Seedance 2.5 Explained: Specs, Pricing, How to Use It, and How It Differs from 2.0

The short answer

If you are searching for Seedance 2.5, you probably want to know four things: what it is, whether you can use it, how much it costs, and how to integrate it. Here they are:

• What it is: Seedance 2.5 is ByteDance's next-generation video generation model. It was previewed at the Volcano Engine FORCE conference on 2026-06-23, alongside an upgraded Seedance 2.0.
• Can you use it: Yes. Seedance 2.5 is live on hiapi as three models: seedance-2.5/text-to-video, seedance-2.5/image-to-video and seedance-2.5/reference-to-video.
• How much it costs: On hiapi, text-to-video and image-to-video cost $0.204 per second at 720p and $0.516 at 1080p. A 5-second 720p clip is $1.02. Reference-to-video starts at $0.1215 per second at 480p.
• How to use it: Send a task to the same POST /v1/tasks endpoint you use for other video models, then poll for the result. Try it on the Seedance 2.5 model page.

Below, we expand on each point, with a focus on what the upgrades mean for people who actually want to build with it.

What Seedance 2.5 is and why it matters

Seedance is ByteDance's family of video generation models. It is also the engine behind the video features in products like Jimeng (Dreamina) and Doubao. In other words, many people have already made videos with it without knowing its name.

In video generation, the top tier is currently a race between Seedance, Veo, Sora, and Kling. The competition has moved past "can it generate video?" to how long, how sharp, and how controllable the output is. The upgrades in Seedance 2.5 target all three.

For coverage of the June preview, see the IT Home report "ByteDance's AI video generation model Seedance 2.5 to launch in early July" (in Chinese): ithome.com/0/967/374.htm.

The 5 previewed upgrades, and what you get on hiapi

The launch event announced five upgrades at once. Some are available through the API on hiapi today; some belong to ByteDance's own Dreamina app.

1. 30-second single-shot generation (available)

Most comparable models top out at 15 to 20 seconds. Anything longer has to be generated in segments and stitched together. Seedance 2.5 generates 4 to 30 seconds in one task on hiapi. Reference-to-video can also pick the length itself if you set duration to -1.

The key benefit is no stitching, not just ten-plus extra seconds of footage. The worst part of segmented generation is the visual jumps and broken camera continuity at the seams. One continuous 30-second shot can cover a full product demo, a short-drama shot, or an opening sequence.

2. Native 4K (not on hiapi's API)

ByteDance previewed native 4K with 10-bit color. On hiapi, Seedance 2.5 outputs 720p or 1080p (reference-to-video also offers 480p). If you need 4K output today, Seedance 2.0 on hiapi supports a 4K resolution.

3. Up to 50 reference assets (available)

This is the most practical upgrade of the five. Reference-to-video on hiapi takes up to 10 reference videos, 30 reference images and 10 audio clips in one task, 50 assets in total. You refer to them in the prompt as @video1, @image1, @audio1 and so on. Image-to-video also accepts up to 30 reference images and 10 audio clips.

Controllability is the biggest pain point in video generation today. No matter how detailed your prompt is, faces and scene styles often drift. The more references you give, the better the model knows what you want. That makes character consistency and consistent brand style noticeably more reliable, which matters most for teams producing large volumes of content with the same characters and tone.

4. 3D clay-model camera control (Dreamina only)

You use a 3D clay model (an untextured gray model) to preview and control camera movement and composition. It lets you lay out a camera path in a simple 3D scene first, then render along that path. This is part of ByteDance's Dreamina app and isn't exposed through the API on hiapi. Through the API, you steer the camera with the prompt, first/last frames, or a reference video.

5. Local editing + native audio (audio available)

Local editing, changing one area of the frame after generation, is a Dreamina feature and isn't in the hiapi API. Native audio is: generate_audio is on by default and produces dialogue, effects and background sound in sync with the picture, so you skip separate voiceover and track syncing in post.

How it differs from Seedance 2.0

The table compares both models as they run on hiapi today.

hiapi prices for the default routes, checked October 5, 2026.

A few takeaways:

• 2.5 is longer and cheaper per second. At 720p and 1080p, Seedance 2.5 costs less per second than the default Seedance 2.0 route, and it can run twice as long in one shot.
• 2.0 still has 4K. If 4K output matters more than length, stay on 2.0.
• Switching is small, but not just a name change. 2.5 splits into three model IDs. Reference-to-video needs at least one reference video, and in image-to-video, first/last frames can't be combined with reference images or audio. The endpoint and the polling flow are the same.

How much it costs

A 5-second text-to-video clip at 720p costs $1.02, a 10-second clip $2.04, and a full 30-second shot $6.12. See hiapi pricing for the current numbers.

How to use it on hiapi

1. Try a prompt on the Seedance 2.5 model page before writing code.
2. Send a task. A minimal text-to-video request:

{
  "model": "seedance-2.5/text-to-video",
  "input": {
    "prompt": "A night train crossing a neon city in one continuous tracking shot, synchronized ambience and rail sounds",
    "resolution": "720p",
    "duration": 8,
    "aspect_ratio": "16:9",
    "generate_audio": true
  }
}

3. Poll the task, then download the video. The response gives you a task ID right away; query GET /v1/tasks/{taskId} until the status is success, then save the video from output. Output links expire, so copy the file to your own storage. One API key works for every model on the platform, so an existing Seedance 2.0 integration only needs the new model ID and input fields.

FAQ

How much does Seedance 2.5 cost? What is the price? On hiapi, text-to-video and image-to-video cost $0.204 per second at 720p and $0.516 at 1080p. Reference-to-video costs $0.1215, $0.2715 and $0.68 per second at 480p, 720p and 1080p.

How do I use and integrate Seedance 2.5? Send a task to POST /v1/tasks with one of the three model IDs (seedance-2.5/text-to-video, /image-to-video or /reference-to-video), then poll for the result. One API key works for all models.

How is Seedance 2.5 related to Jimeng (Dreamina) and Doubao? The Seedance family is the underlying model behind the video features in ByteDance products like Jimeng (Dreamina) and Doubao. When you generate video in those products, this model family is doing the work.

Does Seedance 2.5 support image-to-video? Yes. seedance-2.5/image-to-video takes a first frame, a first and last frame, or up to 30 reference images, with 720p or 1080p output and 4 to 30 seconds per clip.

How does Seedance 2.5 compare with Kling, Veo, and Sora? All four are in the top tier of video generation. Seedance 2.5 stands out on single-shot length (up to 30 seconds) and on how many references it accepts (up to 50 assets). For a price and spec comparison of two of the others, see Kling 3.0 vs Veo 3.1.

Is Seedance 2.5 available now? Yes. It is live on hiapi as text-to-video, image-to-video and reference-to-video models.

Key takeaways

• Seedance 2.5's upgrades are about scale: up to 30 seconds in one shot and up to 50 reference assets, with synchronized audio.
• On hiapi it runs at 720p or 1080p and costs $0.204 per second at 720p, less than Seedance 2.0's default route.
• Native 4K, clay-model camera control and local editing were previewed for ByteDance's own apps and aren't part of the hiapi API.
• Integration uses the same task API as 2.0; you switch to one of three model IDs and adjust the input fields.