metadata

license: cc0-1.0
task_categories:
  - text-generation
  - text-classification
language:
  - en
tags:
  - code
  - art
pretty_name: mg_sd_washed
size_categories:
  - 10K<n<100K

MagicPrompt_SD_Washed

It's a version of datasets of Gustavosta/MagicPrompt-Stable-Diffusion.

When I want to train a model using origin data, some bad prompts broke model and waste many time.

So I washed the origin datasets:

😄 delete some meanless words like some artists name with misspelling
😂 delete many spaces that make 100mm to 10 0m
😭 some url in datasets
😭 and many unknown words

And this version is doing well in my train test!😍