Sound effects · MMAudio (open source)
MMAudio V2
An open-source model that turns a text description into sound effects. Pick it for the cheapest effects, up to 30 s.
Anonymized Low costWhat sets it apart
MMAudio comes from a research paper on joint video, audio and text training, and in Sub Rosa it turns a text description into sound effects of up to 30 s. Its checkpoints are CC BY-NC 4.0, so check licensing before commercial use.
- Research model, paper Dec 2024
- Checkpoints under CC BY-NC 4.0
- Default output 8 s, up to 30 s here
Recommended for: a door, rain, footsteps for a scene.
Strengths and limits
Where it shines
- Cheapest sound effects in the catalog per second
- Up to 30 s per sound
Think twice when
- Less refined than ElevenLabs on complex sounds
- 1 to 30 s
- WAV output
Use cases
Low-cost effects for a rough cut
Door slam in an empty hallway with a short echo.
Lowest price per second of the three.
Ambient texture
Wind through tall grass with distant birds, 20 seconds.
Up to 30 s per sound.
Against its rivals
- ElevenLabs Sound EffectsElevenLabs is more refined, with loops, at a higher price per second.
- Sonilo Sound EffectsSonilo is licensed for commercial use, but costs more.
Version history
Each version, dated, with what it changed over the one before.
- MMAudioEarlier, not in Sub Rosa
- Paper on video to audio synthesis trained jointly on video, audio and text
- MMAudio V2Latest
- Large 44 kHz v2 checkpoint listed in the repository; release date not published
The model in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
MMAudio V2mmaudio-v2-text-to-audio | Audio | Anonymized | $0.0041 per 10 seconds of sound |
Specifications
| Specification | MMAudio V2 |
|---|---|
| Weights | Open, CC BY-NC 4.0 (weights) |
| Accepts | text, video |
| Produces | audio |
| Price | $0.0041 per 10 seconds of sound |
| Privacy | Anonymized |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.