Discover the Ultimate Eastern Art Creation Workflow with AI
Workflow Overview
This workflow consists of two modules (Online and Local), leveraging the Flux.1 model to generate high-quality images in an Eastern Pixar style. The themes include:
Online Module: A close-up shot of a Chinese little girl standing amidst pink and orange flowers, with a blurred background and depth-of-field effect.
Local Module: A female swordsman in red Hanfu standing by a window, with colorful lights and a blurred background. Workflow features and goals:
Style Enhancement: Uses the Lora model (“miluo丨光影物语丨东方皮克斯风格_v1.6”) to enhance Eastern Pixar style and lighting effects.
Resolution Support: Supports different resolutions (Online: 768x1024, Local: 768x1248).
Modular Design: Split into Online and Local modules for flexible use. This workflow is ideal for Eastern-style art creation, movie concept design, or digital illustrations.
Core Models
Flux.1 (miluo-flux-pro_模型配置更智能_v1.0)
Function: An efficient text-to-image model for high-quality artistic images.
Source: Download from Liblib.art (link: https://www.liblib.art/modelinfo/dde591404d9c439ea0aa780c8db6d3d1?from=personal_page&versionUuid=a2d8874ea62a4bb4a39d9b49b00b4ebd), place in ComfyUI/models/checkpoints/.
Lora Model
miluo丨光影物语丨东方皮克斯风格_v1.6: Enhances Eastern Pixar style and lighting, with weights of 1.0 (Online) and 0.4 (Local).
Source: Download from Liblib.art (link: https://www.liblib.art/modelinfo/140a3cf9394f44809085585473b100fc?from=personal_page&versionUuid=b3dedab2d8c04d82856e39a03915720e), place in ComfyUI/models/loras/.
VAE Model (ae.sft)
Function: Decodes latent images into pixel images for high-quality output.
Source: Download from Flux.1-related repositories, place in ComfyUI/models/vae/.
Component Explanation
Below are the key nodes in the workflow, their purposes, functions, and installation methods:
CheckpointLoaderSimple
Purpose: Loads the Flux.1 model.
Function: Outputs model, CLIP, and VAE objects.
Installation: Built into ComfyUI.
LoraLoader
Purpose: Loads the Lora model for fine-tuning.
Function: Enhances Eastern Pixar style and lighting effects.
Installation: Built into ComfyUI.
CLIPTextEncode
Purpose: Encodes positive and negative text prompts.
Function: Converts prompts into conditioning data for generation.
Installation: Built into ComfyUI.
ttN text and ttN concat
Purpose: Processes and concatenates prompts.
Function: Combines prompt segments (style, scene, effects) into a complete prompt.
Installation: Requires ttNpipe plugin (provides ttN text and ttN concat nodes), install via ComfyUI Manager or GitHub.
EmptyLatentImage
Purpose: Creates an initial latent image.
Function: Generates a custom-resolution latent space (Online: 768x1024, Local: 768x1248).
Installation: Built into ComfyUI.
RandomNoise
Purpose: Generates random noise.
Function: Serves as the sampling input with a seed of 2594433241217.
Installation: Built into ComfyUI.
BasicScheduler
Purpose: Calculates sampling steps and timesteps.
Function: Uses a “simple” scheduler with 30 steps.
Installation: Built into ComfyUI.
KSamplerSelect
Purpose: Selects the sampler.
Function: Outputs the “euler” sampler.
Installation: Built into ComfyUI.
SamplerCustomAdvanced
Purpose: Performs advanced sampling (Online module).
Function: Combines noise, guidance, and latent image to generate results.
Installation: Built into ComfyUI.
KSampler
Purpose: Performs standard sampling (Local module).
Function: Generates a latent image with 30 steps.
Installation: Built into ComfyUI.
BasicGuider
Purpose: Combines model and conditioning (Online module).
Function: Outputs a guider object.
Installation: Built into ComfyUI.
VAELoader
Purpose: Loads the VAE model (Online module).
Function: Outputs a VAE object.
Installation: Built into ComfyUI.
VAEDecode
Purpose: Decodes the latent image.
Function: Generates a pixel image.
Installation: Built into ComfyUI.
PreviewImage
Purpose: Previews the generated image.
Function: Displays the output in real-time.
Installation: Built into ComfyUI.
Workflow Structure
Online Module
Model Loading and Fine-Tuning Group
Role: Loads the Flux.1 model and applies Lora fine-tuning (weight 1.0).
Input Parameters: Model file (miluo-flux-pro_模型配置更智能_v1.0), Lora file.
Output: Fine-tuned model and CLIP object.
Prompt Processing Group
Role: Concatenates and encodes prompts.
Input Parameters: Prompt segments (“miluo_pks,pixar style”, scene description, depth-of-field effect).
Output: Encoded conditioning data.
Generation Preparation Group
Role: Creates latent image and sampling parameters.
Input Parameters: Resolution (768x1024), noise seed, sampling steps (30).
Output: Latent image and sampling conditions.
Sampling and Generation Group
Role: Executes image generation.
Input Parameters: Fine-tuned model, conditioning, noise, sampler (Euler).
Output: Latent image.
Decoding and Display Group
Role: Decodes and previews the image.
Input Parameters: Latent image, VAE model.
Output: High-resolution pixel image.
Local Module
Model Loading and Fine-Tuning Group
Role: Loads the Flux.1 model and applies Lora fine-tuning (weight 0.4).
Input Parameters: Model file, Lora file.
Output: Fine-tuned model and CLIP object.
Prompt Processing Group
Role: Encodes prompts.
Input Parameters: Full prompt (“miluo_pks,pixar style, A female swordsman...”).
Output: Positive and negative conditioning data.
Generation Preparation Group
Role: Creates latent image.
Input Parameters: Resolution (768x1248), noise seed (374229838361640).
Output: Latent image.
Sampling and Generation Group
Role: Executes image generation.
Input Parameters: Fine-tuned model, conditioning, latent image, sampling steps (30).
Output: Latent image.
Decoding and Display Group
Role: Decodes and previews the image.
Input Parameters: Latent image, VAE model.
Output: High-resolution pixel image.
Inputs and Outputs
Expected Inputs:
Resolution: Online 768x1024, Local 768x1248 (adjustable).
Seed: Online 2594433241217, Local 374229838361640 (randomizable).
Prompt: Online (“miluo_pks,pixar style, In a close-up shot, a Chinese little girl...”), Local (“miluo_pks,pixar style, A female swordsman...”).
Negative Prompt: Not set (suggest adding “Worst quality, blurry”).
Sampling Steps: 30.
Guidance Scale: 1.0.
Batch Size: 1.
Final Output:
Online: An Eastern Pixar-style close-up image of a Chinese little girl, PNG format, 768x1024 resolution.
Local: An Eastern Pixar-style image of a female swordsman, PNG format, 768x1248 resolution.
Notes and Tips
Resource Requirements: Generation requires 8GB+ VRAM; an NVIDIA GPU is recommended.
Model Files: Ensure miluo-flux-pro_模型配置更智能_v1.0 and ae.sft are in the correct paths.
Plugin Installation: Requires ttNpipe plugin (for ttN text and ttN concat nodes).
Module Selection: Online module is enabled by default; disable it and enable the Local module for local runs.
Prompt Optimization: Replace prompts as needed, maintaining structured descriptions (style, scene, effects).
Example Illustration
Online Run:
Input: Prompt “miluo_pks,pixar style, In a close-up shot, a Chinese little girl... depth of field, large aperture, background blurring”.
Output: An Eastern Pixar-style close-up of a Chinese little girl with flowers, blurred background, vibrant colors.
Local Run:
Input: Prompt “miluo_pks,pixar style, A female swordsman, dressed in red Hanfu... depth of field, large aperture, background blurring”.
Output: An Eastern Pixar-style image of a female swordsman in red Hanfu by a window, with warm lighting and rich details.