Mirelo SFX 2.0 launches: swap video audio without regenerating footage, up to 10 min

ordax · x · 2026-10-14

Sound-foundation-model lab Mirelo launched SFX 2.0, a video-to-sound model whose pitch is changing the audio without regenerating the video:

Detects and preserves dialogue, strips unwanted non-dialogue audio like music, and generates effects, foley and ambience synced to on-screen action, fitting existing speech (including ducking) and staying coherent across cuts for videos up to 10 minutes (unlimited on Enterprise).
Targets two pain points: co-generated video+audio clips force full regeneration to change sound, and sound libraries/text-to-sound tools can't see the footage so every effect must be hand-placed.
Available now in Mirelo Playground, Editor, the API, Mirelo MCP, and on Fal.