One model. One take. Picture and sound, directed from your references.
Seedance 2.5
30 seconds in one pass · 50 references · native 1080p
Sound rolls with the picture
Dialogue, footsteps, room tone and score are generated in the same pass as the frames. Nothing to sync afterwards.
- Lip-synced dialogue in the language of your prompt
- Ambience that follows the camera
- Score that ends when the shot ends
Cast the film from 50 references
Faces, wardrobe, products, locations and style frames go in together. The model holds every one of them for the full thirty seconds.
Fix the frame, not the film
Mark one region and describe the change. Everything outside the mark stays exactly as it was, frame for frame.
Swap the label on the bottle
Clear the crowd behind her
Change the coat to red
Show the move once
Film a stand-in doing the action. Seedance 2.5 transfers the timing and the weight of that motion to any character in your references.
Every frame you publish in
Pick the aspect ratio before you generate. Composition is planned for that frame, not cropped into it.
Spec sheet
Seedance 2.5 is ByteDance's video generation model. On Plot Party it runs in the browser generator and through the API and MCP, so the same clip you direct here can be scripted from Claude, ChatGPT or your own pipeline.
- Model
- Seedance 2.5 by ByteDance Seed · model id seedance-2.5
- Clip length
- 4 to 30 seconds per generation, in 1-second steps, as one continuous take
- Resolution
- 480p, 720p or native 1080p
- Aspect ratio
- 16:9, 9:16, or adaptive, which follows the frame of your references
- Audio
- Dialogue, ambience and score generated in the same pass as the picture, lip-synced to the prompt language
- References
- Up to 50 per generation: 30 images, 10 videos and 10 audio clips, cited in the prompt as @Reference1, @Reference2 …
- Modes
- Omni (text plus references), first and last frame, extend a source video, region edit a source video
- Prompt
- Up to 10,000 characters of natural language, including dialogue lines
- Render time
- Typically around five minutes for a 1080p clip
- Access
- Plot Party generator, REST API and MCP server
2.0 → 2.5
What changed from Seedance 2.0 to Seedance 2.5, as the two models run on Plot Party.
- Longest clip per generation
- 2.015 s2.530 s
- References per generation
- 2.0122.550 (30 images · 10 videos · 10 audio)
- Region editing
- 2.0Not available2.5Mask one region of a source video, keep the rest frame for frame
- Extend a source video
- 2.0Via omni continuation2.5Dedicated extend mode, 4 to 30 s
- Modes
- 2.0First frame · first and last frame · omni2.5Omni · first and last frame · extend · edit
- Aspect ratios
- 2.0Adaptive · 16:9 · 9:16 · 4:3 · 3:42.5Adaptive · 16:9 · 9:16
- Resolution
- 2.0480p · 720p · 1080p · 4K2.5480p · 720p · 1080p
- Prompt length
- 2.05,000 characters2.510,000 characters
| FEATURE | SEEDANCE 2.0 | SEEDANCE 2.5 |
|---|---|---|
| Longest clip per generation | 15 s | 30 s |
| References per generation | 12 | 50 (30 images · 10 videos · 10 audio) |
| Region editing | Not available | Mask one region of a source video, keep the rest frame for frame |
| Extend a source video | Via omni continuation | Dedicated extend mode, 4 to 30 s |
| Modes | First frame · first and last frame · omni | Omni · first and last frame · extend · edit |
| Aspect ratios | Adaptive · 16:9 · 9:16 · 4:3 · 3:4 | Adaptive · 16:9 · 9:16 |
| Resolution | 480p · 720p · 1080p · 4K | 480p · 720p · 1080p |
| Prompt length | 5,000 characters | 10,000 characters |
From the crew
“We storyboarded a 30-second spot on Monday and had a client-ready cut with sound by Tuesday lunch. The lip-sync held on every take.”
“The audio isn't an afterthought. Footsteps land on the beat, the room tone matches the set. That saves me a full sound pass.”
“Our trailer was shot entirely from concept art. Product shots, character turns, a whole storefront that never existed.”
“I write the dialogue, it delivers the dialogue. In two languages, same voice, same face. My editors thought I hired an actor.”
“Same character across eleven shots, three of them with the camera moving hard. I stopped checking for face drift after the fourth.”
“I feed it a reference clip of the move I want and it just copies the blocking. That's a week of previz in one prompt.”
“Eight references in, one clip out. Costume, prop, location, all consistent. My art department asked what changed.”
“1080p straight out, no upscale step, no re-render for the vertical crop. We ship to five placements from one generation.”
Questions
What is Seedance 2.5?
Seedance 2.5 is ByteDance's video generation model. It produces a single continuous take of 4 to 30 seconds with picture and sound generated together, and it can be steered with up to 50 image, video and audio references. On Plot Party you can run it in the browser generator or call it through the API and MCP with the model id seedance-2.5.
How is Seedance 2.5 different from Seedance 2.0?
The clip limit doubles from 15 to 30 seconds in one pass, and the reference limit goes from 12 to 50, now including video and audio references. Seedance 2.5 adds two modes that 2.0 does not have: region editing, which changes one masked area of a source video and leaves the rest untouched, and a dedicated extend mode. Prompts can be twice as long, up to 10,000 characters. Seedance 2.0 remains available on Plot Party for 4:3 and 3:4 output and 4K.
How long can a Seedance 2.5 video be, and at what resolution?
Any length from 4 to 30 seconds, in 1-second steps, as one continuous shot rather than stitched clips. Output is 480p, 720p or native 1080p. In edit mode the output keeps the length of the source video you are editing.
Does Seedance 2.5 generate audio and lip-synced dialogue?
Yes. Dialogue, footsteps, room tone and score are generated in the same pass as the frames, so there is nothing to sync afterwards. Write the lines in your prompt and the character speaks them, lip-synced, in the language you wrote. You can also give it up to 10 audio references, for example a voice or a music cue, alongside your images.
How many references can I use, and what kinds?
Up to 50 per generation: as many as 30 images, 10 videos and 10 audio files. Audio cannot be the only reference. Mention each one in the prompt as @Reference1, @Reference2 and so on, in the order you uploaded them, so the model knows which face, product, location or motion clip you mean.
How does Seedance 2.5 keep characters consistent?
Upload a clear reference photo of each character, product and location and name it in the prompt. The model holds every reference for the full length of the take, so the same face, wardrobe and set persist across the shot instead of drifting. For repeatable movement, add a short video of the action as a motion reference and the timing carries over to your character.
Which aspect ratios does Seedance 2.5 support?
Pick 16:9 or 9:16 before you generate, or leave it on adaptive and the model composes for the frame of your references. Composition is planned for the chosen frame rather than cropped from a wider one.
Can I edit or extend an existing video with Seedance 2.5?
Yes. Edit mode takes a source video of 4 to 30 seconds, a mask over the region to change and a description of the change; everything outside the mask stays identical frame for frame. Extend mode continues a source video with new footage of 4 to 30 seconds. Both accept extra image references for the new content.
How is Seedance 2.5 billed on Plot Party?
You pay in Plot Points per second of video generated, so a 10-second clip costs twice a 5-second one. Resolution changes the rate: 1080p costs more per second than 720p, and 480p costs less. Edit mode is billed on the length of the source video. Current rates and plans are on the pricing page.
Is Seedance 2.5 available through an API?
Yes. The Plot Party API and MCP server expose Seedance 2.5 as seedance-2.5 with the same modes as the generator: omni, first and last frame, extend and edit. Upload references first, then pass their URLs in reference_urls. The API docs list every parameter.
Can I use well-known characters or brands in a prompt?
No. Prompts that mention well-known intellectual property are blocked before generation. Use your own characters, products and reference images instead.
That's a take.
Thirty seconds, picture and sound, from the references you already have.