This extension for AUTOMATIC1111's Stable Diffusion WebUI enables the generation of complex scenes with multiple distinct subjects. It targets users aiming to create composite images with foreground elements placed onto a background, streamlining a process that would otherwise require extensive manual editing.

How It Works

The script orchestrates a multi-stage generation process. It first creates a background image based on the primary prompt. Then, it generates foreground subjects using a separate prompt, potentially with multiple lines for distinct entities. A depth analysis (using MiDaS) is performed on the foreground subjects to enable background removal. These subjects are then composited onto the background, followed by an img2img pass for a seamless blend. This approach automates complex compositing and blending, leveraging depth estimation for accurate subject isolation.

Quick Start & Requirements

• Installation: Copy the repository URL into the Stable Diffusion WebUI's extension tab or clone the repository into the extensions folder. A UI restart may be necessary.
• Prerequisites: Requires AUTOMATIC1111/stable-diffusion-webui.
• Links: AUTOMATIC1111/stable-diffusion-webui

Highlighted Details

• Supports distinct prompts and seeds for each foreground subject.
• Offers control over foreground placement, depth cut thresholds, and blend settings.
• Includes options for random superposition and face correction during the final blend.
• Provides guidance on prompt engineering for better subject isolation and blending.

Maintenance & Community

• Developed by Extraltodeus.
• Mentions copying functions from thygate for depth analysis integration.
• Encourages using the discussion tab for non-bug related queries.

Licensing & Compatibility

• The README does not explicitly state a license.
• Compatibility is tied to the AUTOMATIC1111/stable-diffusion-webui, implying potential licensing implications if the latter's license restricts commercial use or linking.

Limitations & Caveats

The extension currently only renders the final blend to the UI; users must manually save all generated images. It is incompatible with the "high res fix" feature, and users should keep standard UI size sliders around 512x512. Interrupting a generation may result in an "List index out of range" error, though it's noted as not causing persistent issues.