How to Use Seedance 2.0 by API: Models, Prices, First Video

Generate a Seedance 2.0 video from a text prompt, in one prompt. Paste this into your agent:

Set up https://monid.ai/SKILL.md, and then use Monid to generate a 5-second 720p, 16:9 video with Seedance 2.0 Mini of a small paper boat drifting across a puddle on a city sidewalk after rain, then download the file.

One prompt renders and saves a clip. 1 min setup · 3-8 hrs saved.

You want Seedance 2.0 inside your own workflow, not in a web editor: a request goes in, an MP4 comes out, and you know what it cost. There are four Seedance models to choose from, they bill per token rather than per clip, and the link to your video expires. This guide covers the first video, the choice between models, and the arithmetic behind the bill. It runs through Monid, the OpenRouter for agent tools: one key, one balance, every tool priced per call.

What is the easiest way to use Seedance 2.0 through an API?

Through Monid. Install it once, add a key, and your agent calls a Seedance endpoint with a prompt and gets back a video link, with no separate account to open for the model.

• Pay for what you run. Each render is billed on the tokens it actually uses, at $3.50 per million tokens on Seedance 2.0 Mini (Verified 2026-10-16). Buy balance as you need it: no monthly plan and no minimum top-up. Discovery and inspection are free; only a run bills.
• One install, every tool. The same key and balance call the video endpoints and every other tool on Monid. A pipeline that pulls data and then makes a video from it stays on one bill.
• Lives inside your agent. Install the CLI, add your key, run the prompt above. Your agent reads SKILL.md and learns the rest, so setup takes about a minute, and nothing runs between renders.

How do you generate your first Seedance 2.0 video?

Inspect the endpoint, then send one request with a text prompt and wait for the result. The text-to-video call takes a content array, a resolution, a duration from 4 to 15 seconds, a ratio, and three switches: generate_audio, watermark and priority.

For agents

Grab an API key at app.monid.ai, then paste this to your agent and hand it the key:

set up https://monid.ai/SKILL.md

It learns the whole discover, inspect, run workflow itself. More in the agent quickstart.

For humans

npm install -g @monid-ai/cli
monid keys add -k <your-api-key> -l main

Step 1. Read the schema and the price

What it does. Shows the body shape, the price matrix and the notes before you spend anything.

monid inspect -p bytedance -e /v1/video/seedance-2.0-mini

What comes back. The input schema, a rate for each resolution, the endpoint notes and the typical run time. The notes are the only place the token formula is written down. What it costs. Nothing. Inspect is free.

Step 2. Generate on Mini

monid run -p bytedance -e /v1/video/seedance-2.0-mini -w \
  -i '{"content":[{"type":"text","text":"A small paper boat drifting across a puddle on a city sidewalk after rain, low camera angle, soft morning light, gentle ripples, cinematic"}],"resolution":"720p","duration":5,"ratio":"16:9","generate_audio":false}'

-w waits inline. Generation takes tens of seconds to minutes: the median run time inspect reports for Mini is about 124 seconds. The completed run returns a video_url. We did not render this clip today, because a 5-second 720p render is about 108,000 tokens, which the cost section prices at roughly $0.378 (Verified 2026-10-16), so the fields above come from the endpoint listing and not from a run of ours.

Step 3. Download within 24 hours

curl -L "<video_url>" -o clip.mp4

The endpoint notes say the link expires in about 24 hours, so save the file as the first step after a run.

Which Seedance model should you use?

Start on Mini and move up when the clip tells you to. Monid lists four Seedance video endpoints from bytedance, all stable on 2026-10-16, and they share one request shape.

Three things follow from the listing. Mini's rate is exactly half of Seedance 2.0's at 720p, and the schema does not change, so the cheapest way to find out what a prompt does is to render it on Mini. Fast sits between them on price and shares Mini's median run time, so it suits drafts you want back sooner. Seedance 2.0 is the only one that offers 1080p and 4k, and 2.5 is the only one that goes past 15 seconds.

The listing describes Mini as the cheapest tier for bulk generation, Fast as a quicker tier for drafts and iteration, Seedance 2.0 as the highest-quality tier and 2.5 as the flagship. Whether the quality gap is worth the rate for your clips is a judgement you make by looking at one render from each, and the rule that saves money is to look at Mini's first.

For two models head to head, see MiniMax against Seedance for AI video.

How much does a Seedance 2.0 video cost?

A clip costs its token count times the rate, and the token count is set by resolution, aspect ratio and length. The endpoint notes give the formula: tokens are about width times height times 24 frames times seconds, divided by 1,024, which is about 21,600 tokens a second for 720p at 16:9.

Worked out from that formula and the rates above, a 5-second 16:9 clip comes to:

These are calculations from the listed rates (Verified 2026-10-16), not bills from runs of ours; the bill on a real run uses the actual tokens.

What this means for budgeting

Duration is linear. A 10-second clip has twice the tokens of a 5-second one, so on Mini a 15-second 720p clip is about 324,000 tokens, roughly $1.134 (Verified 2026-10-16). duration is your first cost control.

A lower rate does not mean a cheaper video. Seedance 2.0 lists 4k at $4.00 per million tokens and 720p at $7.00, but 4k has nine times the pixels, so about nine times the tokens. The endpoint note puts a 4k video at roughly five times the cost of a 720p one, and the table agrees: $3.888 against $0.756 (Verified 2026-10-16). Read the token count, not the rate.

Audio costs nothing extra. generate_audio defaults to true, and the schema says it has no price impact. Set it to false only when you will lay your own sound over the clip.

Reference videos have a floor. The notes say runs that use reference video have minimum-token floors, and the usage reflects them automatically. Do not estimate those runs with the plain formula.

Current rates sit on the video generation tools page and at monid.ai/tools.

Can you use image, video and audio references?

Yes. The content array takes four item types, and the prompt text can refer to them by number. This comes from the live schema on 2026-10-16.

• Text (text): the prompt, up to 6,000 characters. Refer to other items as @Image1, @Video1 or @Audio1, numbered per type in array order, and say what each supplies. Mark sounds with () for music, <> for effects, {} for dialogue and 【】 for subtitles.
• Images (image_url): first_frame pins the opening frame and is the default, last_frame pins the closing one and needs a first frame, and reference_image is soft guidance for subject or style. Seedance 2.0, Mini and Fast take up to 9 reference images; 2.5 takes up to 30.
• Video (video_url, role reference_video): up to 3 clips of 2 to 15 seconds, 15 seconds in total, on the 2.0 family; 2.5 takes up to 10 clips of 2 to 30 seconds.
• Audio (audio_url, role reference_audio): same counts as video. On the 2.0 family audio cannot be the only input, so include an image or a video. On 2.5 audio may be the only reference.

Image to video is a first-frame call:

monid run -p bytedance -e /v1/video/seedance-2.0-mini -w \
  -i '{"content":[{"type":"text","text":"The camera pushes in slowly as steam rises from the cup"},{"type":"image_url","image_url":{"url":"https://example.com/cup.jpg"},"role":"first_frame"}],"resolution":"720p","duration":5}'

The URL must be public https or an asset:// id. Inline base64 data URLs are not accepted. Reference images and videos that contain real human faces are rejected by the provider, so a product shot, a scene or an illustration works and a photo of a person does not.

What goes wrong when you generate Seedance videos?

Four things, all from the endpoint notes and schema read on 2026-10-16.

Why did my video link stop working?

Because a video_url expires about 24 hours after the run. Treat it as a delivery link, not a reference. A pipeline that stores the URL and not the file loses every clip older than a day, and the fix is one download step, which is also the first step in automating short-form AI video.

Why did a 4k render cost more than I expected?

Because the rate is lower and the count is higher. A rate column that falls from $7.70 at 1080p to $4.00 at 4k (Verified 2026-10-16) reads like a discount. It is a bigger bill, because a 5-second 4k clip is about 972,000 tokens against about 108,000 at 720p. Price from the token formula before you loop.

Why did my edit come back as a new video?

On Seedance 2.5, the model infers the task from your content roles and your wording. The notes say that without an explicit instruction such as "Video edit: remove everyone in @Video1 except the protagonist", a request with reference clips is read as plain reference-to-video, which succeeds, produces a new video and bills in full with no error. Editing and extension also need ratio set to adaptive, and editing needs duration set to auto.

Why did duration: "auto" reserve so much balance?

On 2.5, auto holds the price of a 30-second video up front and releases the unused part when the run settles on actual tokens. Pass an explicit duration when you want a tight hold on your balance.

Which endpoint should I use for which job?

Prices Verified 2026-10-16 with monid inspect. The first row is the one to render on first. The others are for a specific reason: 1080p or 4k, clips longer than 15 seconds, or editing and extending existing footage.

When is another route a better fit?

You want a visual editor. If you are trimming, captioning and exporting by hand, a video app that embeds the model is quicker than a request body. An API fits pipelines and agents, not one-off edits.

You need a specific real person on screen. Reference images with real faces are rejected by the provider, so a fixed presenter or actor needs a different class of tool.

You need the result in seconds. The median run time is about two minutes on Mini and three on Seedance 2.0, and the 95th percentile for 2.5 is over six. Anything that blocks a user on generation needs a queue and a callback, not a wait.

You cannot predict clip length. Billing follows tokens, so a loop over prompts with free-running durations has an open-ended bill. Fix the duration, estimate with the formula, then loop.

And the disclosure: this is Monid's blog and we resell all four endpoints. The advice here is to render on the cheapest one first, which is advice to spend less.

Conclusion

The quickest way to use Seedance 2.0 by API is a single request through Monid: a content array with your prompt, a resolution and a duration, a video_url back, billed on tokens from $3.50 per million on Mini (Verified 2026-10-16). Render on Mini first, then move to Seedance 2.0 for 1080p or 4k, or to 2.5 for longer clips and editing.

Two things matter more than the model you pick. The bill is tokens, and tokens come from pixels and seconds, so estimate with the formula before you loop and read the token count, not the rate. And save the file at once, because the link lasts about a day.

How we estimated the time saved: wiring an asynchronous video API into a project is roughly 2 to 5 hours (account and key, request builder, waiting for the result, saving the file), and a token-based cost estimator plus handling the reference rules is roughly 1 to 3 hours more. That is 3 to 8 hours. The 1 minute is the install, key and first call once you already have a Monid key.

Free next step: run monid inspect -p bytedance -e /v1/video/seedance-2.0-mini and read the price note, then work out your own clip's token count with the formula. Start at monid.ai.

FAQ

Published September 2026.