zimage-ncnn-vulkan  by nihui

AI image generation powered by ncnn and Vulkan

Created 2 months ago
363 stars

Top 77.8% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

This project offers a portable, high-performance implementation of the Z-Image generative model using ncnn and Vulkan. It enables efficient image generation across diverse Intel, AMD, NVIDIA, and Apple-Silicon GPUs without requiring heavy runtimes like CUDA or PyTorch, targeting users needing fast, cross-platform AI image synthesis.

How It Works

zimage-ncnn-vulkan employs the ncnn framework for efficient, cross-platform neural network inference and Vulkan for GPU acceleration. It implements the Z-Image diffusion transformer model, optimizing execution for broad hardware compatibility. This approach bypasses CUDA or PyTorch dependencies, providing a self-contained, performant AI image generation solution.

Quick Start & Requirements

Pre-compiled binaries for Windows, Linux, and macOS are available, requiring no additional runtime environments. Users must download z-image-turbo and z-image models separately. Minimum requirements: 16GB RAM and a Vulkan-capable GPU. Windows users need (Half system RAM) + (GPU memory) >= 16GB due to WDDM. Recommended: 32GB RAM, 16GB GPU with tensor core support. Releases: https://github.com/nihui/zimage-ncnn-vulkan/releases. Models: https://huggingface.co/nihui-szyl/z-image-ncnn/tree/main.

Highlighted Details

  • Self-contained portable executables eliminate CUDA or PyTorch installations.
  • Broad GPU support via Vulkan: Intel, AMD, NVIDIA, Apple-Silicon.
  • Efficient inference powered by the ncnn framework.
  • Integrated support for WebP, JPEG, and PNG encoding/decoding.

Maintenance & Community

The project is primarily maintained by nihui. No specific community channels (e.g., Discord, Slack) or detailed roadmap information are provided in the README.

Licensing & Compatibility

The main project license is not explicitly stated. It incorporates permissively licensed dependencies: ncnn (BSD-style), libwebp (BSD-style), libjpeg-turbo (BSD-style), libpng (BSD-style), zlib-ng (Zlib), and dirent (MIT-like). Commercial use compatibility is likely permissive but requires explicit license verification.

Limitations & Caveats

The software is in an "early development stage" and may contain bugs. Windows users face specific RAM requirements due to WDDM limitations. Users experiencing issues should ensure GPU drivers are up-to-date.

Health Check
Last Commit

4 days ago

Responsiveness

Inactive

Pull Requests (30d)
14
Issues (30d)
11
Star History
321 stars in the last 30 days

Explore Similar Projects

Starred by Alex Yu Alex Yu(Research Scientist at OpenAI; Cofounder of Luma AI), Lianmin Zheng Lianmin Zheng(Coauthor of SGLang, vLLM), and
2 more.

HunyuanVideo by Tencent-Hunyuan

0.3%
12k
PyTorch code for video generation research
Created 1 year ago
Updated 3 months ago
Starred by Chip Huyen Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"), Lyumin Zhang Lyumin Zhang(Author of ControlNet), and
4 more.

Fooocus by lllyasviel

0.1%
48k
Image generator for streamlined prompting and generation using SDXL
Created 2 years ago
Updated 2 months ago
Feedback? Help us improve.