FLUX ControlNet: The Ultimate Tool for Copyright-Free Image Generation

Workflow Overview

This workflow uses FLUX ControlNet to redraw images, primarily aiming to regenerate style and content through AI technology while avoiding copyright issues. It preserves the original composition and structure while creating completely new visual effects.

Core Models

1. Base Algorithm_F.1: Main model "基础算法_F.1" for image generation

2. FLUX ControlNet: "FLUX.1-dev-ControlNet-Union-Pro-InstantX.safetensors" for structure control

3. T5XXL Text Encoder: "t5xxl_fp8_e4m3fn" for text processing

4. Meta-Llama-3.1-8B: Image captioning model "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit"

Key Components

1. ControlNetApplyAdvanced: Advanced ControlNet application

   • Strength 0.3, start step 0, end step 0.6

   • Uses depth control type

2. Joy_caption_two: Automatic image captioning

   • Uses Meta-Llama for detailed descriptions

   • Outputs "Descriptive" and "long" format

3. FluxGuidance: FLUX guidance node

   • Guidance strength 3.5

   • Optimizes generation quality

4. ImageScaleByAspectRatio V2: Image scaling

   • Uses lanczos algorithm

   • Maintains original aspect ratio

5. Image Comparer: Image comparison tool

   • Displays original and generated images side-by-side

   • Supports sliding comparison

Workflow Structure

1. Model Loading Group:

   • Loads UNET, ControlNet, CLIP and VAE models

   • Sets model precision to fp8_e4m3fn

2. Input Processing Group:

   • Loads input image (e.g. "image (8).png")

   • Adjusts image size (824×1024)

3. Caption Generation Group:

   • Automatically generates image descriptions

   • Includes English-Chinese translation

4. ControlNet Application Group:

   • Applies FLUX ControlNet

   • Sets control parameters

5. Generation Output Group:

   • Uses KSampler for generation (25 steps, euler sampler)

   • Decodes and outputs final image

Inputs and Outputs

Input Parameters:

• Original image (recommended 1400×933)

• Automatically generated description

• Random seed (140685274328837)

Output Results:

• Redrawn image

• Comparison view of original and generated images

Notes

1. Requires at least 8GB VRAM

2. Includes VRAM cleaner node (PurgeVRAM)

3. Adjustable ControlNet strength (0.3-0.6)

4. Recommended to use highly descriptive original images

5. Output resolution automatically adapts to input aspect ratio