Music and effects from a prompt.
Describe a bed or a hit and Diffcut makes it on your Mac with ACE-Step. Or drop in a boom, whoosh or riser, synthesised on the spot.
- ~6 s
- to make a 30-second music bed on an M5
- 1–240 s
- any length, levelled and ready to place at the playhead
- 12
- synthesised effects that need no model at all
Say what you hear.
Genre, instruments, mood and tempo, in plain words. Diffcut ships with 16 examples; here are six of their prompts, exactly as written.
- Lo-fi study
lo-fi hip hop, dusty drums, warm Rhodes piano, vinyl crackle, mellow, 80 BPM, instrumental - Epic orchestral
cinematic orchestral, soaring strings, French horns, timpani, heroic, 100 BPM, instrumental - Calm corporate
calm corporate background, soft piano, light plucks, airy pads, optimistic, 100 BPM, instrumental - Cinematic boom
cinematic boom, deep sub impact, trailer hit, long reverb tail - Tension riser
riser, rising synth swell, white noise sweep, tension build - Fast whoosh
fast whoosh, air swoosh, quick pass-by transition
“Use prompt” on any example fills in its prompt, length and seed, so Generate makes a similar take. Then change what you like.
Twelve effects, no model.
For the hits every edit needs, Diffcut synthesises the sound itself: instant, and the same every time you ask.
- Riser
- Downlifter
- Whoosh
- Impact
- Boom
- Swell
- Pop
- Click
- Tick
- Swipe
- Ding
- Error
Made on your Mac.
The model runs on the GPU through MLX. Your prompts and the music they make stay on the machine.
- ACE-Step 1.5The turbo model, MIT licensed, in Diffcut’s own Swift code.
- LevelledMusic at −18 LUFS and effects at −16 LUFS, peaks at −1 dBTP.
- Ready to placeEach piece is ready to drop in at the playhead.
Questions
Which model makes the music?
ACE-Step 1.5 (the turbo model, MIT licence), running on your Mac’s GPU through MLX in Diffcut’s own Swift code.
How long can a piece be?
Anything from 1 to 240 seconds. It arrives levelled (music at −18 LUFS, effects at −16 LUFS) and ready to place at the playhead.
Can it write songs with lyrics?
Not from Diffcut today. The prompt describes instrumental beds and effects; there is no lyrics field in the app.
Do the synthesised effects use AI?
No. The 12 synthesised effects are generated by code on the spot, so they need no model and no download.
More of Diffcut
Make the cut.
Diffcut is free, open source and early: the first Mac build isn’t out yet. Leave your email and we’ll write once, when it’s ready.