Mirelo SFX 2.0 Video-to-Audio: a sound file for a silent clip
Mirelo SFX 2.0 Video-to-Audio watches your clip and returns synchronized sound effects as a separate audio file. Settings, prompts and what to check first.
You have a clip. It looks right, and it is silent. Maybe it came out of a video model, maybe from your phone, maybe from an old render. Mirelo SFX 2.0 Video-to-Audio watches that clip and writes sound effects to fit what it sees: the door that closes on frame 40, the steps on gravel, the wind behind the hill. What you get back is audio only, a separate file you drop onto your own timeline.
Who is it for?
Editors and makers who already have a video tool they like and only miss the sound. Since the result is a file of its own, you keep full control: lay it under the picture, cut it, raise one hit, mute a part. If you would rather get a finished video with the new soundtrack already in it, there is a sister model for that, Mirelo SFX 2.0 Video-to-Video. The two take almost the same settings; the difference is what comes back.
The short version of how it works
- Upload the video. It must be at least one second long; it is the only required input.
- Optionally describe the sounds you want in the prompt, up to 2000 characters. Leave it empty and the model decides from the picture alone.
- Set the duration. The slider goes from 1 to 600 seconds, starts at 10, and is capped by the length of your video.
- Pick a format under advanced: WAV (default), MP3, AAC or FLAC.
Do you need a prompt at all?
Try without one first. The model reads the footage, and for a clear action (a glass set down, a car passing) it will find the sound on its own. The prompt is for the things the picture cannot say: what the room sounds like, whether the street is busy or empty, whether the mood is calm or tense.
A quiet kitchen early in the morning: soft clink of a spoon in a ceramic cup, the low hum of a refrigerator, a distant bird outside the window. Warm, close, no music.Heavy, metallic footsteps on a steel floor, a hiss of steam from a pipe, a deep mechanical hum filling the room. Large industrial space, echoing.The two advanced switches that save a take
- Negative prompt: sounds to avoid, such as music, voices or traffic. If you plan to add your own score, 'music' here is almost always right. The negative prompts guide explains the idea.
- Preserve speech: on by default. If your clip already has someone talking, the voice is kept and the new effects are built around it. Turn it off only if the speech is a leftover you want replaced.
Which format?
WAV for editing, since it is uncompressed. FLAC for the same quality in a smaller file. MP3 or AAC for a quick preview or for the web; for those two you may set a bitrate from 32 to 320 kbps, and left empty the encoding is automatic.
Where it falls short
It writes sound effects and ambience. It does not compose a song and it is not a voice actor, so for music or a spoken line use another audio model. Sound is also only as good as the picture is readable: a dark, blurry or very busy clip gives the model less to catch, and you will hear it guess. And the length is limited by your video, since the result cannot be longer than the clip you upload. If you would rather start from words alone, Mirelo SFX 2.0 Text-to-Audio makes sound without any video, and every audio model is on the models page.
Questions
What do I get back from Mirelo SFX 2.0 Video-to-Audio?
A separate audio file with the generated sound effects, in WAV, MP3, AAC or FLAC. The video itself is not returned.
How long can the video be?
At least one second. The sound duration slider runs from 1 to 600 seconds and is capped by the length of your video.
Will it keep the speech that is already in my clip?
Yes, by default. The preserve speech switch under advanced settings keeps the original voice while new effects are generated around it.
