Neural Pixel is a fast, Vulkan-powered image generation tool that runs on almost any GPU from 2014+ (requires at least 2GB VRAM for SD 1.5 or 3GB for SDXL). Skip the CUDA/ROCm headache and Python dependency hell, Neural Pixel is simple, portable, and high-performing!

Compatibility

• Neural Pixel supports leading image generation models such as SDXL and FLUX, plus a broad range of community models, extensions, and runtimes.
• For more information about ADetailer and how to convert the files, check this.
• By default, the release ZIP packages include support for Vulkan and CPU inference only.
• No video support at the moment.
• See docs for the full list of supported models, features, formats, platforms, and backends.

Linux Setup & Running

1. Requirements

• OS: Linux kernel >= 5.14 (Tested on RHEL 9, Fedora 42, and Arch).
• Dependencies: GTK >= 4.12, libpng, zlib.
• Vulkan backend (Optional): Vulkan driver/loader/tools & >= 2GB of VRAM.

2. How to Run

• Download the
• Extract the archive and execute the run_neural_pixel file.
• Tip: For debugging, launch from a terminal and enable Terminal Verbose under Extra Options.

Windows Setup & Running

1. Requirements

• Microsoft Visual C++ Redistributable latest: vc_redist.
• A GPU or iGPU with at least 2GB of VRAM for Vulkan.

2. How to Run

• Download the
• Extract the archive and execute the neural_pixel.bat file.

Recommended checkpoints

• Due to hardware limitations, I used q5_1 quantization and the DMD2 speed LoRA (0.7 weight) to generate these examples. The Sam semi-realistic model is the only exception, as it has the LoRA built-in. If you use the base model, your results may differ slightly.

[!WARNING] The links to get the models may display NSFW, mature, or suggestive preview images. Viewer discretion is advised.

Settings used

• Sampler: Euler A (more detail) or LCM (smoother/less detail)
• Scheduler: ays, exponential, or smoothstep
• Steps: 8-12

Build (Linux)

You'll need GTK 4 and the libpng development libraries installed. Then, clone this repository using:

git clone https://github.com/Luiz-Alcantara/Neural-Pixel.git

Next, navigate into the cloned directory and run:

mkdir build && cd build && cmake .. && make

Build (Widows)

To build on Windows follow Windows Build.

Build Stable-Diffusion.cpp

To build sd.cpp follow the instructions on its github page: Stable-diffusion.cpp.

Credits

• This project is a GUI for stable-diffusion.cpp.

Donations

Crypto