Seedance 2.0 Open Source? Access, Pricing and API (2026)

If you searched for Seedance 2.0 open source, here is where things stand in November 2026: ByteDance has not released the model weights, the training code or a licence that lets you host it yourself. Seedance 2.0 is a proprietary model that you reach through ByteDance’s own apps or through an API. The confusion comes from GitHub repos, Hugging Face pages and Reddit threads that use the name, so this post sorts out what is actually public, what you can run locally instead, what Seedance 2.0 costs per clip, and how to call it from code.

Is Seedance 2.0 open source?

No. To see why the question keeps coming up, it helps to separate three terms that often get mixed together.

• Open source means the source code is published under a licence that lets anyone use, study, modify and redistribute it. For an AI model, people usually also expect the training and inference code to be public.
• Open weights means the trained parameters (the large files that hold what the model learned) can be downloaded, so you can run the model on your own hardware. The licence may still be restrictive. Our coverage of Kimi K3.1 Pro’s custom licence and Baidu’s ERNIE licence that excludes some regions shows how much the fine print varies.
• Proprietary (closed) means the weights stay on the company’s servers. You send a request, the company runs the model, and you get a result back.

Seedance 2.0 sits firmly in the third group. There is no weights download, no official model card with files to pull, and no permission to self-host. The Seedance 2.0 open source question has the same answer for the earlier generation: our guide to Seedance 1.0 covers a family that was also only available as a hosted service.

ByteDance has published a technical write-up about Seedance 2.0 describing it as a multimodal audio-video generation model. A paper is not a release. It tells you how the model was built in broad strokes; it does not give you anything you can run.

What Seedance 2.0 actually is

Seedance 2.0 is ByteDance’s video generation model, released in April 2026. It turns a text prompt, an image, or a mix of reference media into a short video clip with sound. On Unicorn API it comes in two versions:

• Seedance 2.0: the higher-quality version, for text-to-video or image-to-video.
• Seedance 2.0 Fast: the quicker, cheaper version of the same model family, with the same options.

Both share the same capabilities and request fields:

• Output: 480p or 720p, 4 to 15 seconds, with native audio (sound generated together with the picture rather than added afterwards).
• Aspect ratios: 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, or adaptive.
• Inputs: a prompt, a start image (image), an optional end_image to control the final frame, and reference media via reference_image_urls, reference_video_urls and reference_audio_urls.
• Extras: with_audio to toggle sound, return_last_frame to get the final frame back (useful for chaining clips), and web_search.

That reference-media support is a big part of why people want Seedance 2.0 open source access in the first place: you can give it a product photo, a voice clip and a style reference in one request. It is also exactly the kind of capability that is hard to reproduce with today’s open-weight models.

What the Seedance 2.0 open source GitHub, Hugging Face and Reddit results really are

Search for Seedance 2.0 open source and you will find results that look like a release. They are not. Here is how to read each type.

GitHub repositories

The most common GitHub results are open-source applications built around Seedance, such as web app templates (often Next.js) for generating, editing and managing videos. The app code is open: you can fork it, restyle it and deploy it. But look inside and you will find an environment variable for an API key. The template sends prompts to a hosted Seedance endpoint and pays per clip. The model never touches your server.

These templates are genuinely useful if you want to ship a video product quickly. Just do not mistake “open-source Seedance app” for “open-source Seedance model”.

Hugging Face pages

On Hugging Face you will mostly see demo Spaces (small hosted web apps) that forward your request to a paid API, plus the occasional repository that claims to hold Seedance weights. ByteDance has not published official Seedance 2.0 weights there. Treat any “Seedance 2.0 weights” upload as fake at best and malware at worst: model files in some formats can execute code when loaded.

Reddit threads

Threads in communities that focus on local generation keep asking whether Seedance 2.0 open source weights will ever appear. The consensus matches what we describe here: it is closed, it is available through ByteDance’s apps and hosted platforms, and local users should look at open-weight models instead.

A quick check for any “open” Seedance project

1. Does it ask for an API key? Then the model runs elsewhere.
2. Is the publisher ByteDance’s verified organisation? If not, the weights are not official.
3. Does the README name a licence for the model, not just the code? A code licence like MIT on a wrapper says nothing about the model.
4. Are the “weights” a few megabytes? Real video model weights are many gigabytes. Small files are wrappers or fakes.

Can you run Seedance 2.0 locally?

No. Without the weights there is nothing to load, no matter how much GPU memory you have. If running on your own hardware is a hard requirement (for data privacy, offline use or unlimited generation), the realistic Seedance 2.0 open source alternative is an open-weight video model. The best-known options are:

• Wan from Alibaba, a family of open-weight text-to-video and image-to-video models with a large community of workflows.
• HunyuanVideo from Tencent, an open-weight video model with its own licence terms.
• LTX-Video from Lightricks, designed to be comparatively light to run.

Most of these are diffusion models: they start from random noise and remove it step by step, guided by your prompt, until a coherent video appears. They run in tools such as ComfyUI, a node-based interface for building generation pipelines.

Going local is a real trade-off, not a free upgrade:

	Seedance 2.0 via API	Open-weight model, self-hosted
Weights available	No	Yes
Hardware	None; runs on the provider’s GPUs	A capable GPU with a lot of VRAM, or rented cloud GPUs
Setup time	Minutes: get a key, send a request	Hours to days: drivers, workflows, model files, tuning
Cost model	Pay per clip	Pay for hardware and electricity; clips are then marginal cost
Native audio and multi-reference input	Yes	Varies by model; often needs extra tools
Data leaves your machine	Yes	No
Fine-tuning (training on your own footage)	No	Possible, licence permitting
Licence risk	Provider terms	Read the model licence: some restrict commercial use or regions

The rule of thumb: if you generate a few dozen clips a month and care about output quality, an API is cheaper than buying a GPU. If you generate thousands of clips, need full privacy, or want to fine-tune on your own footage, open weights start to make sense.

How much does Seedance 2.0 cost?

Because Seedance 2.0 is not open source, every clip costs money. On Unicorn API, price depends on duration and resolution, and is charged per output video. Credits cost $0.01 each on monthly plans (Starter $25, Creator $50, Studio $100, Pro $200 per month), and top-up credits cost $0.015 each and never expire. Full details are on the pricing page.

Clip	Seedance 2.0	Plan cost	Seedance 2.0 Fast	Plan cost
4 s, 480p	40 credits	$0.40	32 credits	$0.32
5 s, 480p	50 credits	$0.50	40 credits	$0.40
5 s, 720p	110 credits	$1.10	90 credits	$0.90
10 s, 480p	100 credits	$1.00	80 credits	$0.80
10 s, 720p	220 credits	$2.20	180 credits	$1.80
15 s, 720p	330 credits	$3.30	270 credits	$2.70

A few things fall out of this table:

• Fast is about a fifth cheaper at every setting (32 vs 40 credits at 4 s 480p, 90 vs 110 at 5 s 720p).
• 720p costs a little over twice as much as 480p for the same length. Draft at 480p, render finals at 720p.
• Cost scales linearly with length. A 15-second clip costs three times a 5-second one.

To put that in budget terms, the Starter plan’s 2,500 credits buys about 27 five-second 720p clips with Seedance 2.0 Fast, or 22 with standard Seedance 2.0, or 78 four-second 480p Fast drafts. On top-up credits, a 5-second 720p Fast clip costs $1.35.

You never have to guess: POST /v1/models/price returns the exact price of a request before you run it, which is handy when your app lets users pick duration and resolution. The request format is in the API documentation.

How to use Seedance 2.0 through the API

Since there is no Seedance 2.0 open source package to install, the practical route is an HTTP API. Video generation takes time, so it works as a job: you submit a request, get a job ID back, then poll until the video is ready.

1. Create a key on the API keys page and export it as UNICORN_API_KEY.
2. Optionally try prompts first in the video playground in your browser.
3. Send POST /v1/jobs with the model ID and an input object.
4. Poll GET /v1/jobs/{id} until status is "succeeded", then read outputs[0].url.

The simplest request (curl)

curl https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedance-2.0-fast",
    "input": {
      "prompt": "A slow drone shot over a rocky coastline at sunset, waves breaking"
    }
  }'

# Then check the job until it has succeeded:
curl https://api.unicornapi.net/v1/jobs/JOB_ID \
  -H "Authorization: Bearer $UNICORN_API_KEY"

A complete Python example

This script generates a vertical 5-second 720p clip with sound using Seedance 2.0 Fast, waits for it, and prints the video URL. At these settings it costs 90 credits.

import os
import time
import requests

API = "https://api.unicornapi.net"
HEADERS = {"Authorization": f"Bearer {os.environ['UNICORN_API_KEY']}"}


def generate_video(prompt, model="bytedance/seedance-2.0-fast", **options):
    body = {"model": model, "input": {"prompt": prompt, **options}}
    r = requests.post(f"{API}/v1/jobs", json=body, headers=HEADERS, timeout=30)
    r.raise_for_status()
    job = r.json()

    while job["status"] != "succeeded":
        if job["status"] == "failed":
            raise RuntimeError(f"Job failed: {job}")
        time.sleep(5)
        r = requests.get(f"{API}/v1/jobs/{job['id']}", headers=HEADERS, timeout=30)
        r.raise_for_status()
        job = r.json()

    return job["outputs"][0]["url"]


if __name__ == "__main__":
    url = generate_video(
        "A street food vendor flips noodles in a wok at a night market, "
        "steam rising, handheld camera, sizzling and crowd ambience",
        duration=5,
        resolution="720p",
        aspect_ratio="9:16",
        with_audio=True,
    )
    print("Video ready:", url)

Image-to-video and reference inputs

To animate a still, pass a start image URL. To control where the shot ends, add end_image. To switch to the higher-quality model, change the model ID:

url = generate_video(
    "The camera slowly orbits the sneaker as light sweeps across it",
    model="bytedance/seedance-2.0",
    image="https://example.com/sneaker-start.png",
    end_image="https://example.com/sneaker-end.png",
    duration=6,
    resolution="720p",
    aspect_ratio="1:1",
)

That request costs 132 credits. The same pattern works with reference_image_urls, reference_video_urls and reference_audio_urls when you want to steer style, motion or sound. If you work inside Cursor or Claude Code, you can also connect the API to your coding agent and ask it to generate clips directly.

Common mistakes with Seedance 2.0

• Downloading “leaked” weights. There is no legitimate Seedance 2.0 open source download. Files claiming otherwise waste your time and can compromise your machine.
• Assuming an open-source template means free generation. The template is free; every clip it makes still goes through a paid API. Budget for usage before you launch.
• Rendering every draft at 720p. Prompt iteration is where most clips get thrown away. Do that at 480p with Seedance 2.0 Fast (32 to 40 credits) and only render keepers at 720p.
• Expecting 1080p. Seedance 2.0 outputs 480p or 720p. If you need 1080p, pick a model that offers it, or upscale afterwards.
• Asking for 15 seconds by default. Long clips cost three times as much as 5-second ones and give the model more room to drift. Generate shorter shots and chain them with return_last_frame and image.
• Not handling the job lifecycle. Treat generation as asynchronous: poll with a pause between calls, stop on failure, and save the output URL to your own storage once the job succeeds.
• Skipping the licence on open-weight alternatives. “Open” does not always mean “use commercially anywhere”. Some model licences exclude regions or large companies.

Seedance 2.0 alternatives and which option to pick

If you came looking for Seedance 2.0 open source because you wanted it cheaper, a hosted alternative may suit you better than self-hosting. The same API key gives you other video models with different strengths:

Model	Best for	Example price
Seedance 2.0 Fast	Text or image to video with audio, cheaper drafts	90 credits, 5 s 720p
Grok Imagine Video 1.5	Very low cost per second, image-to-video only, up to 1080p	22.5 credits (rounded up to 23), 5 s 720p
Grok Imagine Video 1.0	Longer clips, 6 to 30 seconds, text or image input	27 credits, 6 s 720p
PixVerse V6 Max Image To Video	Per-second image-to-video with optional audio, up to 1080p	12 credits per second at 720p with audio
Kling 4.0 Keyframe To Video	Controlling a shot through first, last and middle keyframes	28.5 credits per second at 720p

Browse the full model catalog for the rest. To choose:

• You need Seedance’s look, multi-reference input and native audio: use Seedance 2.0 for finals and Seedance 2.0 Fast for drafts.
• You are animating stills on a tight budget: Grok Imagine Video 1.5 costs a fraction of Seedance per clip.
• You need clips longer than 15 seconds: Grok Imagine Video 1.0 goes up to 30.
• Your footage cannot leave your network, or you want to fine-tune: self-host an open-weight model such as Wan, HunyuanVideo or LTX-Video and accept the setup work.

The short version: Seedance 2.0 is closed, and that is unlikely to change on its own schedule just because people search for it. If you want to use it today, an API call is the fastest path, and the price endpoint tells you exactly what each clip will cost before you spend a credit.