Skip to contentSkip to content

Download a Portable Release

Current GitHub Releases contain a complete Windows portable directory, not the old standalone app.exe / PyInstaller package. Download every volume for the hardware variant you need, extract it, and run Win-Start.bat. Python, uv, PortableGit, the matching dependencies, and the models are included.

How current release packages are built

When a version tag triggers .github/workflows/build-and-release.yml, the release workflow creates CPU, CUDA 13.0, CUDA 12.6, and ROCm 7.2.1 portable packages separately:

  1. CI first downloads manga-translator-ui-portable.7z from the portable release. This base already contains Python 3.12, uv, and PortableGit; the tagged source tree is then overlaid. PortableGit is retained from the base instead of being downloaded again for every version build.
  2. It exports the matching dependency group from the locked uv.lock and installs those packages directly into packaging/python inside the archive.
  3. The ROCm 7.2.1 variant separately installs the Windows Radeon ROCm 7.2.1 SDK and matching PyTorch build; the cuda13.0 dependency group uses PyTorch cu130, while cuda12.6 uses PyTorch cu126.
  4. It adds the model files and smoke-tests PyQt6, PyTorch, and ONNX Runtime.
  5. It creates approximately 1990 MiB 7-Zip volumes and attaches every volume to the versioned GitHub Release.

packaging/build_packages.py and the PyInstaller specs remain available for local builds, but current CPU, CUDA 13.0, CUDA 12.6, and ROCm 7.2.1 GitHub Release assets use the portable-Python workflow above. It is therefore normal for the release directory not to contain app.exe.

Download and choose a variant

Open the latest version on GitHub Releases, then download every file with the prefix for your hardware:

Release filename prefixBest forRuntime
manga-translator-cpu-vX.Y.Z.7z.*Any Windows x64 computer; use it without a discrete GPU or when compatibility is uncertainCPU PyTorch / ONNX Runtime
manga-translator-cuda13.0-vX.Y.Z.7z.*Turing (compute capability 7.5) or newer NVIDIA GPUs whose driver supports CUDA 13.0; recommended for RTX 50-series GPUsCUDA 13.0 / PyTorch cu130
manga-translator-cuda12.6-vX.Y.Z.7z.*Required for GeForce 10-series GPUs; also for other NVIDIA GPUs needing the compatibility build. Drivers supporting CUDA 13.0 or newer can run it through backward compatibilityCUDA 12.6 / PyTorch cu126
manga-translator-rocm7.2.1-vX.Y.Z.7z.*AMD GPUs supported by Windows ROCmExperimental Radeon ROCm 7.2.1; AMD driver 26.2.2 is required

The ROCm 7.2.1 package works only with supported AMD GPUs. Use the CPU package when uncertain, and never mix volumes from different runtimes.

CUDA 13.0 no longer supports NVIDIA architectures before Turing. The installer explicitly recognizes GeForce 10-series model names and forces CUDA 12.6; it also conservatively falls back to CUDA 12.6 when compute capability cannot be detected.

Download and extract the volumes

  1. Download all .7z.001, .7z.002, and later volumes for the selected variant into one directory without renaming them.
  2. Use 7-Zip or another extractor that supports split 7z archives, and extract only .7z.001; the remaining volumes are read automatically.
  3. Extract to a writable, short path such as D:\manga-translator-ui\. Do not run inside the archive or combine volumes from different versions.

Extraction fails when any volume is missing, incomplete, or renamed. Normal volumes are approximately 1990 MiB; the final volume is usually smaller.

Start the application

  1. Install the Microsoft Visual C++ 2015–2022 Redistributable (x64).
  2. Open the extracted directory and double-click Win-Start.bat.
  3. The launcher adds the bundled PortableGit\cmd to PATH, then uses packaging\python\python.exe to run desktop_qt_ui\main.py. Initial model loading can take time, but it does not reinstall the bundled Python dependencies or models.
  4. Run Win-Install-or-Update.bat to check dependencies, synchronize source, or switch versions or branches. The maintenance program prefers the bundled PortableGit\cmd\git.exe.

Update or change hardware variants

  • Update in place: run Win-Install-or-Update.bat and select [2] Update. This requires network access and synchronizes the source and current dependencies.
  • Download a new release package: use this to repair a damaged environment or switch between CPU, NVIDIA, and AMD. Extract to a new directory, then migrate any config/, result/, and logs/ data you want to keep.

Do not copy another variant's packaging/python over the current one. The three PyTorch runtimes are mutually exclusive.

Troubleshooting

  • There is no app.exe: current releases start through Win-Start.bat and the bundled Python runtime; this is expected.
  • Extraction fails: verify that every volume has the same version and hardware prefix, finished downloading, and is being extracted from .001.
  • The launcher exits immediately: install the x64 Visual C++ runtime, check whether antivirus quarantined scripts, DLLs, or Python files, then run Win-Install-or-Update.bat to inspect the environment.
  • The NVIDIA build cannot use the GPU: update the driver and confirm that it supports the CUDA version selected by the package; use the CPU package if the problem remains.
  • The AMD build cannot load PyTorch: confirm that the GPU is supported by Windows ROCm 7.2.1 and install AMD driver 26.2.2; use the CPU package when unsupported.