The first public release of Gentastic - a modern, no-Python Windows app for running FLUX and Stable Diffusion XL image models locally on your own GPU. Text-to-image, image-to-image, and image editing, with automatic model download and automatic runtime detection.
Download
Grab Gentastic-portable-win-x64.zip below (about 311 MB), unzip it anywhere, and run Gentastic.App.exe. No installer, no .NET, no Python. The build is universal and auto-selects the best backend: CPU, Vulkan (AMD / Intel), or NVIDIA CUDA. Windows x64; a DirectX 12 / Vulkan-capable GPU is recommended (CPU works but is slow).
Features
- Text-to-image and image-to-image, plus keep-face and image editing (FLUX.1 Kontext and FLUX.2 klein).
- Curated model catalog with one-click download from Hugging Face: FLUX.2 klein (the fast default), FLUX.1 schnell and dev, and SDXL models. Optional adult (18+) models behind an age gate.
- GPU VAE decode on Vulkan and CUDA via conv-direct, about 8x faster at 1024 than the previous CPU path.
- Per-model GPU-memory estimates in the picker, with a warning when a model may not fit your card, plus friendlier out-of-memory messages.
- Presets, a gallery with generation parameters embedded in each PNG, and auto-save to Pictures/Gentastic.
- Automatic backend detection (CUDA, ROCm, Vulkan, CPU) and an on-demand CUDA runtime download, so NVIDIA users do not need the CUDA Toolkit.
- Dark / Light / System themes, live progress and cancel.
Notes
- FLUX.1-dev and its finetunes are non-commercial. Set a Hugging Face token in Settings to download gated models.
- Third-party licenses are bundled in the download (THIRD_PARTY_NOTICE.md and the licenses/ folder) and linked from the About page. Gentastic itself is BSD-3-Clause.
Full Changelog: https://github.com/ideaconnect/gentastic/commits/v0.1.0