## ComfyUI Workflow: LLM-Driven Image Generation with Flux

This advanced ComfyUI workflow combines free language models, prompt enhancement, efficient image generation, and automatic uploading to create a powerful and user-friendly AI art generation pipeline.

## Prerequisites

1. **API Keys**:

  - OpenRouter LLM API key (free)

  - ImgBB API key (free)

2. **Custom Node**:

  - Install the custom SaveImage node for ImgBB uploading: https://github.com/revirevy/Comfyui_saveimage_imgbb

3. **ComfyUI Installation**:

  - Ensure you have ComfyUI installed and set up on your system

4. **Additional Custom Nodes**:

  - Install the following custom nodes:

    - ComfyUI_experiments

    - ComfyUI-ModelDownloader

    - ComfyUI-SUPIR

    - KJNodes for ComfyUI

    - Klinter_nodes

    - pythongosssss/ComfyUI-Custom-Scripts

    - ReActor Node for ComfyUI

    - Various ComfyUI Nodes by Type

## Required Nodes

### Primitive Nodes

1. CLIPTextEncodeFlux

2. FluxPromptEnhance

3. FluxPromptGenerator

4. ImgBBUploader

5. LLM_prompt_generator

6. OllamaTextDescriber

7. Power Lora Loader (rgthree)

8. Primitive string [Crystools]

9. Primitive string multiline [Crystools]

10. Seed (rgthree)

11. Show any [Crystools]

### Custom Nodes

1. **ComfyUI Core Nodes**:

  - SaveImage

  - CheckpointLoaderSimple

  - DualCLIPLoader

  - VAEDecode

  - KSampler

  - VAELoader

  - CLIPTextEncode

  - KSamplerAdvanced

2. **Additional Custom Nodes**:

  - AIO_Preprocessor (ComfyUI Nodes for Inference.Core)

  - ModelSamplerTonemapNoiseTest (ComfyUI_experiments)

  - TonemapNoiseWithRescaleCFG (ComfyUI_experiments)

  - ModelDownloader (ComfyUI-ModelDownloader)

  - SUPIR_model_loader_v2_clip (ComfyUI-SUPIR)

  - ColorMatch, Superprompt, JoinStringMulti, EmptyLatentImagePresets (KJNodes for ComfyUI)

  - concat (Klinter_nodes)

  - ShowText|pysssss, MathExpression|pysssss, StringFunction|pysssss (pythongosssss/ComfyUI-Custom-Scripts)

  - ReActorRestoreFace (ReActor Node for ComfyUI)

  - JWDatetimeString (Various ComfyUI Nodes by Type)

## Workflow Features

- **Prompt Generation**: Utilizes free Ollama service and OpenRouter API for LLM-based prompt generation

- **Multiple Prompt Generation**: For each LLM API call, the workflow generates 5 interesting random prompts

- **Iterative Processing**: The workflow loops through each of these prompts, processing them individually

- **Prompt Enhancement**: Improves generated prompts for better results

- **Efficient Processing**: Uses Lightning Flux for as few as 4-5 steps

- **Image Uploading**: Automatically uploads generated images to ImgBB

- **Improved CLIP_L Prompting**: Enhanced prompting for better image generation

- **Noise Correction**: Includes logging and noise correction features

## Model Details

- **Checkpoint**: fluxArtfusion4Steps_v12.safetensors

## Workflow Process

1. **Initial LLM Call**: The workflow starts by calling the LLM API (either Ollama or OpenRouter).

2. **Prompt Generation**: The LLM generates 5 interesting and random prompts based on the initial input or parameters.

3. **Looping Process**:

  - The workflow enters a loop, processing each of the 5 prompts individually.

  - For each prompt:

    a. The prompt is enhanced using the FluxPromptEnhance node.

    b. The enhanced prompt is fed into the image generation pipeline.

    c. The image is generated using the Lightning Flux model (4-5 steps).

    d. Noise correction and other post-processing steps are applied.

    e. The resulting image is uploaded to ImgBB.

4. **Output**: After processing all 5 prompts, the workflow will have generated and uploaded 5 distinct images.

## Usage Tips

- Ensure all required custom nodes are properly installed before running the workflow.

- Double-check that your API keys for OpenRouter and ImgBB are correctly entered in the appropriate nodes.

- Experiment with different prompt inputs to see how they affect the final output.

- Use the logging feature to understand the workflow process and identify any areas for optimization.

- Consider fine-tuning the noise correction parameters if you notice any artifacts in the generated images.

- Take advantage of the multi-prompt feature by reviewing all 5 generated images from each API call to select the best results.

- Consider adjusting the number of prompts generated per API call if you want more or fewer options (note that this may require modifying the workflow).

- Pay attention to how different prompts from the same API call can lead to varied results, which can be useful for understanding the LLM's creative range.

This workflow combines cutting-edge AI technologies with user-friendly features, making it an excellent tool for both beginners and advanced users in the field of AI-assisted image generation. It leverages the power of multiple prompt generation to provide a diverse set of outputs from a single LLM API call, combining efficiency with creativity and allowing users to explore a range of possibilities from each interaction with the language model.