Skip to main content
LaoZhang API sells four video model families: MiniMax H3, Muse, Seedance, and Wan. All of them run as async jobs: submit, poll the status, then download the MP4. For short clips around 720p, start with MiniMax H3 or Muse: they cost the least per clip and work with a token from the regular groups.
minimax-h3-en-t2v

MiniMax H3

$0.05 per second at 768P, 37.5% below MiniMax’s official price, so a 5-second clip is $0.25. Supports 2K, first and last frames, and reference images, video, and audio.
muse-video-en-t2v

Muse

Meta’s video model, billed per video: $0.30 for 5–10 seconds ($0.03 per second at 10 seconds), $0.55 for 20, $0.80 for 30. Uses the OpenAI-style /v1/videos endpoint and can extend an existing clip.
The card images are frames from each guide’s text-to-video example. What the two models share:
  • Every clip comes with music and sound effects.
  • Call them with a token in the default, vip, or svip group set to Usage first; one token works for both.
  • Failed jobs are refunded, and video links stay valid for 7 days.

​
What one clip costs

Cost per clip at 720p or higher with no reference video, based on public prices as of October 31, 2026 (USD):
  • MiniMax’s official H3 prices are $0.08/s at 768P and $0.13/s at 2K, from MiniMax pay-as-you-go pricing (checked October 24, 2026).
  • Seedance bills by tokens. The figures are estimates for 16:9 with no input video; the job’s reported usage decides the final charge.
  • With a reference video, MiniMax H3, Seedance, and Wan also bill the reference video’s length. See the pricing section of each model’s guide.
  • Longer clips: Muse has 20-second ($0.55) and 30-second ($0.80) models. H3 goes up to 15 seconds ($0.75 at 768P).
  • Check what each job actually cost in the call log. For current unit prices, see Models and pricing and each model’s guide.

​
Pick a model by task

  • Short clips around 720p with sound: for about 5 seconds use MiniMax H3 at 768P. For 8–10 seconds use Muse, which costs the same at 10 seconds.
  • A flat price that is easy to budget at scale: Muse. The price per video doesn’t change with duration tier, aspect ratio, or input images.
  • One continuous 20- or 30-second shot, or extending an existing clip: Muse’s muse-video-20s or muse-video-30s.
  • 2K output, or reference video and audio to steer motion and sound: MiniMax H3.
  • Existing OpenAI /v1/videos code, for example a former Sora 2 integration: Muse takes the same request shape. Change the model ID and duration; see the FAQ.
  • 1080p output, or footage of real people or virtual humans: Seedance 2.0 / 2.5.
  • Restyling or editing an existing video: video editing in Wan 2.7 / 3.0.

​
Endpoints and tokens

The endpoints are not interchangeable: MiniMax H3 only accepts /hailuo/v2, and Wan doesn’t use /v1/videos. Usage first is the simplest billing mode for every video model:
  • MiniMax H3, Wan, and Seedance also work with Usage-based tokens.
  • Muse also works with Per-call tokens.
You pick the group when you create a token in Token management; it never goes into the request’s model. See billing and token groups.

​
FAQ

​
Which video model is cheapest?

At 720p or higher, MiniMax H3 at 768P is cheapest for clips of about 5 seconds ($0.25), and Muse is cheapest for 8–10 seconds ($0.30). Both include audio and work with a default group token.

​
How do I migrate code that called Sora 2?

Sora 2 has been retired. Muse uses the same calls as Sora 2: POST /v1/videos to create a job and GET /v1/videos/{id} to poll it, and multipart first-frame uploads use the same input_reference field. Change three things:
  • Set model to muse-video.
  • Set seconds to 5, 8, or 10. For 20 or 30 seconds, switch to muse-video-20s or muse-video-30s.
  • Use a token in the default group with the Usage first billing mode.
Download the result from video_url. Field details are in the Muse video generation API guide.

​
Which model should I use for 1080p?

Seedance 2.0 / 2.5 or Wan. Both output 1080p and bill by seconds or tokens; see the price table above. MiniMax H3’s high-resolution tier is 2K, and Muse always outputs at the 720p tier.

​
Do the videos have sound?

MiniMax H3 and Muse always include an audio track, and there is no switch to turn it off. Wan 3.0 outputs sound by default; pass audio: false for a silent clip at the same price. Seedance uses generate_audio to control sound.