Mimic v1.4.4

Add macOS support for Apple Silicon

Enable Metal acceleration, native macOS integration, and microphone
permissions. Build app and DMG release artifacts with optional Apple
signing and publish macOS updater metadata.

Mimic v1.4.3

Fish Audio Drama 3

A new engine that acts. Drama 3 is a 35B mixture-of-experts model from
Fish Audio, running locally on audio.cpp.

• Tone and sound directions. Add sound now has whisper, shout,
  angry and sad alongside laugh, sigh, pause and the rest. Drama 3 also
  reads any direction you type in brackets, like [nervous laugh] or
  [slowly, almost asleep].

• Scenes with several voices. Add sound › Switch speaker drops a
  speaker chip into the script, and that voice speaks the words after it.
  Up to 16 cloned voices in one take.

• Download it from Settings › Engines (about 24 GB). It needs
  Windows and an NVIDIA GPU: 12.9 GB of VRAM plus about 11 GB of RAM,
  measured on an RTX 5070 Ti.

• Licensed under the Fish Audio Community License: free for research and
  non-commercial use, and for businesses under US$500k a year.

Recommended settings

• Every slider in Controls now shows the engine's recommended value.

• When a setting is in a range known to cause trouble, Mimic says so
  next to Generate and offers Use recommended. For Drama 3 that's
  Temperature below 1.0, which can leave silence where a new speaker
  starts, and a Top K limit of 30 or less.

Fixes

• Sound tags on the second and later lines of a script were read as
  plain words. They now work on every line, for every engine.

• The Add sound menu scrolls when it's taller than the window, instead
  of running off screen.

Mimic v1.4.2

bug fixes and security improvements

Mimic v1.4.1

What's new

• Mimic on your phone. The browser version now has a real phone layout:
  • One panel at a time: a tab bar at the bottom switches between
    Voices, Script and Takes.
  • Generate takes you to Takes, so you can watch the take being made.
    If a take finishes while you're on another tab, a dot appears on Takes.
  • Built for touch: a swipeable row of engines, a full-width Generate
    button, bigger buttons and a larger waveform to tap.
  • Settings and other dialogs open as full-width sheets that scroll.
  • Tablets in portrait get the same layout.
• Easier Tailscale setup. Settings › Use from another device now
  includes a step-by-step guide to Anywhere (Tailscale). If Tailscale
  isn't installed yet, or HTTPS still needs turning on for your devices,
  Mimic explains the problem right under the buttons, with a button that
  opens the right Tailscale page.

Fixes

• Problems starting sharing now show up next to the sharing buttons,
  instead of only in the status bar at the bottom of the window.
• On touch screens, Mimic no longer shows keyboard-shortcut hints like
  "Ctrl S saves it".

Mimic v1.4.0

What's new

• Use Mimic from another device. Open Mimic in a browser on your
  laptop or phone while this PC's GPU does the work. Start it from
  Settings › Use from another device:

  • This network: a link for devices on the same Wi-Fi.
  • Anywhere (Tailscale): an HTTPS link that only your own devices
    can reach, from any network. Recording from the other device works too.

  Nothing is shared until you press a button, and sharing stops when
  Mimic closes. The link carries a secret, and New link signs out
  every device that used the old one. In the browser, Save downloads the
  take to that device, and imports upload clips from it.

• AMD Radeon GPUs on Linux (experimental). Engines now run on AMD
  graphics through ROCm. RX 7000 and 9000 cards work out of the box, and
  RX 6000 cards run in compatibility mode. Radeon integrated graphics
  (680M, 780M, 890M) run when you set Run engines on to GPU.

• Smarter GPU detection on every Linux distro. Mimic finds GPUs by
  their hardware ID, with or without lspci or nvidia-smi. It
  recognizes Intel Arc and Core Ultra graphics even when "Arc" isn't in
  the name, and picks the right GPU on laptops that have two.

• Mimic tells you how to fix a GPU it can't use. If a driver,
  runtime or permission is missing, Settings shows the exact command for
  Arch, Debian/Ubuntu or Fedora. If an engine falls back to the CPU, Mimic
  says why.

• Switching between GPU and CPU really switches. After you change
  Run engines on, Settings › Engines offers each downloaded
  engine's build for the new choice (Get GPU build).

• A new recording passage. Type your name and read a short, natural
  passage that starts with it ("Hey, it's …"). The new voice is named
  after you, and the transcript matches what you said, numbers included.

Responsible use

• Before your first clone, Mimic asks you to confirm you have permission
  to use the voice.
• Takes you save or download are marked as AI-generated in the file's
  information.
• KittenTTS now shows its real license (Stellon Labs Community) and
  credit. Credits and licenses for every engine and library ship with the
  app.

Fixes

• PyTorch and its audio library are now always installed as matching
  versions. Intel GPU installs could get mismatched ones.
• Security hardening: stricter protection against injected content in
  the window, deletes can't reach outside Mimic's folders, settings and
  the sharing secret are private to your account on Linux, and the Fish
  server only listens on this PC.

Mimic v1.3.2

Bug Fixes

Mimic v1.2.0

What's new

• New voice button. Cloning a voice now starts from one obvious
  place: the red New voice button at the top of Voices. Choose
  Record your voice or Use an audio file, and Mimic writes the
  transcript for you. You can still drop audio files anywhere on the
  window, and Mimic now shows when it's ready to take them.

• Controls panel. Open Controls next to Generate to tweak a
  take:

  • Speed on every engine, without changing the pitch
  • Temperature, Top P, Top K and repetition penalty for Kitten,
    Qwen and Breeze
  • Pause between sentences for Kitten and NeuTTS
  • Guidance (CFG) for Breeze in Design and Direction modes
  • Seed, moved here from Options

  Settings are saved per engine, and each one has its own Reset.

• In-app updates. Mimic checks for new versions when it starts and
  can install them and restart for you. You can also check in Settings ›
  About. Arch builds from the PKGBUILD show you when a new version is
  out and how to update with makepkg.

• A livelier look. There's a short startup animation, the logo is
  bigger on large screens, and it comes alive now and then and while a
  take is being made.

Fixes

• Import a file works again. It used to fail with invalid args 'path' for command 'import_path'.
• Linux packages (.deb, .rpm and AppImage) are attached to the
  release again. v1.1.0 shipped with only the Windows installer.

Upgrading

Install this version by hand one time. After that, Mimic updates itself.

Mimic v1.1.0

Mimic v1.1.0

A polish release: Mimic gets a proper identity, and the project page
shows what it does.

New

• New logo and app icon. Five sound bars that trace an "M", with a
  red "recording" bar in the middle. The mark also appears next to the
  name in the app's header.
• Rebuilt project page. The README now has an animated banner, a
  real-speed demo of a take being generated, screenshots on Windows and on
  Linux with Caelestia colors, and a clearer install and quick-start
  guide.
• Measured numbers. Charts of how much memory each engine needs on
  GPU and CPU, and how fast Qwen3-TTS runs on each.

Fixed

• The Engines list now shows KittenTTS's real download size (about 1 GB
  of model plus PyTorch, not 3 GB).

Under the hood

• Each release now publishes once, after the Windows and Linux builds
  both finish, with these notes taken from the version tag.
• For testing: MIMIC_DATA_DIR runs Mimic with a separate set of
  profiles and settings, and the built-in test runner can type, play takes
  and switch themes.

No engine or model changes. Your profiles, settings and downloaded
engines carry over.

Mimic v1.0.0

Mimic v1.0.0

Mimic is a local voice-cloning app. Write a script, pick a voice, and generate speech on your own GPU or CPU. Nothing leaves your computer.

Highlights

• Voice profiles. Record a voice by reading a short passage, import a clip, or drag audio files onto the window. Each profile keeps its clip, transcript and favorite engine settings.
• Automatic transcripts. Imported clips are transcribed locally with faster-whisper on the CPU. No typing needed.
• Multiple engines. KittenTTS 2, Breeze TTS 2, Fish S2 Pro, Qwen3-TTS (0.6B and 1.7B), and NeuTTS Air on Linux.
• Download engines from inside the app. Settings › Engines shows what each engine needs and whether it fits your PC, with one-click download and remove. Pin the engines you use to the top bar.
• Runs on GPU or CPU. Mimic detects your hardware and runs models on an NVIDIA GPU, an Intel Arc GPU (experimental), or the CPU. The memory bar shows real VRAM or RAM use.
• One engine in memory at a time. Switching engines unloads the previous one automatically, and idle engines unload after a set time.
• Takes are never saved behind your back. They stay in a temporary cache until you press Save (Ctrl+S) or Save as (Ctrl+Shift+S). WAV, FLAC or MP3.
• Compare engines. Generate the same line with every engine, optionally blind, and keep the best take.
• Linux support. Native on Wayland and X11, with a Caelestia theme that follows your shell's live colors and a clean window on Hyprland (class mimic).
• Keyboard first. Ctrl+Enter generates, Space plays, Ctrl+K opens the command palette, Ctrl+, opens Settings.

Downloads

• Windows: Mimic_1.0.0_x64-setup.exe
• Linux: .deb, .rpm or AppImage
• Arch Linux: build with packaging/arch/PKGBUILD (makepkg -si)

Downloading engines requires uv, and importing audio requires ffmpeg.

Good to know

• CPU memory: on a CPU, Qwen3-TTS 0.6B needs about 6 GB of RAM and 1.7B about 11 GB. KittenTTS 2 is too heavy for most laptop CPUs.
• NeuTTS Air (Linux only) needs a free Hugging Face account. Accept the model terms, then paste a read token in Settings › Engines and memory. It also needs espeak-ng, cmake and base-devel.
• Fish S2 Pro runs only on Windows with an NVIDIA GPU (via WSL). Breeze TTS 2 needs an NVIDIA GPU and is licensed for non-commercial use only.
• Model licenses: each model has its own license. Check them before commercial use.

Only clone voices you have permission to use.