zsxkib/mmaudio

Official View on Replicate →

Add sound to video using the MMAudio V2 model. An advanced AI model that synthesizes high-quality audio from video content, enabling seamless video-to-audio transformation.

Capabilities

Reference Images Negative Prompt Seed

Cost

Community model (estimated from hardware time)

Input Parameters

Name	Type	Description	Default	Constraints
`cfg_strength`	number	Guidance strength (CFG)	`4.5`	min: 1
`duration`	number	Duration of output in seconds	`8`	min: 1
`image`	string (uri)	Optional image file for image-to-audio generation (experimental)	`—`	—
`negative_prompt`	string	Negative prompt to avoid certain sounds	`"music"`	—
`num_steps`	integer	Number of inference steps	`25`	—
`prompt`	string	Text prompt for generated audio	`""`	—
`seed`	integer	Random seed. Use -1 or leave blank to randomize the seed	`—`	min: -1
`video`	string (uri)	Optional video file for video-to-audio generation	`—`	—

cfg_strength number

Guidance strength (CFG)

Default: 4.5 min: 1

duration number

Duration of output in seconds

Default: 8 min: 1

image string

Optional image file for image-to-audio generation (experimental)

negative_prompt string

Negative prompt to avoid certain sounds

Default: "music"

num_steps integer

Number of inference steps

Default: 25

prompt string

Text prompt for generated audio

Default: ""

seed integer

Random seed. Use -1 or leave blank to randomize the seed

min: -1

video string

Optional video file for video-to-audio generation

Version: 62871fb59889 Updated: 2/26/2026 4.8M runs