All in one, batteries-included software to easily generate and upscale AI art using Stable Diffusion
Disclaimer: This software is still in beta, bugs can and will occur at some point. Please use the Issues tab to report a bug. Thank you for using SDToolkit!
- No setup required, just install and run
- Video2x upscaler
- GFPGAN face restoration
- Ability to choose a context image for AI to try to match
- Result image viewer with zoom
- Optimized for low VRAM GPUs
- Many helpful tooltips that make it very easy to use
- Decent amount of configuration options
SDToolkit offers a built-in GFPGAN for face restoration. This is quite a powerful tool to remove face artifacts and beautify the result. It's definitely recommended to use GFPGAN for every prompt that might have faces in it.
SDToolkit offers a built-in Video2x upscaler. Please remember that it isn't perfect, and upscaling by a very large factor can cause artifacts.
- OS: Windows 64-bit
- At least 35GB of disk space
- At least 8GB of GPU VRAM is recommended
- At least 16GB of RAM is recommended
- CUDA-enabled GPU is recommended (Nvidia)
Since SDToolkit is all-in-one software, the model, upscaling software (video2x/GFPGAN) and execution environment (Conda/Python) is included in the setup.
Tested with i7-12700KF and GTX 1080 with full precision. (around 6.7GB of VRAM usage)
Half precision is recommended for RTX cards.
Just download and run the setup from the Releases tab. Be reminded that it's a huge file, you should have at least 35GB of free disk space. Users running the v0.2-beta release, please refer to this hotfix for a comprehensive installation.
The software is licensed under GNU Affero Public License and the SD model is licensed under CreativeML Open RAIL-M. You're required to accept both to use this software. You're free to use the generated art files for commercial purposes as long as they conform to the license terms and conditions.
This software uses Stable Diffusion v1-4 model licensed under CreativeML Open RAIL-M. You're free to download it yourself at CompVis's huggingface repository.
Scripts and model execution software provided by basujindal's fork of stable diffusion.
Upscaling model and algorithm is provided by video2x.
GFPGAN 1.3 model and algorithm provided by Tencent's GFPGAN.