Seedance 2.0 AI Videos: Access, Pricing and API (2026)

Seedance 2.0 AI videos are short clips, 4 to 15 seconds long, generated by ByteDance’s Seedance 2.0 model from a text prompt, a still image, or a mix of image, video and audio references, with sound produced in the same pass. The model went viral in early 2026 because its clips looked closer to real camera footage than most of what came before. This post covers what the model can and cannot do, where you can use it, whether any of the “free” offers hold up, what each clip actually costs, and how to generate clips from code.

What are Seedance 2.0 AI videos?

Seedance is ByteDance’s family of video generation models. Like most modern video generators, it is a diffusion model: it starts from random noise and removes that noise step by step until a coherent sequence of frames emerges, guided by your prompt and any reference material you supply. Version 2.0 is the release that made the name widely known, and the one most of the viral clips you see online were made with.

Through an API you get two versions:

• Seedance 2.0: the full model, aimed at the highest quality output.
• Seedance 2.0 Fast: the quicker, cheaper version with the same inputs and options.

Both share the same specification, which is worth knowing precisely because many sites describe the model loosely:

• Inputs: text-to-video (a prompt only) or image-to-video (a start image, optionally an end image). You can also pass reference images, reference videos and reference audio by URL.
• Length: any whole number of seconds from 4 to 15.
• Resolution: 480p or 720p.
• Aspect ratios: 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, or adaptive (which follows the shape of your input image).
• Audio: native audio, meaning the model generates sound alongside the picture instead of you adding a soundtrack later.

So when people talk about Seedance 2.0 AI videos, they mean short, single-scene clips with sound. A “movie scene” you see on YouTube is almost always several clips edited together, with the creator choosing the best take of each shot. That editing step matters as much as the model, and we come back to it below.

You may also see Seedance 2.5 mentioned on Reddit and in demo videos. This post sticks to 2.0, which is what you can call through Unicorn API today.

Is Seedance 2.0 free? Official website, apps and “free unlimited” claims

Search results for Seedance 2.0 AI videos are full of apps and sites promising free or even “free unlimited” generation. Here is how to read those offers.

Who actually runs the model

ByteDance builds Seedance and offers it through its own products and its cloud platform. Most websites with “seedance” in the domain are not ByteDance: they are third-party wrappers that buy access to the model and resell it with their own interface, credit system and terms. The same is true of many mobile apps with the model’s name in the title. That is not automatically bad, but check the company behind the site, what happens to your uploads, and whether paid output comes with a watermark.

Why “free unlimited” does not exist

Every second of video takes serious GPU time to generate. No one can give that away without limit. In practice, “free” means one or more of these:

• A handful of trial credits on sign-up, enough for a few clips.
• Low resolution, short duration or a visible watermark on free output.
• A slow queue that free users sit in behind paying users.
• A daily allowance that resets, often small.

A free trial is a perfectly good way to see whether Seedance 2.0 AI videos suit your project. It is not a way to produce a steady stream of content. We cover the same pattern in more detail in our look at free AI video generators from text and the free options for Kling 3.0.

What it costs when you pay

On Unicorn API, the cheapest Seedance 2.0 clip, 4 seconds at 480p on the Fast version, costs 32 credits. Monthly plans give credits at $0.01 each (Starter $25, Creator $50, Studio $100, Pro $200 per month), so that clip costs about $0.32. If you would rather not subscribe, top-up credits cost $0.015 each and never expire, which puts the same clip at about $0.48. You can check the exact price of any request before running it, so there are no surprises. Full details are on the pricing page.

Seedance 2.0 vs Seedance 2.0 Fast: price per clip

Both versions are priced per clip, by duration and resolution. The pattern is simple: price scales linearly with seconds, and 720p costs a little over twice as much as 480p.

Clip	Seedance 2.0 Fast	Seedance 2.0
4 s, 480p	32 credits	40 credits
4 s, 720p	72 credits	88 credits
5 s, 480p	40 credits	50 credits
5 s, 720p	90 credits	110 credits
8 s, 480p	64 credits	80 credits
8 s, 720p	144 credits	176 credits
10 s, 720p	180 credits	220 credits
15 s, 480p	120 credits	150 credits
15 s, 720p	270 credits	330 credits

Put per second, Fast costs 8 credits at 480p and 18 at 720p, while the full model costs 10 and 22. Fast is about 20% cheaper at 480p and about 18% cheaper at 720p.

What a monthly budget buys

The Starter plan’s $25 gives 2,500 credits. That covers about 78 four-second 480p drafts on Fast, or about 22 five-second 720p finals on the full model, or a mix of both. A 15-second 720p clip on Seedance 2.0 is the most expensive single request at 330 credits, about $3.30 on a plan.

Which one to use

The sensible workflow for Seedance 2.0 AI videos is to draft cheap and finish expensive:

1. Iterate on your prompt with Fast at 480p and a short duration. Most of your attempts will be throwaways, so make them cheap.
2. Once a prompt reliably gives you the shot you want, render it on the full Seedance 2.0 at 720p.
3. Only extend duration when the short version works. A 15-second clip that goes wrong in second 3 costs as much as one that goes right.

Many people find Fast good enough for social posts at 9:16 and keep the full model for hero shots.

How to make Seedance 2.0 AI videos with the API

If you want to try before writing code, the video playground runs both versions in the browser. For anything repeatable, use the API. Video generation takes time, so it works as a job: you submit a request, get a job id back, and poll until the video is ready.

Step 1: get a key

Create an API key on the keys page and export it in your shell as UNICORN_API_KEY. The same key works for every model in the catalog, so you can switch between Seedance and other video models without new accounts.

Step 2: text-to-video in Python

This script checks the price, submits the job, polls until it finishes and prints the video URL.

import os
import time
import requests

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

body = {
    "model": "bytedance/seedance-2.0-fast",
    "input": {
        "prompt": (
            "A slow drone shot over a rocky coastline at sunset, "
            "waves breaking against dark rocks, seagulls calling, "
            "warm golden light, gentle wind noise"
        ),
        "duration": 5,
        "resolution": "480p",
        "aspect_ratio": "16:9",
        "with_audio": True,
    },
}

# 1. Check the exact price before spending credits
price = requests.post(f"{API}/v1/models/price", json=body, headers=HEADERS)
price.raise_for_status()
print("Price:", price.json())

# 2. Submit the job
resp = requests.post(f"{API}/v1/jobs", json=body, headers=HEADERS)
resp.raise_for_status()
job_id = resp.json()["id"]
print("Job:", job_id)

# 3. Poll until the video is ready
while True:
    r = requests.get(f"{API}/v1/jobs/{job_id}", headers=HEADERS)
    r.raise_for_status()
    job = r.json()
    if job["status"] == "succeeded":
        print("Video:", job["outputs"][0]["url"])
        break
    if job["status"] == "failed":
        raise RuntimeError(f"Job failed: {job}")
    time.sleep(5)

The output URL points to the finished MP4. Download it and store it yourself if you need it long term.

Step 3: image-to-video with curl

Image-to-video gives you far more control, because the model starts from a frame you chose instead of inventing one. Pass the start frame in image and, if you want the shot to land somewhere specific, a final frame in end_image. Setting aspect_ratio to adaptive keeps the shape of your input image.

curl -X POST https://api.unicornapi.net/v1/jobs \
  -H "Authorization: Bearer $UNICORN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedance-2.0",
    "input": {
      "prompt": "The woman turns toward the camera and smiles, hair moving in a light breeze, cafe chatter in the background",
      "image": "https://example.com/start-frame.jpg",
      "duration": 6,
      "resolution": "720p",
      "aspect_ratio": "adaptive",
      "with_audio": true,
      "return_last_frame": true
    }
  }'

# Then poll with the id from the response:
curl https://api.unicornapi.net/v1/jobs/JOB_ID \
  -H "Authorization: Bearer $UNICORN_API_KEY"

That 6-second 720p clip on the full model costs 132 credits. The return_last_frame option asks for the clip’s final frame back, which is the key to building longer sequences: feed that frame in as the image of the next request and the shots connect. The API documentation lists every field, including the reference URL inputs and web_search. If you work in Cursor or Claude Code, the MCP setup lets your coding agent call the same models directly.

Prompting tips for better Seedance 2.0 clips

The difference between forgettable and impressive Seedance 2.0 AI videos is mostly in the prompt and the input image. A few habits help a lot.

Write like a shot list, not a story

Each clip is one shot. Describe one subject, one action and one camera movement. “A chef plates a dessert, close-up, slow push in, soft kitchen light” works. “A chef cooks a three-course meal for her family and they laugh together” asks for a whole scene in one shot, and the model will compress or skip parts of it.

Name the camera

Terms like drone shot, tracking shot, handheld, static wide shot, push in, pull back and low angle give the model a clear motion target. Without them, you get whatever motion it guesses.

Describe the sound

Because audio is generated with the picture, the prompt should say what you want to hear: waves, footsteps on gravel, rain on a window, a crowd murmuring. If you need a specific sound or voice character, pass a file in reference_audio_urls. If you plan to add music in your editor, set with_audio to false so the generated track does not fight yours.

Use references for consistency

Keeping a character or product looking the same across shots is the hardest part of AI video. Generate a strong still first, then use it as the start image or pass it in reference_image_urls for every shot. The same character from the same reference holds together far better than the same character described in words each time.

How Seedance 2.0 compares with other video models

Seedance 2.0 is not the only option, and for some jobs it is not the best value. Here is how it sits against other video models on the same API, priced for a 5-second 720p clip where the model supports it.

Model	Input	Resolutions	Length	5 s at 720p
Seedance 2.0	Text or image, native audio	480p, 720p	4-15 s	110 credits
Seedance 2.0 Fast	Text or image, native audio	480p, 720p	4-15 s	90 credits
Grok Imagine Video 1.5	Image only	480p, 720p, 1080p	1-15 s	22.5 credits (rounded up to 23)
Grok Imagine Video 1.0	Text or image	480p, 720p, 1080p	6-30 s	27 credits for 6 s (minimum length)
PixVerse V6 Max Image To Video	Image, optional audio	360p to 1080p	Set per request	60 credits with audio, 44 without
Kling 4.0 Keyframe To Video	Keyframes (first, last, in between)	720p, 1080p	Set per request	about 143 credits

How to read this:

• Pick Seedance 2.0 when you want native audio generated with the picture, text-to-video without preparing an image first, and rich reference inputs (image, video and audio).
• Pick Grok Imagine Video 1.5 when cost per second is what matters and you already have a start frame. It is much cheaper, offers 1080p, and is strong on motion, but it only animates images and can be busy at peak times.
• Pick Grok Imagine Video 1.0 when you need a single clip longer than 15 seconds, up to 30.
• Pick Kling 4.0 Keyframe To Video when you need precise control over where the shot goes, with several keyframes along the way. It costs more per second.
• Pick PixVerse V6 Max when you need 1080p image-to-video with optional audio at a middle price.

Since one key covers all of them, a common setup is to use Seedance 2.0 for shots that need sound and text-only starts, and a cheaper image-to-video model for silent B-roll.

Common mistakes with Seedance 2.0 AI videos

Most wasted credits come from the same handful of errors.

1. Drafting at full price. Running every experiment at 720p on the full model costs nearly three times as much per second as Fast at 480p. Find the shot first, then upgrade.
2. Going long too early. A 15-second clip has more chances to drift: hands change shape, faces shift, objects appear. Short clips joined in an editor usually look better than one long take.
3. Expecting 1080p or 2K. Some marketing pages describe higher resolutions. Through the API the outputs are 480p or 720p. If you need sharper footage, use an upscaler afterwards or pick a model that renders 1080p natively.
4. Cramming a story into one prompt. One shot per request. Plan a shot list and generate each shot separately.
5. Ignoring the audio. Native audio is on by default for many workflows. If you never describe the sound, you get generic ambience. Either direct it or turn it off.
6. Not chaining frames. For multi-shot sequences, request the last frame of each clip and use it as the next start image. Skipping this is why many AI “movie scenes” jump between slightly different characters and lighting.
7. Trusting free unlimited offers with client work. Watermarks, queues and unclear terms cause problems when a deadline arrives. Know who runs the service and what rights you have to the output.
8. Not checking the price. The price endpoint returns the exact cost before you run anything. Build it into scripts that generate in batches so a typo in the duration does not multiply your bill.

Avoid these and Seedance 2.0 AI videos become a predictable, budgeted part of a production workflow instead of a slot machine. Start with a few Fast drafts in the playground, settle on prompts that work, then move them into code.