---
license: mit
task_categories:
- text-to-image
- image-to-text
tags:
- comfyui
- stable-diffusion
- prompts
- vision
size_categories:
- n<1K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: image
    dtype: image
  - name: comfyui_prompt
    dtype: string
  splits:
  - name: train
    num_bytes: 886307.0
    num_examples: 1
  download_size: 892784
  dataset_size: 886307.0
---

# ComfyUI Vision-to-Prompt Dataset

This dataset contains raw images paired with exceptionally detailed prompts optimized for text-to-image generation in ComfyUI and Stable Diffusion. 

The prompts were generated using Llama 3.2 Vision by meticulously analyzing the core subjects, art styles, lighting, and camera composition of the original images.

## Dataset Structure

This dataset uses the standard Hugging Face `ImageFolder` format:
* **Images** are stored in the `data/` directory.
* **Prompts** are linked to the images via the `data/metadata.jsonl` file.

Because of this structure, you can view the images and their corresponding prompts side-by-side in the Dataset Viewer at the top of this page.