UI for human-in-the-loop controlled image synthesis
Planned features:
- Generation of images based on text
- Hyperparameters control
- Scatter/select creation, with progressive refinement
- Branch/continue generation of images (tree of creation)
- Latent space editing
- Latent space constraining for continuation
- More? post your ideas here!
...
...
The server is meant to be installed in machines (even remote) that have a 10 GB+ GPU and do not need to have a user interface. In case of local installs both the server, and the UI can run on the same machine.
# create a virtual environment if needed (we assume you know how to do it)
# install PyTorch for your CUDA version - example below (windows, PyTorch 1.7.1 for CUDA 11.0)
pip install torch===1.7.1+cu110 torchvision===0.8.2+cu110 torchaudio===0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
pip install -r requirements.txt
For now this has been tested and verified ont the following configurations:
OS | GPU Config | PyTorch | Working |
---|---|---|---|
Windows 10 | CUDA 11.0 + cuDNN 11.0 | 1.7.1 (CUDA 11.0) | Yes |
This 'React' APP can be built and served by any web server, and will use 'socket.io' websockets to connect anc control the server.
...