Introduction

Mirelo SFX 2.0, the closed video-to-sound model from German startup Mirelo, now runs inside ComfyUI. A partner-node commit merged on October 21 (#17265) added two audio nodes, Mirelo Video to SFX and Mirelo Text to SFX, and both ship in ComfyUI v0.42.4. There are no open weights: the nodes call Mirelo through Comfy's partner API and bill per second of sound, at about $0.0133 per second, which works out to roughly $0.80 a minute.

We covered the model itself in Mirelo SFX 2.0 Keeps Dialogue, Replaces Music. This piece is only about what's new: the ComfyUI nodes, their settings, and the two official templates. ComfyUI hasn't posted an announcement yet, so everything below comes straight from the node code and the template library.

What Shipped

Both nodes sit under partner/audio/Mirelo and offer one model id, sfx-2.0.

Mirelo Video to SFX watches a clip and generates sound timed to the action on screen.

  • Video length: 1 to 600 seconds, so up to 10 minutes per run.
  • Keeps dialogue by default. The preserve_speech toggle is on out of the box and mixes the original dialogue under the new sound. Turn it off and you get sound effects only.
  • Two outputs. You get the new mix plus a separate speech track, which is the dialogue pulled out of the soundtrack (silent when there's no speech).
  • Optional dialogue input. You can plug in a clean dialogue recording to use instead of the video's own soundtrack.
  • Prompt is optional. The node's tooltip says the model describes the video itself and doesn't always follow the prompt, so treat text as a nudge. A negative prompt (a short comma-separated list) tells it which sounds to avoid.

Mirelo Text to SFX makes sound from a description alone.

  • Duration: 0.1 to 600 seconds, default 5.
  • Prompt is required, up to 2,000 characters, plus an optional negative prompt.
  • Output: the template describes it as a 48 kHz mono audio file.

Price and Templates

The price badge in the node code is $0.0132533115 per second for both nodes. That's about $0.07 for a 5-second effect, $0.80 for a minute, and just under $8 for a full 10-minute pass. You pay with Comfy credits, like other partner nodes, and you need to be signed in to your Comfy account.

The workflow template library added two Mirelo templates dated October 22:

  • Mirelo SFX 2.0: Video to SFX loads a short sample clip, adds picture-locked Foley, and saves three things: the SFX track, the speech track, and the video carrying the new mix.
  • Mirelo SFX 2.0: Text to SFX saves an MP3 and includes a prompt guide.

One catch: the template entries list ComfyUI 0.43.0 as their minimum version, which isn't out yet. The nodes themselves are already in v0.42.4, so if a template won't load, build the graph by hand: Load Video → Mirelo Video to SFX → Save Audio, or Mirelo Text to SFX → Save Audio.

Where to Run It Today

  • ComfyUI: update to v0.42.4 or newer and search the node list for "foley" or "sound effects".
  • Mirelo's own app and API, fal, Runway and Morphic, where SFX 2.0 was already live before the ComfyUI nodes arrived. Morphic
    , describing both video-to-sound and text-to-sound for clips up to 10 minutes.

Try It Yourself

The Text to SFX template's guide suggests one short line that names the sound, not a story: source, then material or surface, then space or distance, then intensity. The default prompt in the template is a good place to start:

Heavy rain on a tin roof with distant thunder

A few more in the same shape:

Slow footsteps on wet gravel, close mic, outdoors at night
Heavy wooden door creaks open, then latches, small stone room
Fast whoosh past the camera, then a deep metallic impact, large hall

For Video to SFX, leave the prompt empty on your first pass and let the model read the picture. If it misses something, add a short steer, like the template's own: Pulsing magenta energy hum from the crashed ship cockpit, soft electrical crackle, quiet rocky cavern wind, distant metal settling. Use the negative prompt for things you keep hearing and don't want, like music, crowd noise.

Conclusion

If you already make video in ComfyUI, this closes a gap: you can now go from a silent clip to a scored, Foley-timed cut without leaving the graph, and keep the dialogue on its own track for mixing. At about $0.80 a minute, it's priced for finishing passes rather than endless rerolls, so set a fixed seed once you like a result. It's hosted only, with nothing to download, and the polished templates will be easiest to use once ComfyUI 0.43.0 lands.

—Anabel ♡