Mirelo SFX 2.0 Video to Audio
Mirelo SFX 2.0 watches your video and writes the sound for it. Send a video and get back soundtracks that are synchronized to the picture, as audio files you can drop into any editor. Speech in your footage is kept by default and the sound effects are generated around it, or you can ask for sound effects only.
Use cases: Post-production for AI-generated video | Foley for ads and social content | Replacing or filling in a soundtrack | Sound design drafts for editors
How it works
SFX 2.0 describes what happens in each moment of the video, then generates sound that follows it. Add a text prompt to steer the result. Mention sounds the picture does not show, such as a siren in the distance or wind in the trees, and they are worked into the soundtrack.
Every request returns one soundtrack per sample, so you can pick the take you like best. Outputs that would clip are scaled down to a peak of -1 dBFS.
| Parameter | What it does |
|---|
video_url | The video to add sound to. |
text_prompt | Optional. Steers the soundtrack. Name sounds that are off-screen or ambience to have them worked in. Up to 2,000 characters. |
negative_prompt | Optional. Sounds to keep out, as a short comma-separated list. Up to 1,000 characters. |
preserve_speech | true (default) keeps the speech in the video's own audio and generates the rest around it. false generates sound effects only. |
duration | Seconds of sound to generate, from 1 to 600. Defaults to 10 and is clamped to the video's length. Over 10 seconds, sound is generated in overlapping 10-second windows. |
num_samples | 1 to 4 variations per request. Defaults to 2. |
seed | Reproduce a result. -1 or unset is random. |
upload_audio_format | wav (default), mp3, aac or flac. |
upload_audio_bitrate_kbps | Optional bitrate for mp3 (32 to 320, in fixed steps) and aac (32 to 256) outputs. Unset uses a high-quality default. |
return_speech | Also return the speech separated from your video as its own audio file, in speech. Not billed. |
Output
audio holds one file per sample. speech is present when return_speech is true. model names the release that made the output.
Pricing
Billed per second of generated audio, per sample. Two 10-second samples are 20 billed seconds.
- Video to Video returns your video with the soundtrack already mixed in.
- Text to Audio generates sound effects from a text prompt alone.
mirelo-ai/sfx2.0/video-to-audio