This repository provides a custom script for AUTOMATIC1111's Stable Diffusion WebUI, enabling users to apply different prompts to distinct regions of an image. It targets users who need fine-grained control over image generation, allowing for complex compositions and stylistic variations within a single output.

How It Works

The script operates by dividing the image into specified regions, either horizontally, vertically, or in a 2D grid. Users define prompts for each region, separated by "BREAK" keywords. It supports two primary calculation modes: "Attention" mode, which is faster and suitable for most use cases, and "Latent" mode, which is slower but offers better separation for elements like LoRAs. New features include region specification via inpainting masks or prompt analysis.

Quick Start & Requirements

• Installation: Install as a custom script within the Stable Diffusion WebUI.
• Prerequisites: Stable Diffusion WebUI (AUTOMATIC1111 or Forge), Python.
• Usage: Refer to the official documentation for detailed setup and usage examples.

Highlighted Details

• Supports 2D region division using ADDCOL and ADDROW separators.
• Offers "Mask" mode for defining regions via custom drawn or uploaded masks.
• Includes "Prompt" mode for automatically identifying regions based on prompt keywords.
• Latent mode allows for better LoRA separation but is slower.

Maintenance & Community

The project is actively maintained, with recent updates including support for reForge and bug fixes for Forge compatibility. Community discussions and support are likely available through the project's GitHub repository.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Users should verify licensing for commercial use or closed-source integration.

Limitations & Caveats

Latent mode with LoRA/LoCon/LoHa may experience slower generation speeds or compatibility issues depending on VRAM and WebUI specifications. There are known issues with mask uploading/loading in Gradio and potential LoRA corruption in Latent mode, with suggested workarounds provided.