1. Workflow Overview

This workflow specializes in high-resolution image upscaling with detail restoration, featuring:

• Tiled Processing: Splits images into tiles (TTP_Image_Tile_Batch) to prevent VRAM overload.

• Smart Redraw: Uses ControlNet and prompts to enhance details during upscaling.

• Multi-Model Pipeline: Combines FLUX.1-dev for generation and 4x_NMKD-Siax for upscaling.

• Comparison Tool: Side-by-side preview of original vs. upscaled images.

2. Core Models

3. Key Components

4. Workflow Structure

Group 1: Input & Preprocessing

• Inputs: Source image (e.g., ComfyUI_00043_.png).

• Outputs: Tiled image data.

• Key Nodes: LoadImage, TTP_Tile_image_size.

Group 2: Tiled Redraw

• Inputs: Image tiles + optional prompts.

• Outputs: Enhanced tiles.

• Key Nodes: SamplerCustomAdvanced, FluxGuidance.

Group 3: Assembly & Output

• Inputs: Processed tiles.

• Outputs: Final HD image (4K+).

• Key Nodes: TTP_Image_Assy, VAEDecodeTiled.

5. Inputs & Outputs

• Inputs:

  • Source image (drag-and-drop).

  • Optional text prompts (auto-generated if empty).

  • Tile size (default: 1024x1024).

• Outputs:

  • Upscaled image (saved to ComfyUI/output).

  • Comparison preview (Image Comparer).

6. Notes

1. VRAM: 16GB+ recommended. Reduce tile size (e.g., 768x768) for lower usage.

2. Models:

   • Manually download flux1-dev-fp8 and 4x_NMKD-Siax.

3. Plugins:

   cd ComfyUI/custom_nodes
   git clone https://github.com/ssitu/ComfyUI-TTP-Toolset.git
   git clone https://github.com/Extraltodeus/ComfyUI-JoyCaption.git

4. Optimization: Adjust denoise (default: 0.1) to balance detail/speed.