stablediffusiongui-win-ROCm6

• stablediffusiongui-windows AMD ROCm 6

  ROCm is x2 faster than vulkan.

  Binaries for Radeon RX 7900XTX , 7900XT, 7800XT, 7700XT, 7600, 6950XT, 6900XT, 6800XT and 6800.

  Other Radeon, use vulkan.

• INSTALL

1. You need to install AMD HIP SDK

2. Download binaries for Radeon

3. Extract zip.

4. Double click StableDiffusionGUI-x.x-win64.exe. This exe is installer for windows. I recommend NOT install CUDA Backends. stable-diffusion_cuda.dll is unnecessary, file size is very large.

5. Make .bat's desktop shortcut if needed.

   bat files are in installed folder which you specified when you install.

• START

  Double click stablediffusiongui.exe.

  If AMD HIP SDK is detected (HIP_PATH is detected in PATH), backend become ROCm HIP, stable-diffusion_hipblas.dll. If not, backend is vulkan, stable-diffusion_vulkan.dll. (HIP_PATH is configured by AMD HIP SDK installer automatically. So, you do not need configure HIP_PATH and HIP_PATHbin manually.)

  Depend on your PC, you may need to configure HIP_VISIBLE_DEVICES. Two sample .bat files are included. Try them!

These are commandline options. -avx, -avx2, -avx512, -hipblas, -vulkan, -cuda. hipblas means ROCm. cuda means nvidia. For example, if you want to run FORCE VULKAN, add -vulkan. Sample .bat file is included.

• UNINSTALL

  Use windows OS's uninstaller or use Uninstall.exe .

• KNOWN ISSUE

  SDXL speed is slow: Compared with Comfy, SDXL's speed is slow, takes x2 sec/it. though, FLUX speed is slightly faster then Comfy.This SDXL-SLOW issue is sd.cpp's. radeon, nvidia, intel, and even cpu are affected.

sd.cpp.gui.wx - Stable Diffusion GUI

A cross-platform GUI for Stable Diffusion C++, built using wxWidgets.

Licensed under the MIT License.

Features

• Text-to-image (text2img) generation
• Image-to-image (img2img) generation
• Inpaint & Outpaint, save / load masks, built-in inpaint / outpaint editor
• Tiling for low memory
• Headless computation with SDGUI Server even in containerized mode
• Model quantization in-fly or manually
• Built-in upscaling capabilities
• ControlNet integration
• Model conversion to GGUF format
• Optional integration with CivitAi Model Downloader
• Hardware detection for optimal diffusion performance
• Integrated model management tools
• VAE support
• TAESD support
• Customizable generation presets
• Customizable output file names
• ComfyUI compatible Prompt styles
• Generation queue for batch processing
• Real-time progress monitoring during generation
• Save and load metadata directly from images
• Multi-language GUI support

Supported Backends

• CPU (AVX, AVX2, AVX512)
• CUDA
• VULKAN

---

OS Compatibility

Refer to the GitHub Releases page for available installers.
For compiling on unsupported operating systems, consult the Wiki.

---

---

Use at your own risk!
These releases are preview versions and may contain bugs. If you encounter issues, please report them via the GitHub Issues page.