Install Omnichar Studio

You need Python 3.11 or newer. The web UI ships as a Python package, so there is no Node step. --install --extra all installs everything: the engine, the model runtime, the trainer and the UI.

macOS / Linux
Windows
pip

git clone https://github.com/omnichar/OmniChar
cd OmniChar/core
./webui.sh --install --extra all
./webui.sh

Use webui.bat. webui.sh is a bash script and won’t run in PowerShell.

git clone https://github.com/omnichar/OmniChar
cd OmniChar/core
.\webui.bat --install --extra all
.\webui.bat

From the repo root, install the whole app from PyPI, then start it:

pip install -r requirements.txt
inline-studio

Open http://127.0.0.1:8848 in your browser. On NVIDIA, --install reads your GPU’s compute capability and installs the matching CUDA build of PyTorch, RTX 50-series included. Everything goes into core/.venv. An environment already active in your shell is left alone.

No GPU? Deploy Omnichar on RunPod, or use Omnichar Cloud in your browser.

Update

Pull and re-run the install. It upgrades the engine and the prebuilt UI package together.

git pull
./webui.sh --install

The app checks PyPI once a day and prints a notice at launch when either half is behind. Set INLINE_NO_UPDATE_CHECK=1 to turn the check off.

Hardware support

Hardware	Status	Extra steps
NVIDIA, Linux	Tested	None
NVIDIA, Windows	Supported	None. `--install` picks the CUDA build for your card.
Apple Silicon (MPS)	Code path exists, untested	None. No int8 on MPS, so a model must fit in unified memory.
AMD ROCm, Linux	Untested, reports welcome	Install a ROCm build of PyTorch after the normal install.
AMD ROCm, Windows	Untested, reports welcome	Pass AMD’s own wheel index with `--torch-index`.
CPU only	Works, very slow	`./webui.sh --cpu`

RTX 50-series

RTX 50-series cards need a CUDA 13 build, and --install picks cu130 for you. CUDA 13 needs driver R580 or newer. On an older driver, --install falls back to cu128 and says so. That build works but is frozen at torch 2.11, so update the driver when you can. To see what your driver reports and which index would be used, without installing anything:

.\webui.bat --print-torch-index

Paste that line into any bug report about CUDA.

AMD ROCm on Linux

Install normally first, then replace PyTorch, so nothing overwrites your ROCm build afterwards. Pick the index that matches your ROCm version from pytorch.org.

cd core
./webui.sh --install --extra runtime
uv pip install --python .venv/bin/python --force-reinstall \
  --index-url https://download.pytorch.org/whl/rocm6.2 torch
.venv/bin/python -c "import torch; print(torch.cuda.is_available(), torch.version.hip)"

The last line should print a HIP version, not None. Don’t run uv sync or --recreate afterwards: both put the default PyTorch back. Either way, open an issue with how it went.

Troubleshooting

The UI is a blank page on Windows. You’re on a build older than v1.3.0. Update and the app sets the right content type itself. The CUDA build is wrong for my card. Name the index yourself, for example --torch-index cu130. This also replaces an installed torch, which a plain re-run does not.

Get started

Omnichar Cloud

Omnichar Studio

ComfyUI node

Install Omnichar Studio

Update

Hardware support

RTX 50-series

AMD ROCm on Linux

Troubleshooting

​Update

​Hardware support

​RTX 50-series

​AMD ROCm on Linux

​Troubleshooting

Update

Hardware support

RTX 50-series

AMD ROCm on Linux

Troubleshooting