Seedance 2.5: playground and API

Model guide

Build with Seedance 2.5.

Best for

ByteDance Seedance 2.5. Generates video from a prompt alone, or from reference material: up to 8 reference images and videos in any mix (5 on the older inline route), plus up to 2 reference audio tracks. Use the playground to set up a single step, then open a workflow when you want to connect more tools.

Inputs explained

Prompt (required): Set the prompt used by this model. Type: text.
Start Frame (optional): The video opens exactly on this image. Use it when you have already composed the shot (framing, background, camera position) and want it kept. Costs one of the registered materials (up to 8 per run, 5 on the older inline route). Pair it with an End Frame for a start-to-end shot. It cannot be combined with any reference media (Reference Images, Reference Audio, Audio): Seedance runs a start/end-frame task OR a reference task, never both (the mix is refused before you are charged). With a Reference Video wired, this image is sent as one more reference image instead, so the video no longer opens on it. A Start Frame shot cannot take your own voice: with Generate Audio on, the model makes its sound. Type: image.
Reference Image 1 (optional): An identity and style guide, not a framing guide. The model may reframe freely, so a carefully composed shot wired here comes back as a DIFFERENT shot. If you want the video to open on this exact image, use Start Frame instead. Type: image.
Reference Video (optional): Set the reference video used by this model. Type: list of video.
Reference Audio (optional): Audio the model listens to; this track never plays in the result. In an edit (a Reference Video wired and a prompt that changes it, like "replace the man in @Video1"), wire the audio of that same clip here, turn Generate Audio off, and add this line to the prompt: "Match the framing of @Video1 in every frame: same face size and position, same distance from the camera. Never move closer or crop tighter." Without it, the audio can pull a wide shot into a close-up. The mouth then follows the timing and big shapes of this track, close but not frame perfect, and the clip comes back silent. Check its length against the clip (short clips have come back about 6 percent fast), retime it in the Video Editor if it is off, then lay the audio back in. Keep Generate Audio off when the track is a copyrighted song, because with it on Seedance can refuse the generated audio for copyright. With Generate Audio on, Seedance re-voices the track (timbre, timing and words), so the lips will not match it exactly. Without a Reference Video, it cannot be combined with a Start Frame (the mix is refused before you are charged). mp3 or wav only. Type: list of audio.
Aspect Ratio (optional): Set the aspect ratio used by this model. Type: text. Options: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive. Default: "16:9".
Duration (optional): Set the duration used by this model. Type: text. Options: Auto (model decides), 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30. Default: "5".
Resolution (optional): Set the resolution used by this model. Type: text. Options: 480p, 720p, 1080p. Default: "720p".
Generate Audio (optional): On: the clip comes back with sound the model generates, guided by any Reference Audio. Off: the clip comes back SILENT, and it never keeps your own track. For lip sync in an edit, turn this off, wire the audio of the clip being edited into Reference Audio, and add this line to the prompt: "Match the framing of @Video1 in every frame: same face size and position, same distance from the camera. Never move closer or crop tighter." The mouth then follows the timing and big shapes of that track, close but not frame perfect; check the take length against the clip and retime it in the Video Editor if it is off, then lay the audio back in. Keep it off when that audio is a copyrighted song, because with it on Seedance can refuse the generated audio for copyright. Outside an edit, with this on, Seedance re-voices any Reference Audio, so the lips will not match your own track exactly. Type: boolean. Default: true.
Watermark (optional): Set the watermark used by this model. Type: boolean. Default: false.
Content filter (optional): On by default. Seedance screens your prompt, your reference material and the finished clip, so a run can fail after it has been submitted when something gets flagged (copyright is the common one). Turning this off asks Seedance to skip that safety and IP moderation. It costs 10 percent more per run, and you are responsible for holding the rights to whatever you generate. Type: boolean. Default: true.
Reference Image (from Seedance 2.0) (optional): An image arriving from the Seedance 2.0 card. It is used as a reference, not as an opening frame: the model may reframe freely. To make the video open on an exact image, wire the Start Frame port above. Type: image.
End Frame (optional): Optional last frame. Beside a Start Frame the video ends exactly on this image (and, like the Start Frame, it cannot be combined with reference media). With no Start Frame it is treated as one more reference image, so it guides the ending rather than pinning it. Type: image.
Reference Images (optional): Identity and style guides. The model may reframe freely, so these steer WHO and WHAT is in the shot, not how it is framed; to pin the framing use Start Frame. Start Frame, End Frame and any reference VIDEO all draw on the same budget of registered materials per run (up to 8, or 5 on the older inline route). Type: list of image.
Audio (optional): Optional reference audio. Shares the reference-audio slot with Reference Audio, and wiring both concatenates them: up to 2 tracks and 30 seconds between the two. It behaves exactly like Reference Audio (see that port for lip sync in an edit) and never plays in the result. Without a Reference Video, it cannot be combined with a Start Frame. Type: list of audio.
Aspect Ratio override (set by Seedance 2.0) (optional): Leave this empty. It is filled in automatically when the Seedance 2.0 node routes a run into this model, and while it has a value it OVERRIDES the Aspect Ratio field above ("auto" lets the model choose from your reference media). To change the shape of your video, use Aspect Ratio, not this. Type: text. Options: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, auto.

Tips before you run

Start with the default settings and change one input at a time. Keep the result you like on your board so the next step can use it.

If you add a reference file, use a link the workflow can read. Check the inputs and credit estimate on the board before starting a run.

Working with this model

Start Frame, reference images and the budget

Start Frame is first_frame_url. Wire it when the video must open on an exact image you already composed. image_url is NOT the Start Frame: it is the port the Seedance 2.0 card bridges into, and it is always sent as a plain reference image, so the model may reframe freely. Setting image_url to get a first-frame run gets you a reference-mode run, and you pay for it.

References (Reference Images, Reference Video, and a Start or End Frame) draw on one shared budget of registered materials, images and videos together. The budget is up to 8 per run, or 5 on the older inline route.

Content filter, and what turning it off costs

Seedance moderates a run in three separate places: your prompt, the reference images and video you register, and the FINISHED CLIP. The last one is the surprising one. A run can pass the first two checks, generate, and then fail after submission with a message like "the request failed because the output video may be related to copyright restrictions". Nothing you can see on the canvas predicts it.

Content filter is on by default and that is the behaviour every existing board already has. Switch it off and Wireflow sends content_filter: false, which asks Seedance to skip that safety and IP moderation.

Two things to know before you do:

It costs 10 percent more. The run is billed at a higher rate, so the quote on the node, the credit gate before the run, the charge and the final settlement all go up by 10 percent. The number on the card is the number you pay, either way.
The rights are yours to hold. Skipping the filter does not give you a licence to anything. You are responsible for having the rights to the material you feed in and to whatever comes out.

Leave it on unless a run has actually been refused and you know the footage is yours to use.

Realistic talking-head video

Animating an avatar still? Make a Realistic AI Avatar has the tested recipe: reference mode for head turns, and ship the raw 720p output with no video upscale.

Common questions

How do I use Seedance 2.5 on Wireflow?

Set the inputs above and choose Open in a workflow. Sign in to save your board, then check its credit estimate before you run.

Can I use Seedance 2.5 through an API?

Yes. Save a workflow with this model, then call the Wireflow run API with its workflow ID. The API tab has examples you can copy.

What changes the credit price?

The price shown uses the model defaults. Your inputs and model settings can change the total. The workflow shows an estimate before you run.

Can I connect this to another model?

Yes. Open it in a workflow and connect its output to another node with a matching input type.

Input

Result Ready / example