This is a dataset consisting of prompts from the top 50k most reacted and most commented images from CivitAI.
The only used images are images generated by models utilizing the T5 Text Encoder, thus mostly using natural prose or something close to that.
Those prompts were sanitized and the shorter ones removed. 150 of the most often reoccuring (quality tags, unnecessary names, etc...) tags have been removed. This resulted in 7.5k images and prompts in total.
For those 7.5k images, short natural… See the full description on the dataset page:
https://huggingface.co/datasets/Aconexx/CivitAI-Flux-Prompts.
