This extension addresses the challenge of generating and upscaling high-resolution images (2K+) with limited VRAM (<=6GB) for Stable Diffusion users. It provides advanced tiling techniques for both diffusion and VAE processes, enabling users to create or enhance large images that would otherwise be impossible on lower-end hardware.

How It Works

The extension implements several state-of-the-art tiling techniques, including Tiled Diffusion and its variants (MultiDiffusion, Demofusion), alongside original Tiled VAE and Tiled Noise Inversion methods. This approach breaks down large images into smaller, manageable tiles that are processed sequentially. By tiling the diffusion and VAE operations, the extension significantly reduces VRAM requirements, allowing for the generation and upscaling of ultra-large images on consumer-grade GPUs.

Quick Start & Requirements

• Install: Install as a Stable Diffusion WebUI extension.
• Prerequisites: Stable Diffusion WebUI, Python 3.10.6, PyTorch 2.0.1, CUDA 11.8 or 12.1.
• Documentation: Detailed documentation and examples are available on the wiki. A quickstart tutorial is also provided.

Highlighted Details

• Supports ultra-large image generation (txt2img) and upscaling (img2img).
• Features Regional Prompt Control for localized prompt application.
• Integrates with ControlNet, StableSR, SDXL, and Demofusion.
• Offers original Tiled VAE and Tiled Noise Inversion methods.

Maintenance & Community

The project is actively maintained. Further details on community engagement or roadmaps are not explicitly provided in the README.

Licensing & Compatibility

Licensed under CC BY-NC-SA 4.0. Versions after March 28, 2023, prohibit commercial sales of the code itself, though derived artworks are not restricted. This license may restrict integration into closed-source commercial applications.

Limitations & Caveats

The CC BY-NC-SA 4.0 license restricts commercial use of the code. Users should be aware of the specific date cutoff for commercial restrictions.