Abel66999/stable_diffusion_prompts_instruct

Stable diffusion prompts for instruction models fine-tuning

Overview

This dataset contains 80,000+ prompts summarized to make it easier to create instruction-tuned prompt enhancing models. \ Each row of the dataset contains two values:

• —a short description of a image
• —a full prompt corresponding to that description in a stable diffusion format

Hope this dataset can help creating amazing apps !

How to use

You can download and use the dataset easily using the following code snippet:

from huggingface_hub import hf_hub_download

import pandas as pd

train_path = hf_hub_download(repo_id='groloch/stable_diffusion_prompts_instruct',
                             filename='train.csv',
                             repo_type='dataset')
test_path = hf_hub_download(repo_id='groloch/stable_diffusion_prompts_instruct',
                            filename='test.csv',
                            repo_type='dataset')

train_data = pd.read_csv(train_path).to_numpy()
test_data = pd.read_csv(test_path).to_numpy()

Generation

This dataset was partially generated by a LLM. \ It is based on Gustavosta/Stable-Diffusion-Prompts \ The short descriptions (x column) were generated using mistralai/Mistral-Nemo-Instruct-2407, more preciseliy one quantized (Q6KL) version found there: bartowski/Mistral-Nemo-Instruct-2407-GGUF. The script used for the generation can be found in the gen subfolder.

Disclaimer

Some values in the x column may contain hallucination or be ill-generated. The original dataset comes from scraped internet data and could contain bias towards certain styles.

Stats

Number of rows (total): 81,910 \ Number of rows (train): 73,718 (90.00%) \ Number of rows (test): 8,192 (10.00%)

Generation time: ~ 10 hours \ Generation hardware: 1 x AMD RX7900 XTX

License: Apache 2.0