How to use Seedance

Before you start

You need an account with an active subscription and, if you plan to use image or reference modes, your source files ready. Seedancee shows a credit estimate before every generation, so you will not be surprised by the cost. There are no free trial credits: Starter, from $4.99 a week, covers Seedance 2.0 Mini, 2.0 Fast and 1.5 Pro at up to 720p and 10-second clips, and Pro adds Seedance 2.5 and 2.0, up to 1080p (4K on 2.0), each model's full length and reference videos. The pricing page lists the plans and the credit packs subscribers can add.

• Have a one-sentence idea of the shot: who or what, doing what, where, seen how.
• Collect references if you need consistency: a character sheet, a product photo, a short clip with the camera move you want.
• Decide the delivery format first (vertical for social, 16:9 for web), because aspect ratio is set before generation, not after.

The eight steps

1. 01

   Choose a mode

   Text to video invents everything from words. Image to video animates a frame you provide. Reference to video uses images, clips or audio to control identity, motion or sound. Edit and extend work on a clip you have already generated (Seedance 2.5 only). Start with text to video if you are exploring; switch to image or reference mode once you need a specific look to persist.

2. 02

   Choose a model

   Seedance 2.5 is the default and the right choice for longer clips (4–30s), heavy referencing (up to 30 images, 10 videos, 10 audio clips) and anything you plan to edit or extend. Seedance 2.0 is the choice when you need 4k output on a 4–15s clip. Seedance 2.0 Mini is the cheapest way to draft ideas at 480p, 720p. Seedance 1.5 Pro is a previous-generation option with a fixed-lens setting and no reference inputs. The model picker shows the same capability matrix we publish on the model pages.

3. 03

   Add references

   Upload a first frame for image to video, or add reference images, video and audio in reference mode. Label what each reference is for: the model needs to know whether an image is a character to keep, a style to copy, or a layout to follow. Keep reference video short and clean; the provider caps clip length and total duration, and a busy reference confuses the result. Reference audio should be a single voice or a single beat, not a mixed track.

4. 04

   Write the prompt

   Structure beats adjectives. Cover subject, action, environment, camera and lighting in that order, in plain sentences. State what should not happen ('no cuts', 'no text on screen') because the model will otherwise improvise. If you want speech, put the exact line in quotation marks and say who says it. Our prompt guide has the full framework and a library of copyable prompts.

5. 05

   Configure the shot

   Set duration, resolution, aspect ratio and audio. Shorter clips finish faster and cost less, so draft at 4–6 seconds and 720p, then re-run the keeper at full length and resolution. Audio is free on the Seedance 2.x models and doubles the rate on 1.5 Pro, so leave it on whenever the shot needs sound. Seedance 2.5 can also return the last frame, which is useful for chaining shots.

6. 06

   Generate

   Press generate. The job is submitted to the provider asynchronously; you can leave the page and the result lands in your library. Typical waits range from under a minute for short Mini clips to several minutes for long 2.5 clips at 1080p. If the provider fails the job, the reserved credits are released automatically.

7. 07

   Refine

   Watch the result at full size, then decide: re-run with a tighter prompt, change one setting at a time, or on Seedance 2.5 use edit to change part of the clip or extend to continue it (ByteDance's page says a clip can be extended up to twice). Keep a note of what you changed; the workspace stores every generation with its exact settings so you can compare.

8. 08

   Export

   Download the MP4 (or MOV where the model offers it) from the library. Outputs are copied to our storage as soon as they finish, so you are not dependent on the provider's expiring links. If you want to share a clip publicly, publish it from the library; published clips go through moderation before they are indexable.

A worked example

Here is a complete first project. It uses text to video on Seedance 2.5 so you can follow along without any assets.

A florist in a green apron arranges white tulips in a glass vase on a wooden counter. Morning light through a shop window on the left, soft shadows. Slow dolly in from a medium shot to a close-up of her hands. Quiet shop ambience, no music, no cuts.

Generate it once at these draft settings. Then change exactly one thing, for example 'slow dolly in' to 'static camera', and generate again. Comparing two clips that differ by one sentence teaches you more about the model than ten unrelated prompts.

Try the steps here

Below is the same generator that runs in the workspace, embedded as a lightweight demo and preloaded with the worked example. Switch modes and models to see which inputs each one asks for (steps 1 to 3), edit the prompt (step 4), then change duration, resolution and audio and watch the credit estimate update (step 5). Configuring is free and needs no account. Pressing Generate (step 6) hands the draft to the workspace, which asks you to sign in and choose a plan if needed, and checks your credits against the estimate before anything is submitted.

Choosing a mode: a quick decision table

Mistakes that waste credits

• Generating at 30 seconds and 1080p before the prompt is right. Draft short, then scale.
• Stacking six adjectives instead of one clear camera instruction.
• Uploading a reference video with cuts, captions or a watermark and expecting a clean result.
• Asking for a copyrighted character or a real person's likeness; it violates the content policy and may be blocked by provider moderation, which still costs time.
• Changing three settings at once, then not knowing which one fixed or broke the clip.

Using Seedance through the API

If you would rather script generations, the same steps map onto API calls: choose a model id, submit a job with the prompt, references and settings, then poll or receive a callback when the output is ready. Our provider enforces a rate limit of 20 requests per 10 seconds and keeps outputs for 14 days, so production integrations must copy results to their own storage. The Seedance API page documents the Seedancee API, currently in early access.