Mimic separates a track into stems, analyses the timbre of the sound you choose, and rebuilds it as a real, playable instrument on a MIDI keyboard. Stem separators do half the job. AI synthesizers do the other half. Mimic is the first plugin built to do both.
Mimic runs three engineering stages back-to-back. The web demo above mirrors them step-for-step — the shipping plugin runs them natively in your DAW with sub-millisecond latency.
Drop in a track, paste a streaming link, or capture audio from your DAW. Mimic's separation engine — built on Facebook's MIT-licensed Demucs Hybrid Transformer — splits the signal into clean drums, bass, synth/keys and vocal stems on your machine. No cloud upload required.
Pick the element you want — a bassline, a synth pad, a vocal "ahh". Mimic extracts a fingerprint of its spectral content (MFCC, chroma, centroid, bandwidth) and a custom neural model maps that fingerprint into our synthesizer's parameter space. It produces four candidate patches per query.
Load any candidate onto the engine and the source sound becomes a real, polyphonic instrument under your MIDI controller. 32 voices, multi-mode filters, full ADSR + modulation envelopes, 2 LFOs, complete effects chain — every parameter is automatable from your DAW.
Every detail that matters to a working producer or sound designer — and nothing that doesn't.
Three oscillators per voice, full waveform set, FM & AM modulation, sync — the engine isn't a sample player, it's a real subtractive + FM synthesizer.
Mimic doesn't lock you to one match. It generates four variations per query so you can pick the one that fits your track — or refine the best.
Source separation and timbral inference run locally on your machine. No cloud upload of your work-in-progress tracks. No subscription gating.
Every synth parameter is automatable from your DAW. CC1 modulation, CC11 expression, full pitch-bend response. Plays nicely with your existing rig.
Reverb, delay, chorus, distortion (tube, tape, digital), compression, parametric EQ — chosen because producers actually reach for them, not to pad a spec sheet.
Works in Ableton Live, Logic Pro, FL Studio, Reaper, Bitwig, Studio One, and any other VST3/AU host. Both macOS (Apple Silicon + Intel) and Windows from day one.
Mimic isn't aimed at everyone. It's built for the people who actually try to chase a specific sound and end up burning hours doing it the hard way.
You know the sound. You can hear it. You've been digging through your sample library for an hour and nothing's it.
Synthesizer presets are stale by their fifth user. Mimic's variations come from your reference — guaranteed unique starting points.
Your director sent a temp track. They want "something like that". You don't have time to hire the original musician — and you don't need to.
Stem separators do half the job. AI synths do the other half. Mimic is the first plugin built to do both in one playable instrument.
| Capability | Stem Separators AudioShake · LALAL · Demucs |
AI Sound Designers Synplant 2 · Riffusion |
Mimic This plugin |
|---|---|---|---|
| Separates a track into stems | Yes | No | Yes |
| Turns a sound into a playable patch | No | Yes | Yes |
| Works as a VST/AU plugin in your DAW | Mostly online | Yes | Yes |
| Real synth engine you can tweak | No | Limited | Full subtractive + FM |
| Runs on your machine (no cloud upload) | Some yes | Yes | Yes |
| One workflow from source → playable | No | No | Yes |
The current build above is a v0.1 web concept demo — designed to prove the user flow end-to-end before native plugin engineering begins. The full VST/AU build is planned across three production phases (synth engine, AI timbral matching, polish + release). The roadmap targets a beta in roughly six months from kickoff.
Two distinct AI stages. First, a stem-separation model (Demucs Hybrid Transformer) splits the source track into drums, bass, synth and vocals. Second, a custom timbral-inference model analyses the selected element and maps its spectral fingerprint to a candidate set of synthesizer parameters.
You're not getting an audio sample — you're getting a playable synthesizer patch tuned to match.
No. Both the separation model and the timbral inference run on your machine. Your work-in-progress audio never leaves your computer — important for film/game composers under NDA, and good practice for everyone else.
Mimic ships as a VST3 and Audio Unit plugin on macOS (Apple Silicon + Intel) and Windows from v1.0. That covers Ableton Live, Logic Pro, FL Studio, Reaper, Bitwig, Studio One, Cubase, and any other VST3 or AU host. Pro Tools AAX support is on the roadmap for v1.1+.
No tool can. Mimic is a starting-point generator, not a perfect cloner. Vocals can't be turned into a playable vocal instrument (that's a different research problem). Hyper-specific human performance characteristics — breath, micro-timing, exact articulation — are out of scope for a synth-engine target.
What Mimic does well: capturing the timbre, character and tone of a sound so it sits in the same sonic space as the original.
A sampler plays back a recorded audio file at different pitches. The result inherits artefacts of the source recording — room, key, transient. Mimic produces a real synthesizer patch tuned to match the timbral character. You play any note, any velocity, with no time-stretch artefacts and full control over every synth parameter.
Mimic doesn't reproduce or store the source audio in the output. The synth patches it generates are parameter sets — not recordings — and the resulting sound you perform is your own composition through your own MIDI input. As with any creative tool, you're responsible for using it within applicable copyright law in your jurisdiction.
No. Mimic v1.0 will be a one-time purchase. Pricing is to be confirmed at launch — sign up for early access below and you'll get launch-window pricing plus founder-supporter perks.
Mimic is being built in four sequential phases. Each phase ships a real, tested deliverable before the next phase begins — no perpetual pre-order in the dark.
Public web demo proving the full source-to-playable user flow. UI/UX language locked. Architecture and stem-separation model validated.
Custom neural model that maps a target sound's spectral fingerprint into our synthesizer's parameter space. Four candidate patches per query, refinement loop, full evaluation harness.
Full multitimbral synth in JUCE/C++: oscillators, filters, envelopes, LFOs, effects. VST3 + Audio Unit shell with AI parameter binding. Closed alpha with founding subscribers.
Final UI per design brief, preset manager, copy-protection, signed installers for macOS and Windows. Open beta, bug-fix sprint, public release.
Subscribers get launch-window pricing, founder-supporter perks, and the closed beta a month before public release. No spam — one email per real milestone.