MireloVideoToSFX - ComfyUI Built-in Node Documentation - ComfyUI

Generate sound effects that follow a video with Mirelo SFX. The clip is uploaded, the model follows the action on screen and returns a sound-effects mix, plus a separate speech stem with the dialogue pulled out of the soundtrack. Use it to add foley, ambience and impacts to silent or partly silent clips.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	Model to use. Selecting a model reveals the parameters specific to it: `video`, `audio`, `prompt`, `negative_prompt`, `preserve_speech` and `seed`.	DYNAMIC_COMBO	Yes	`"sfx-2.0"`
`video`	The video to generate sound for. Sound is generated for the whole clip, 1 to 600 seconds.	VIDEO	Yes	1 to 600 seconds
`audio`	Dialogue track to use instead of the video’s own soundtrack for `preserve_speech` and for the `speech` output. Leave it unconnected to use the sound embedded in the video.	AUDIO	No	-
`prompt`	What the sound should be like. May be left empty (default: empty) because the model describes the video itself, and it does not always follow the prompt. Maximum 2000 characters.	STRING	Yes	Up to 2000 characters
`negative_prompt`	Sounds to avoid, as a short comma-separated list. May be left empty (default: empty). Maximum 1000 characters.	STRING	Yes	Up to 1000 characters
`preserve_speech`	Keep the dialogue mixed under the generated sound. Off returns sound effects only. (default: True)	BOOLEAN	Yes	false true
`seed`	The seed for the request. Results are non-deterministic, so the same seed does not repeat a clip. (default: 42)	INT	Yes	0 to 4294967295

The node stops with an error when the clip is shorter than 1 second or longer than 600 seconds. A connected audio track sets the dialogue that preserve_speech keeps and that the speech output separates; without one, the dialogue comes from the video’s own soundtrack.

Outputs

Output Name	Description	Data Type
`audio`	The sound-effects mix for the clip, with the dialogue mixed under it when `preserve_speech` is on.	AUDIO
`speech`	The dialogue separated from the soundtrack. Silent, with the same sample rate as `audio`, when no speech is found or speech separation returns nothing.	AUDIO

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): fa09525e6c9b10412823e358652220850fbc7109d81cc207d31ffa8921956495