An AI music generator on the same canvas as the picture.

Three generators and an analyzer

Each one publishes real audio downstream — and a volume and speed control that only ever touch playback, never the bytes.

• music

  prompt or composition plan

  Describe the track, or build it section by section with per-section lyrics and style tags. Style tags can be wired in from a list node and apply across every section. Inpainting replaces a stretch of an existing track.

• voice

  text to speech

  Voice and model selects over whatever text is wired in. Point it at a text node you are editing or an llm node that wrote the script, and the read follows the copy.

• sfx

  text to sound

  Sound effects from a description, with trim handles on the waveform that really cut the clip before it is published — decoded, sliced and re-encoded, always from the original bytes.

• waveform

  analyzer

  Load or receive a clip, play it back, and mark its sections on the wave. Those marks leave as keyframe timings, which is how you cut picture to a track you already have.

AI audio generation — common questions

How does the AI music generator work?

Two ways. Describe the track and get it back in one shot, or build a composition plan section by section — each section with its own lyrics and style tags — so the structure of the song is something you decide rather than something you get.

Can I change part of a track without regenerating all of it?

Yes. Inpainting replaces a stretch of a track you already have while leaving the rest intact, so a chorus that nearly worked doesn't cost you the whole song.

Can the music be cut to the video?

Or the other way round. Wire a keyframe track into the music node and its marks become the section boundaries of the plan. Wire the music node's output back out and every section end arrives downstream as a marker for the video and render nodes to cut against.

What about voiceover?

The text-to-speech node takes a voice and a model and reads whatever is wired into it. Write the script in a text node or generate it with an llm node, and the voiceover re-renders when the script changes.

Can I trim a sound effect?

Yes, and the trim is real — the clip is decoded, sliced and re-encoded, so what goes downstream is what you trimmed, not the original with markers on it. Trims are always re-cut from the source bytes, so they never compound.

How is audio billed?

By size, not per click: speech per 250 characters and music per 15 seconds, with caps of 5,000 characters and 5 minutes so the worst case is knowable before you press the button. Controls say "up to N credits" for the same reason.