Workflow Overview

This workflow is designed for automatic old photo restoration and enhancement.
It includes:

• Fixing blurry or damaged photos

• Enhancing facial details

• Upscaling images for higher resolution

It utilizes Stable Diffusion, ControlNet, Florence2 AI Image Analysis, GFPGAN for Face Restoration, ReActor Face Swap, and R-ESRGAN Super-Resolution.

Core Models

1. Stable Diffusion (Realistic Vision V5.1)

   • Handles base image restoration.

2. Florence-2 (microsoft/Florence-2-base)

   • Analyzes image content for optimized restoration.

3. GFPGAN (GFPGANv1.4.pth)

   • Enhances facial details.

4. ReActor (inswapper_128.onnx)

   • Performs intelligent face reconstruction.

5. ControlNet Modules

   • LineArt: Preserves edge details.

   • Depth Map: Recovers 3D depth.

   • OpenPose: Maintains human pose structure.

6. Super-Resolution Model (R-ESRGAN_4x+)

   • Upscales images for 4x resolution.

Workflow Structure

This workflow consists of five main stages:

1. Input & Analysis

• LoadImage loads the old photo.

• Florence2Run generates an AI description.

2. ControlNet Processing

• Extracts edges, depth, and pose data.

3. High-Resolution Generation

• KSampler & VAEDecode generate an enhanced version.

4. Face Restoration

• ReActorFaceSwap + GFPGAN fix facial clarity.

5. Upscaling & Output

• UltimateSDUpscale + R-ESRGAN perform final image upscaling.

Inputs & Outputs

Inputs

• Old photo

• AI-generated text descriptions

• Face restoration settings

• Super-resolution model settings

Outputs

• Fully restored high-resolution photo

• Up to 4K quality

Considerations

1. Performance Requirements

   • Since this workflow involves ControlNet, GFPGAN, and super-resolution, it is recommended to use an RTX 3090 or higher GPU for smooth processing.

2. Troubleshooting

   • If restoration results are unsatisfactory, try:

     • Adjusting Florence2-generated descriptions to optimize AI repair guidance.

     • Increasing KSampler steps for better detail generation.

   • If facial restoration looks unnatural, tweak ReActorFaceSwap parameters for improved alignment.

3. Compatibility

   • This workflow is compatible with ComfyUI 0.4+, and some nodes may require installation via ComfyUI-Manager.

This guide provides a detailed workflow for AI-based old photo restoration in ComfyUI! 🚀