Generate and Design the Sound Effects your Videos Deserve in Seconds🎵
Joined June 2024
- Tweets1.2K
- Following227
- Followers5.9K
- Likes1.4K
Pinned Tweet
AI Videos finally sound real.
Introducing Mirelo SFX 2.0.
It watches your video, cuts the music you never prompted, keeps the dialogue, and builds a full sound design synced to the action and consistent across any length clips.
You’ve made it look right. We’ll make it sound right.
Live today through our API, on @runwayml and @fal.
Other platforms coming soon. (Thread)
Excited to share that @nvidia is planning to invest in Mirelo!
As a foundation model company training our own audio models from scratch, we’ve already benefited immensely from @nvidia : their compute, the Inception program, joint events, and great exchanges with their researchers.
This planned investment will help us accelerate our mission to build the world’s best AI sound designer, expanding from sound effects to the best audio models for video.
Stay tuned!
Get started: app.runway.com
Our CEO @cjsimongabriel is dedicated to beat @elonmusk on The Cheeky Pint Benchmark
Shout out to @etnshow for having us!
If you haven't yet, watch our Thursday segment on our latest SFX 2.0 model release
Introducing The Cheeky Pint Benchmark.
We made our intern watch every episode of Cheeky Pint to bring you this breakdown of pints drank by guests to the nearest sixth.
Congratulations to @elonmusk who is currently in the lead with three full pints of Guinness.
How to STOP Seedance 2.5 from adding music that you never asked for.
Video models bleed unprompted music into your clips at random.
Mirelo SFX 2.0 keeps the visuals, removes the music and rebuilds the SFX based on the footage.
Don't take our word for it, just listen 👇
Ask your agent to create / edit a video with sounds and prepare to be disappointed.
Make SFX 2.0 your default audio layer to get sound that actually fits your video.
Today the agent ties sound to one video model and loses track of what is on screen.
SFX 2.0 builds sound from the video itself, so your agent can:
> Use any video model for the visuals
> Change the sound without regenerating the video
> Keep dialogue and drop music you did not prompt
> Get consistent sound across a stitched cut
> Advance control to edit or add new sounds
Your agents make your videos look right so make them also sound right.
AI Videos finally sound real.
Introducing Mirelo SFX 2.0.
It watches your video, cuts the music you never prompted, keeps the dialogue, and builds a full sound design synced to the action and consistent across any length clips.
You’ve made it look right. We’ll make it sound right.
Live today through our API, on @runwayml and @fal.
Other platforms coming soon. (Thread)
Your agent can call Mirelo SFX 2.0 at any stage of making a video.
Here is what it handles at each one:
1. While generating
Sound on every variant from any video model, so users can judge pacing and impact before picking one
2. On a clip with native audio
Unprompted music removed and dialogue kept, leaving room for newly generated score with Mirelo
3. After stitching
One pass over the full edit, so clips from different models sound consistent across every cut
4. Any time after delivery
New sound on the same video, without regenerating all visuals
Plug it into your agent through the Mirelo API or MCP 👇
SFX 2.0 can now be your default sound layer for anything you create on @fal
Mirelo retweeted
Mirelo SFX 2.0 is available today in the Mirelo Playground, Mirelo Editor, the API and Mirelo MCP, and on Fal. The full breakdown, with the benchmark method and what the API can do, is on our blog: mirelo.ai/blog/mirelo-sfx-2-…
Mirelo retweeted
We tested Mirelo SFX 2.0 in blind listening tests against other video-to-sound models and against the native audio of today's video models.
Listeners strongly preferred SFX 2.0 over other video-to-sound models, choosing it over Sony's Woosh in 81% of cases. On cinematic AI clips, it was preferred on average over the native audio of every video model we tested, including Seedance 2.5, Veo 3.1 Fast, MiniMax H3 Max, Flux 3 and Kling.
For each comparison, 25 raters judged 30 videos on audio quality and how well the sound matched what happens on screen.
Mirelo retweeted
4. Syncing
Mirelo SFX 2.0 builds the sound design from the video itself.
Text-to-sound tools work from a prompt and never see the footage, so every effect has to be found, trimmed and lined up by hand.
SFX 2.0 uses the video as its main input, so sound effects, foley and ambience follow the action and timing on screen and fit the setting, and a scene in a park doesn't sound like one in a cathedral.
Mirelo retweeted
3. Consistency and long-form
Mirelo SFX 2.0 keeps the same thing sounding the same, from your first shot to your last.
Clips generated one at a time each sound like a separate generation, so once they are stitched together the same car sounds different from shot to shot and the ambience shifts between shots of the same place.
SFX 2.0 generates sound with continuity across cuts, for the whole edit in one pass, so you never have to split a finished video into short clips to process it. Mirelo Studio and the API take videos up to 10 minutes, and unlimited length is available through our Enterprise tier.
Mirelo retweeted
2. Speech preservation
Mirelo SFX 2.0 keeps the dialogue and rebuilds the sound around it.
When a video model gets the dialogue right but the sound around it wrong, the options used to be regenerating the clip and losing the take, or rebuilding the sound design by hand.
SFX 2.0 detects the speech, preserves it and regenerates new effects and ambience, so you still use the take you chose with the sound you wanted.
Mirelo retweeted
1. Music leakage
Mirelo SFX 2.0 removes music your video model added without being asked.
Video models often bleed music into a clip even when you never prompted for it. Re-prompting burns credits without fixing it, and muting the clip takes the dialogue with it.
SFX 2.0 removes the unwanted music and replaces it with effects and ambience that match the picture, so the only music in your video is the music you add yourself.
We tested Mirelo SFX 2.0 in blind listening tests against other video-to-sound models and against the native audio of today's video models.
Listeners strongly preferred SFX 2.0 over other video-to-sound models, choosing it over Sony's Woosh in 81% of cases. On cinematic AI clips, it was preferred on average over the native audio of every video model we tested, including Seedance 2.5, Veo 3.1 Fast, MiniMax H3 Max, Flux 3 and Kling.
For each comparison, 25 raters judged 30 videos on audio quality and how well the sound matched what happens on screen.
Mirelo SFX 2.0 is available today in the Mirelo Playground, Mirelo Editor, the API and Mirelo MCP, and on Fal. The full breakdown, with the benchmark method and what the API can do, is on our blog: mirelo.ai/blog/mirelo-sfx-2-…