What is Seedance 2.5?
Seedance 2.5 is the newest video model from ByteDance's Seed team and the main Seedance model on BytePlus ModelArk, where it is listed as Dreamina Seedance 2.5. It generates picture and sound together from text, images, video clips or audio, in one take of 4 to 30 seconds at up to 1080p, and it can also edit or extend an existing clip. It is a closed model with no open weights.
Seedance 2.0 is still on sale next to it. Version 2.5 doubles the clip length and more than triples the reference limit, but it stops at 1080p where 2.0 goes to 4K, and it costs about half as much again per token.
What changed from Seedance 2.0?
Longer takes, far more references and a draft step, for about half as much again per token and a lower top resolution.
- Length
- Up to 30 seconds per generation, up from 15.
- References
- Up to 50 files (30 images, 10 clips, 10 audio files), up from 15, and audio alone can now drive a video.
- Output
- 1080p in 10-bit color with a new MOV option; Seedance 2.0 still offers 4K.
- Price
- $10.70 per million tokens at 480p and 720p without a video input, against $7.00 for Seedance 2.0.
Who is Seedance 2.5 best for?
Seedance 2.5 is best for video teams and developers who need takes of up to 30 seconds with sound, steered by many reference images, clips and audio files, or edits to footage they already have.
Longer single takes
One generation runs up to 30 seconds and can be extended again, so a short scene does not have to be stitched from 5-second clips.
Video with its soundtrack
Dialogue, music and effects come out with the picture, in 11 languages, instead of needing a separate audio pass.
Scenes built from many references
Up to 30 images, 10 clips and 10 audio files can set the cast, props, setting and voices of one video.
Editing existing footage
It swaps subjects, adds or removes objects, replaces a green-screen background or re-stages camera moves in a clip you already have.
What's included with Seedance 2.5?
30-second takes with extensions, 50 reference files, editing by prompt and timestamp, clay-render and green-screen control, a 480p draft step, and sound in 11 languages.
- 30-second takes and extensions
- Each generation runs 4 to 30 seconds, and a finished clip can be extended forward or backward in more rounds.
- 50 reference files
- Up to 30 images, 10 video clips and 10 audio clips per request, and audio alone can now drive a video.
- Editing by prompt and timestamp
- Target a stretch of a clip to change a character, action or object, and the output keeps the source clip's length and shape.
- Clay-render and green-screen control
- Untextured 3D blockouts set the layout, blocking, camera path and light direction; green-screen edits replace the world around a subject.
- Draft mode
- Make a cheap 480p preview first, then turn the same draft into the final 1080p video within seven days.
- Sound in 11 languages
- Prompts and generated speech work in Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean.
How does Seedance 2.5 work?
Switch the model on
Put at least $30 toward it on BytePlus ModelArk, as account credit, an AI Savings Plan or a Seedance resource pack.
Send a generation task
Call the video generation API with a prompt and your reference files, or start from a 480p draft.
Collect the video
Poll the task until it finishes, then download the file within 24 hours, when the link expires.
How much does Seedance 2.5 cost?
Seedance 2.5 is a pay-as-you-go API on ByteDance's BytePlus ModelArk, at $10.70 per million tokens for 480p and 720p video (about $0.23 a second at 720p) and $11.70 per million tokens at 1080p (about $0.57 a second).
- About $0.10 a second at 480p and $0.23 a second at 720p
- A 5-second 720p clip costs about $1.16
- $6.40 per million tokens when the input includes a video
- About $0.57 a second, or $2.84 for a 5-second clip
- 10-bit color at 24 frames a second
- $7.00 per million tokens when the input includes a video
- 10 million tokens for $64 or 100 million for $640
- Expires 90 days after purchase
- Non-refundable
To switch the model on you need at least $30 in BytePlus credits, a $30 AI Savings Plan or $30 of resource packs. Token use is roughly the input plus output video seconds times width times height times frame rate, divided by 1,024, so a video input adds tokens even though its token rate is lower, and a minimum charge applies: a 5-second 720p clip made from a 30-second input video costs about $4.84. Pack tokens match the video-input rate, so jobs without a video input use about 1.7 to 1.8 pack tokens per token. Only videos that finish are charged. Prices are in US dollars. The model is also offered in ByteDance's Dreamina app, which uses its own credits and shows its prices only after sign-in. Full pricing on docs.byteplus.com. We checked these prices on .
How do readers rate Seedance 2.5?
No one has rated Seedance 2.5 yet. Be the first.
Sign in to rate. One rating per account.
Seedance 2.5 pros and cons
It keeps long takes and many references consistent, but it costs more than Seedance 2.0, stops at 1080p and refuses real faces as references.
Pros
- Holds characters, settings and style steady across a full 30-second take; users on Hacker News and Reddit say that means fewer re-rolls and less stitching.
- Turns a loose pile of reference images into a fairly coherent sequence, a result several early testers singled out.
- Follows character references more closely than Seedance 2.0, with finer facial detail, though testers say that makes it more literal, so storyboards need more care.
- Draft mode lets you check a scene at 480p prices before you pay for the 1080p render.
Cons
- Costs about half as much again per token as Seedance 2.0 on BytePlus, and early users complained that reseller apps charged two to three times their Seedance 2.0 price.
- Tops out at 1080p, while Seedance 2.0 still renders 4K.
- Rejects reference images and videos that show real human faces; you have to use BytePlus's preset digital characters or an authorized real-person asset.
- Many viewers still spot the AI look in faces, motion and pacing, and ByteDance itself says complex motion and scenes with several interacting people need work.
What are the best Seedance 2.5 alternatives?
Wan 3.0 for cheaper 30-second clips, Kling 3.0 for 4K at a lower rate, and Veo 3.1 for Google's API.
Wan 3.0FreeChoose it when you want 30-second clips with sound for less: Alibaba Cloud lists it at $0.10 a second at 720p, under half of Seedance 2.5's rate.
Kling 3.0FreemiumChoose it when you need 4K output or a lower rate with sound, $0.126 a second at 720p on its API, and 15-second clips are long enough.
Veo 3.1FreeChoose it when you build on Google's Gemini API or need 4K; Veo 3.1 Fast costs $0.10 a second at 720p with audio and $0.30 at 4K.
Questions about Seedance 2.5
What is Seedance 2.5?
Seedance 2.5 is ByteDance's video model for creators and developers: it makes clips of up to 30 seconds with sound, steered by up to 50 reference images, video clips and audio files.
Who is Seedance 2.5 for?
Seedance 2.5 is best for video teams and developers who need takes of up to 30 seconds with sound, steered by many reference images, clips and audio files, or edits to footage they already have.
Is Seedance 2.5 free?
No. BytePlus lists no free trial quota for Seedance 2.5, and the model switches on only after you put at least $30 toward it. ByteDance's Dreamina app hands out free credits, but it shows what they buy only after you sign in.
How much does Seedance 2.5 cost?
Seedance 2.5's paid plans are 480p and 720p at $10.70 per million tokens, 1080p at $11.70 per million tokens and Resource pack at $32 per 5 million tokens.
What are the best Seedance 2.5 alternatives?
The closest alternatives on Recaply are Wan 3.0, Kling 3.0 and Veo 3.1.
Where can I try Seedance 2.5?
On its official website, seed.bytedance.com.
How do I get access to the Seedance 2.5 API?
Sign up for BytePlus ModelArk and put at least $30 toward the model, as account credits, an AI Savings Plan or Seedance resource packs. Once the model is switched on, each video is billed by the tokens it uses, from the pack first and then from your balance.
Is Seedance 2.5 the same as Dreamina Seedance 2.5?
Yes. ByteDance's Seed team calls the model Seedance 2.5, while BytePlus lists it in its API as Dreamina Seedance 2.5, model ID dreamina-seedance-2-5-260628. ByteDance's Dreamina app also offers it under that name.
Can I use a real person's face in a Seedance 2.5 video?
Not by uploading it. The API rejects reference images and videos that contain real human faces. BytePlus offers a library of preset digital characters instead, and a separate process for using a real person's likeness with their authorization.
How long can a Seedance 2.5 video be?
One generation runs from 4 to 30 seconds, or you can let the model pick the length. You can then extend a clip over more rounds while keeping the same characters and setting, which ByteDance says makes videos of several minutes possible.
Our verdict
Seedance 2.5, ByteDance's newest video model, is the pick when one continuous 30-second take or a scene built from many references matters more than price.
It doubles Seedance 2.0's clip length, takes 50 reference files and edits by prompt, and early users say long takes stay consistent enough to cut re-rolls.
- Skip it for
- 4K delivery, real-face references or tight budgets, where Kling 3.0, Veo 3.1 or Wan 3.0 fit better.
- Check first
- Price a test job that includes a video input, since a 30-second source clip can push a 5-second 720p result to about $4.84.
Seedance 2.0
Seedance 2.0 mini
Gemini Omni
FLUX 3 Max
Kling 4.0
Grok Imagine Video 1.5