| --- |
| 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](https://huggingface.co/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: |
|
|
| 1. π delete some meanless words like some artists name with misspelling |
| 2. π delete many spaces that make `100mm` to `10 0m` |
| 3. π some url in datasets |
| 4. π and many unknown words |
|
|
|
|
| And this version is doing well in my train test!π |
|
|