Start creating

HKUST

MMAudio V2AI sound for video

Adds a soundtrack to a silent video: sound effects and ambience in time with what happens on screen, from a prompt. The result is the start of your video, as many seconds as you ask for (up to 30), with the new sound. On Leap, it sits beside 174 other models in one studio.

  • Video to audio
  • Up to 30 s

8 s on MMAudio V2: $0.048. Your words wait while you sign in.

  • No subscription
  • Every top model, one account
  • Pay per run, priced before it starts
A silent forge clip, given its hammer and sparks, a real run on Leap
A silent forge clip, given its hammer and sparks

AI sound for video

MMAudio V2 at a glance

MMAudio V2 is a model that adds sound to video from HKUST. It makes video from words or a video, 1 to 30 seconds long.

Made by
HKUST
Makes
Video
Starts from
Words or a video
Length
1 to 30 seconds
Prompt
Up to 4,000 characters
Price per run
$0.006 to $0.18
Run time
About 5 minutes
Updated
October 3, 2026
  1. 01Describe what you want to see
  2. 02See the exact price, then press go
  3. 03Keep it in your library, or take it further

Side by side

More video models on Leap

No prompt needed

Or start from a recipe

Recipes write the prompt for you: add a photo, pick a look, and get a finished result.

Every recipe

Why Leap

Why creators make it on Leap

  1. 01

    Every top model, side by side

    175 models from Black Forest Labs, Google, ByteDance, Kuaishou, Alibaba and more, side by side in one studio.

    See the models
  2. 02

    One click, or full control

    Character Studio, effects and recipes write the prompt for you. Or write your own for any model.

    See the tools
  3. 03

    Your work stays yours

    Everything you make stays in your library while your account is active, and you own what you make here.

    See what people make
  4. 04

    One balance for everything

    The studio, the API and your agents spend the same balance, and every run shows its price before it starts.

    How pricing works

Questions

MMAudio V2, answered

What is MMAudio V2?
MMAudio V2 is a model that adds sound to video from HKUST. Adds a soundtrack to a silent video: sound effects and ambience in time with what happens on screen, from a prompt. The result is the start of your video, as many seconds as you ask for (up to 30), with the new sound. On Leap it runs in one studio beside every other top model, with one account.
What do I give MMAudio V2?
MMAudio V2 starts from words or a video.
How long can the videos be?
1 to 30 seconds. You choose the length before you press go, and the price follows it.
How much does a run cost?
An 8-second clip costs $0.048 at the default settings. By length, a run costs $0.006 to $0.18. You see the exact price before you press go, and a run that fails costs nothing.
Is MMAudio V2 free to try?
No. There is no free credit: you add credit from $5, and your first top-up has no card fee. The cheapest run is $0.006 (a 1-second clip), so $5 covers 833 of those. You pay per run, in dollars, with no subscription.
Can I use what I make commercially?
Yes. You own what you make on Leap, so you can use it commercially, and we never train AI on it. The people and brands shown in a result can still have rights in it; the Terms say more.
Is there an API?
Yes. Leap's REST API runs it as hkust/mmaudio-v2, at the same price per run as the studio, and agents can call it through Leap's MCP server. The request is at the bottom of this page.
Do I need a HKUST account?
No. Leap sends your request to HKUST's model and gives you what it makes. One Leap account runs MMAudio V2 and every other model, in the studio, through the API and from your agents.

Your idea, MMAudio V2.

Every run shows its price before it starts, and one that fails costs nothing.

Make it with MMAudio V2

For developers

Run MMAudio V2 from code

API docs
Model ID
hkust/mmaudio-v2
Makes
Video
Pricing
Per run, quoted first: POST /v1/quotes

Inputs

What the model takes. Leave out anything with a default.

Promptprompt
Text, up to 4,000 characters, required
Videovideo
An image, requiredThe video to work on: an uploaded file, by ID. Up to 30 seconds.
Durationduration
1 to 30, required. Default: 8 s
Num Stepssettings.num_steps
4 to 50. Default: 25The number of steps to generate the audio for.
Cfg Strengthsettings.cfg_strength
0 to 20. Default: 4.5The strength of Classifier Free Guidance.
Mask Away Clipsettings.mask_away_clip
On or off. Default: offWhether to mask away the clip.
Negative promptnegative_prompt
Text, up to 2,000 charactersWhat the image should not contain.
Seedseed
0 to 4294967295The same seed and input give the same image.

The request

The same envelope for every model; only the model and its input change. With prefer: wait=60 the answer holds the result when the run finishes within a minute; a longer one is read again by its ID. A file input takes the ID that POST /v1/files returns for your upload. Agents can call it through the MCP server at app.tryleap.ai/api/mcp.

curl https://api.tryleap.ai/v1/generations \  -H "x-api-key: $LEAP_API_KEY" \  -H "content-type: application/json" \  -H "prefer: wait=60" \  -d '{    "model": "hkust/mmaudio-v2",    "input": {      "video": "file_...",      "prompt": "A paper boat drifting down a rainy street at dusk"    }  }' # Video often outlasts the wait and answers "queued" or "running".# Put its "id" in GENERATION_ID and read it again until it is done:# curl https://api.tryleap.ai/v1/generations/$GENERATION_ID -H "x-api-key: $LEAP_API_KEY"