MMAudio v2 (video→audio) generates short sound effects and ambient audio from a text prompt or video reference. Footsteps, rain, machinery, alien creature roars — describe the sound and MMAudio v2 (video→audio) synthesizes it.

Typically 1 to 22 seconds depending on the engine. Loopable ambient tracks can be stretched with /audio/loop/.

Yes — video-to-audio engines like MMAudio v2 read frames from your video and synthesize a matching soundtrack (footsteps when feet move, splashes when water hits). Upload the silent video to /v1/audio/from-video/ or the page above.

WAV by default. MP3 is available in the format picker.

MMAudio v2 (video→audio) is a premium audio model. About ~1,000–5,000 tokens per clip. $1 = 750,000 tokens.

These models are tuned for sound effects + foley, not music. For melodic instrumental or vocal tracks see /music/ where MusicGen, ACE-Step, Stable Audio handle that case.

Yes — the prompt is descriptive (describe the sound, not lyrics), so any language works as long as the model understands it. English gives the most consistent results.

Yes — /batch/ accepts a list of prompts. Each clip lands in /account/?tab=history. The API is the most-flexible route for folder-tree preservation.

Yes — POST to /v1/audio/generate/ with model="MMAudio v2 (video→audio)" and your prompt (or video for v2a engines). /api/ has the full reference.

Same policy as the rest of Free.ai — self-hosted on our GPUs, premium with a DPA, uploads expire on the share-window schedule. We do not train on your inputs.

Yes — Free.ai grants commercial use of generated audio for game sound design, film foley, podcasts, ads.

5 to 30 seconds per clip. Video-to-audio takes longer (proportional to video length). Use the queue button on /audio/ to close the tab.