Aller au contenu

Install on Windows (preview)

Cette page n’est pas encore traduite : voici la version anglaise.

  • Windows 10 (22H2) or Windows 11, 64-bit (x64). Windows on Arm is not supported yet.
  • Disk space for the app (about 1.5 GB installed), the speech model (about 1.6 GB) and your renders.
  • An AI model you already have access to: a Claude Code or Codex login, an API key, or a local model. You can also start with none; planning then falls back to simple rules.
Mac (Apple Silicon) Windows (preview)
Speech recognition MLX Whisper on the GPU faster-whisper on the CPU, which is noticeably slower than on a Mac. An NVIDIA GPU can be used, see below
H.264 encoding VideoToolbox Media Foundation (built into Windows), falling back to OpenH264 when Media Foundation has no H.264 encoder (for example Windows N without the Media Feature Pack)
API keys macOS keychain Encrypted with Windows DPAPI under your user account
Installer Signed and notarized DMG, updates itself Unsigned NSIS installer on GitHub Releases, so SmartScreen warns. Checks the same releases for updates
Free offline voice Kokoro, about 330 MB, for narration without a key Not available: narration (explainers, courses) needs an OpenAI key or a voice you set in the persona
Shortcuts ⌘ Ctrl (Ctrl+1 to Ctrl+4 for Home, Inbox, All projects, Publish)
  1. Open the latest release and download Reelfold-<version>-win-x64-setup.exe (from v0.2.0 on, next to the Mac DMG).

  2. Run it. Because the installer is not code-signed, Windows shows Windows protected your PC: choose More info → Run anyway. Only run installers downloaded from this repository’s Releases page.

  3. The installer installs Reelfold for your user (no admin rights needed; you can change the folder) and adds a Start menu shortcut. It contains its own Python, ffmpeg and engine, like the Mac app.

Reelfold checks GitHub Releases for updates and installs them when you quit. Updates on Windows are new: if one fails, download the next installer from Releases and run it over the old one. Your projects and settings are kept.

You need Node.js 22+, Python 3.12 (tick Add python.exe to PATH), Git for Windows and an ffmpeg on PATH (for example winget install Gyan.FFmpeg).

  1. In PowerShell, clone the repository and install the engine’s Python packages and the app’s packages:

    Terminal window
    git clone https://github.com/zyziyun/reelfold; cd reelfold
    python -m pip install -r requirements.txt
    npm install

    On Windows requirements.txt installs faster-whisper instead of MLX Whisper.

  2. Start the app:

    Terminal window
    npm run desk

    The app opens as Reelfold and starts its engine with the python on your PATH. If it picks the wrong one, set the Python path in Settings.

Fonts and face models are downloaded by the app on first launch (Settings › Advanced › Downloads). The Whisper model is downloaded the first time you transcribe.

Maintainers can build the same unsigned installer for any branch in CI (Actions → Desk release → Run workflow with windows ticked); it is kept as a workflow artifact for 7 days. Only run installers from this repository’s own Releases or Actions runs.

Open Settings › AI, the same as on the Mac. For the subscription logins, install the CLI first:

  • Claude Code: the native installer (irm https://claude.ai/install.ps1 | iex) or npm install -g @anthropic-ai/claude-code.
  • Codex: npm install -g @openai/codex.

Reelfold finds them on PATH, in %USERPROFILE%\.local\bin and in npm’s global folder (%APPDATA%\npm), so a CLI installed while the app is running is picked up without a restart.

By default faster-whisper runs on the CPU on Windows, because the CUDA libraries it needs (cuBLAS and cuDNN for CUDA 12) are not part of the app. If you have installed them, set the environment variable VSTUDIO_WHISPER_DEVICE=cuda (or auto) for your user and restart Reelfold. If the CUDA libraries cannot be loaded, transcription stops with a message saying so; remove the variable (or set it to cpu) to go back to the CPU.

  • Very deep folders: Windows limits paths to 260 characters unless long paths are enabled (Group Policy Enable Win32 long paths). Keep projects in a short folder such as D:\Reelfold if a render fails with a path error.
  • The publish browser, scheduled publishing and the Create page’s recorder are less tested on Windows.
  • The installer and the app are unsigned, so SmartScreen (and some antivirus tools) warn until signing is set up.
  • There is no free offline voice yet; add an OpenAI key for narrated videos.

How the editions compare: Editions: Mac, App Store, Windows.

Folder What is in it
%APPDATA%\Reelfold App settings, downloaded models and fonts (assets), encrypted API keys, the publish browser’s platform logins
%USERPROFILE%\.config\vstudio The engine’s home: the project list, publish calendar and preferences; new projects go to projects\ here by default
%USERPROFILE%\.cache\video-studio Engine cache: fonts, models, transcripts. Shared with the Claude Code skill
%USERPROFILE%\.cache\huggingface\hub Whisper models downloaded by a source build or the skill

Your original footage is never moved or modified.