This post doesn’t work on mute. 🔇❌ Great sound design is invisible. When it’s right, nobody notices it. They just feel like they’re inside the scene. When it’s wrong, nothing else matters. Not the lighting, not the camera move, not the realism. That’s the quiet problem with AI video... right now. The picture keeps getting better, but the sound often breaks the illusion: inconsistent effects, music that leaks in, backgrounds that don’t fit the dialogue. Mirelo is a research lab working only on sound, and today they released SFX 2.0. It watches your footage and generates sound effects, foley and ambience matched to the action, the timing and the setting. It works on a short AI clip or a full edit, without touching the picture. Watch the before and notice what pulls you out. Then watch the after and notice that nothing does. Free to test. Start caring about the half of the video you hear.show more

Would be useful to know if it can leave intentional silence alone or if it always wants to fill the space.

The helicopter clip is a good stress test. Does it keep the rotor tone consistent when the camera cuts outside and back in?

The fact that it doesn’t touch the picture is the practical part. Most of the time the visuals are already locked by the time you realize the audio is wrong.

The “music that leaks in” thing is real. I’ve had models add cinematic scores to quiet dialogue scenes even when I never mentioned music.

I’ve got a few longer AI sequences sitting around. Wonder how well the ambience holds when you stitch multiple generations together.

The before/after framing makes sense. Once the sound matches, you stop noticing the cuts that were pulling you out.

Most of these tools still treat sound as an afterthought. This one seems to treat the footage as the prompt, which is the part that usually breaks.

I’ve been manually layering foley under AI clips and it still feels off by a few frames. Curious how much of the timing it’s actually reading from the picture.