Workflow Overview

This ComfyUI workflow is designed to generate movie storyboard images in the style of Hayao Miyazaki’s anime using the Flux.1 model with multiple LoRA enhancements. It supports extracting prompts from a detailed story script and cropping the resulting images into multiple storyboard sections. Key functionalities include:

1. Prompt Processing: Extracts and concatenates storyboard prompts from user input.

2. Image Generation: Produces Miyazaki-style images using Flux.1 and LoRA models.

3. Image Cropping: Splits generated images into multiple storyboard regions.

Core Models

1. Flux.1:

   • Function: Advanced diffusion-based image generation model.

   • File: 基础算法_F.1 (manually download and place in models/unet).

2. VAE:

   • Function: Decodes Latent representations into final images.

   • File: ae.sft (place in models/vae).

3. LoRA Models:

   • FLUX_宫崎骏漫画风_v1.0: Enhances Miyazaki manga style.

   • 吉卜力动漫画风风格-_宫崎骏风格奇幻场景_V1: Boosts fantastical scene rendering.

   • In-Context-LoRA_电影分镜头生成: Optimizes movie storyboard generation.

   • Source: Download from Civitai or Hugging Face.

Component Explanation

1. UNETLoader (ID: 12):

   • Purpose: Loads the Flux.1 model.

2. DualCLIPLoader (ID: 11):

   • Purpose: Loads Flux-specific CLIP models (t5xxl_fp8_e4m3fn and clip_l).

3. LoraLoader (ID: 27, 102, 160):

   • Purpose: Loads LoRA models for style enhancement.

4. CLIPTextEncode (ID: 6, 170):

   • Purpose: Encodes text prompts into conditioning vectors.

5. SamplerCustomAdvanced (ID: 13, 173):

   • Purpose: Performs advanced sampling to generate Latent.

6. VAEDecode (ID: 8, 171):

   • Purpose: Decodes Latent into images.

7. ImageCrop+ (ID: 127-129, 179-181):

   • Purpose: Crops images into three regions (top, center, bottom).

8. Text Concatenate (Multiple Nodes):

   • Purpose: Combines multiple storyboard prompts.

9. SaveImage (Multiple Nodes):

   • Purpose: Saves generated images.

Workflow Structure

1. Prompt Processing Group :

   • Nodes: TextInput_ → CR Text Concatenate → Text Concatenate (JPS).

   • Role: Extracts and combines storyboard prompts from the story script.

   • Output: Complete prompt string.

2. Model Loading Group :

   • Nodes: UNETLoader → multiple LoraLoader.

   • Role: Loads Flux.1 and LoRA models.

3. Sampling Group :

   • Nodes: EmptyLatentImage → SamplerCustomAdvanced → VAEDecode.

   • Role: Generates 1024x1536 images.

   • Output: Full image.

4. Storyboard Cropping Group :

   • Nodes: ImageCrop+ → SaveImage.

   • Role: Crops images into three 1024x512 regions (top, center, bottom).

   • Output: Cropped storyboard images.

5. Preview Group :

   • Nodes: SaveImage.

   • Role: Saves full images for preview.

Inputs and Outputs

• Inputs:

  • Prompt: Story script from user input (see TextInput_ ID: 121), including character settings, scenes, and storyboard descriptions.

  • Resolution: 1024x1536 (set by EmptyLatentImage).

  • Seed: Randomized (via Seed node ID: 188).

• Outputs:

  • Full Image: 1024x1536 PNG file.

  • Cropped Images: Three 1024x512 storyboard images (top, center, bottom).

Notes and Considerations

1. Model Preparation: Ensure Flux.1, VAE, and LoRA files are correctly placed.

2. Plugin Requirements: Install Flux support plugins (e.g., ComfyUI-Flux) and cropping plugins (e.g., ImageCrop+).

3. Performance: Requires a high-performance GPU (12GB+ VRAM recommended).

4. Prompt Format: Storyboard prompts should follow the [SCENE-X] format, input line-by-line.