt tags are written @Image1 on the official page and [Image1] on at least one other fal page.
- Reference-to-video allows up to 9 images and at most 12 files across images, video and audio. Whether a tagged image can also be forced to be the exact first frame is suggested by prompt guides ("@Image1 as the first frame") but not documented as a parameter.
- No extend endpoint for Seedance 2.0 on fal was found; the capabilities row marks native extension as false.
- Pricing is per output second (about 0.30 USD per second at 720p in fal pricing at the time of writing) and changes; a 15 second 4K clip is expensive, so test at 480p.
- Seedance 2.5 exists on fal with a different schema; this guide covers 2.0 only.
API notes
Call bytedance/seedance-2.0/image-to-video (prompt, image_url, optional end_image_url, duration, resolution, aspect_ratio, generate_audio, seed) or bytedance/seedance-2.0/reference-to-video (prompt with @Image tags, image_urls array) through the fal client: fal.subscribe waits for the result, or fal.queue.submit returns a request id that you poll for status and then fetch the result. Images are passed as public URLs; upload local pack files first with the fal client storage helper to get a URL. The result carries the MP4 at data.video.url.