sd-forge-layerdiffuse  by lllyasviel

WebUI extension for transparent image layer diffusion

created 1 year ago
4,080 stars

Top 12.3% on sourcepulse

GitHubView on GitHub
Project Summary

This project provides an extension for Stable Diffusion WebUI (Forge) to generate transparent images and layers, enabling detailed effects like semi-transparent objects and intricate textures. It targets users seeking advanced control over image composition and transparency beyond simple background removal.

How It Works

The extension leverages specialized models that modify the latent space of Stable Diffusion to represent transparency. It introduces custom VAE encoders and decoders to process and generate images with alpha channels. Different model variants are available for SDXL and SD1.5, supporting various generation modes including transparent-only, foreground-to-background, background-to-foreground, and joint generation of multiple layers.

Quick Start & Requirements

  • Installation is typically done via the WebUI's extension management system.
  • Requires a compatible version of Stable Diffusion WebUI (Forge).
  • Specific models (.safetensors files) need to be downloaded and placed in the appropriate extensions directory.
  • Refer to the GitHub repository for detailed setup and usage examples.

Highlighted Details

  • Supports native transparent diffusion for effects like glass and fur.
  • Offers automatic model downloading and selection for ease of use.
  • Provides specific models for SDXL and SD1.5, including LoRAs and full model offsets.
  • Includes detailed sanity check examples to verify results.

Maintenance & Community

  • The project is marked as "WIP" (Work In Progress).
  • Further details on community engagement or roadmap are not explicitly provided in the README.

Licensing & Compatibility

  • The README does not specify a license. Users should verify licensing terms before commercial use.

Limitations & Caveats

  • The project is explicitly stated as "WIP," indicating potential instability or incomplete features.
  • Some SDXL workflows currently require a two-step process, with a one-step model planned for release.
  • There are noted differences between training scripts and WebUI samplers (e.g., DPM++ vs. Euler A) that may cause artifacts, with ongoing efforts to resolve these.
Health Check
Last commit

11 months ago

Responsiveness

1 day

Pull Requests (30d)
0
Issues (30d)
1
Star History
57 stars in the last 90 days

Explore Similar Projects

Feedback? Help us improve.