Unlock the Secrets of Ghibli Anime Style with FLUX Technology

Workflow Overview

This is an advanced workflow that transforms ordinary images into Ghibli anime style using FLUX technology stack. It preserves original content while applying the distinctive colors and textures of Miyazaki's animations.

Core Models

1. Ghibli Style Main Model: "【动漫风格】_juaner-宫崎骏吉卜力工作室风格 flux_v2.0"

2. ControlNet Model: "FLUX.1-dev-ControlNet-Union-Pro-InstantX.safetensors"

3. Style LoRAs:

   • "复古旧漫_flux_1.0" (weight 0.6)

   • "明義浮梦_数字插画_F1&majicFlus_V2.0-F1" (weight 0.37)

   • "majicFlus复古美漫_V1" (weight 0.8)

4. PULID Model: "pulid_flux_v0.9.0.safetensors" (facial feature processing)

Key Components

1. ReduxAdvanced: Advanced downsampling

   • Downsample ratio 0.8

   • Uses area method to maintain aspect ratio

2. FaceDetailer: Facial detail enhancement

   • Resolution 768

   • 20 steps euler sampling

   • CFG scale 5

3. easy imageColorMatch: Color matching

   • Uses adain and reinhard algorithms

   • Applied at different processing stages

4. ModelSamplingFlux: FLUX sampler

   • Sampling ratio 1.03

   • Base resolution 1024×1024

5. ApplyPulidFlux: Facial feature application

   • Combined with InsightFace analysis

   • Uses EVA-CLIP processing

Workflow Structure

1. Model Loading Group:

   • Loads UNET, VAE, ControlNet and CLIP models

   • Applies multiple style LoRAs

2. Input Processing Group:

   • Loads input image (e.g. "49395318-...png")

   • Image scaling and aspect ratio adjustment

3. ControlNet Processing Group:

   • Depth processing (DepthAnything)

   • Pose estimation (Openpose)

   • Control strength 0.4-0.8

4. Style Conversion Group:

   • Applies Redux downsampling

   • Uses FLUX guidance (strength 3.5)

   • Ghibli style prompts

5. Output Optimization Group:

   • Facial detail enhancement

   • Color matching adjustment

   • Final image output

Inputs and Outputs

Input Parameters:

• Original image (recommended 768×1360)

• Random seed (220489138248147)

• Style prompt: "ghibli style,anime style"

Output Results:

• Direct Ghibli-style output

• Face-enhanced version

• Color-optimized version

• Original vs generated comparison

Notes

1. Requires at least 12GB VRAM

2. Includes multiple VRAM optimization nodes

3. Adjust Redux downsampling ratio (0.1-0.8) for style intensity

4. Facial LoRA weights adjustable (0.3-0.8)

5. Recommended input resolution near 1024×1024