1. Workflow Overview

This workflow is designed for local image editing and enhancement, including:

• Object Removal: Mask unwanted areas for AI-based content fill.

• Inpainting: Precise regeneration of masked regions.

• Face Restoration: Enhances facial details via CodeFormer.

• Super-Resolution: Upscales images (e.g., 4x) using AI models.

Core Models:

• Stable Diffusion (Flux): Base image generation/inpainting.

• ControlNet (Canny): Preserves structural consistency.

• CodeFormer: Face refinement.

• 4xNomos8kSCHAT-L: Upscaling model.

2. Key Nodes & Installation

Special Dependencies:

• Flux Model: Manually download GGUF model to models/unet.

• CodeFormer: Place model in models/codeformer.

3. Workflow Groups

1. Input & Mask Processing (Upload Image Group)

   • Input: Image with mask (PNG).

   • Key Nodes: LoadImage, GrowMask.

2. Local Inpainting (Flux Inpainting Group)

   • Uses InpaintModelConditioning with prompts.

   • Output: Latent representation of inpainted area.

3. Face & Upscale (Face Restoration Group)

   • ReActorRestoreFace → ImageUpscaleWithModel.

4. Comparison (Image Comparer)

   • Compares original/processed images.

4. Input/Output

• Inputs:

  • Resolution: Default 1024x1024 (adjustable via ConstrainImage).

  • Negative Prompt: Pre-set (e.g., "low quality, watermark").

  • Mask: Must fully cover target area.

• Output: Final image (PNG) with edits and upscaling.

5. Notes

1. Hardware: 12GB+ VRAM recommended (upscaling is resource-heavy).

2. Errors:

   • Incomplete masks cause artifacts.

   • Missing models: Verify paths for GGUF/CodeFormer.

3. Optimization:

   • Reduce KSampler steps for faster runs.

   • Split workflow if VRAM is limited (inpaint → upscale separately).