ComfyUI Prompt Flux/SD3 Generator
This workflow offers a completely self-contained solution for generating and optimizing prompts for both CLIP-L and T5XXL, running entirely within ComfyUI without requiring any external APIs or Ollama.
Key Features
Dual Input Methods: Process images using Florence-2 Vision Language Model or input custom prompts manually
Fully Local Processing: All language models run directly within ComfyUI
Zero External Dependencies: No need for API keys or Ollama setup
Technical Architecture
The workflow utilizes three locally running Mistral-7B-Instruct models in sequence to:
Generate detailed descriptions
Optimize prompt structure
Format specialized outputs for CLIP-L and T5XXL
Components
Florence2 custom nodes for image analysis
Searge-LLM custom nodes for running Mistral-7B locally
Custom prompt styling and optimization scripts
Integrated CLIP-L and T5XXL output formatters
Searge LLM Installation Guide
Create directory:
mkdir ComfyUI/models/llm_gguf
File: Mistral-7B-Instruct-v0.3.Q4_K_M.gguf (4.37 GB)
Place in: ComfyUI/models/llm_gguf/
Description
FAQ
Comments (8)
amazing, almost as good as mine😜😅, does running it locally mean mistral is uncensored or do you run in to that censorship thing also?
The vision model of Florence_2 is censored.
However, Mistral is uncensored. If you set the switch to "2," you can also create uncensored prompts using the manual prompt option.
@denrakeiw hmm I can run Mistral 7B 0.3 on my ollama aswell, ima try that next
Hi, the workflow tells me that it can't find "SDXLPromptStylerbyMileHigh", can you give me the link to download the node? Thanks
@denrakeiw Thank you!
Is it possible to use this workflow to generate prompts for other models like Qwen Edit etc?