Bring New Life to Old Photos: Advanced Damage Repair and Super-Resolution
1. Workflow Overview

This workflow specializes in old photo restoration and enhancement, featuring:
Damage repair: Fix cracks, scratches, and fading.
AI colorization: Convert B&W photos to color.
Super-resolution: Recover details from low-quality scans.
Prompt-based control: Customize styles (e.g., "vintage grain" or "modern HD").
2. Core Models & Tech
Component/Model | Function |
---|---|
FluxKontextPro | Core restoration model (damage repair + color correction). |
LibLibTranslate | Auto-translates prompts (e.g., Chinese → English). |
Image Comparer | Side-by-side comparison tool (slider mode). |
3. Key Nodes & Installation
Node Name | Function | Installation |
---|---|---|
FluxKontextProImageNode | Main restoration/colorization node. | Requires |
LibLibTranslate | Translates user prompts. | Via ComfyUI Manager. |
Image Comparer | Visual comparison of results. | Install |
Dependencies:
Download Flux-Kontext models to
models/flux_kontext/
.Recommended VRAM ≥6GB (≥8GB for high-res photos).
4. Workflow Structure
Input Group:
Upload old photo (e.g.,
WeChat screenshot.png
).Enter prompts (e.g., "correct colors, enhance sharpness").
Processing Group:
FluxKontextPro: Executes repair (default 3 steps).
Aspect Ratio: Matches original dimensions (e.g., 3:4).
Output Group:
Preview restored image (
PreviewImage
).Generate comparison (
Image Comparer
).
5. Inputs & Outputs
Inputs:
Required: 1 old photo (e.g.,
WeChat_20250604.png
).Optional: Custom prompts, seed (e.g.,
1103634540819029
), aspect ratio.
Outputs:
Restored HD image (original dimensions preserved).
Comparison slider (before/after).
6. Notes
Quality Tips:
For complex damage, use detailed prompts (e.g., "repair torn edges + fill missing forehead").
Colorize B&W photos with terms like "natural skin tones" or "1950s palette".
Hardware:
Use RTX 3060+ for resolutions >2000px.
Troubleshooting:
Color shifts: Add prompts like "preserve original white balance".
Over-sharpening: Reduce FluxKontextPro steps (default: 3).