Anime Screenshot Pipeline

Semi-automatic pipeline to extract image training set from anime for generative model training.
Colab version of: https://github.com/cyber-meow/anime_screenshot_pipeline

Feature

• Can be found at cyber-meow github repo README.md

Limitation

• Not all code tested
• Not all steps from github repo is implemented
• Since this notebook is combination of many steps, dependency or package conflict may occurs
• Bad english language spelling and grammar (english is not my primary language :#). Feel free to correct and Pull Request!

Credits

• Author
• TheSkinnyRat
• Base Code Repo
• cyber-meow github repo
• Colab Template and Reference
• Linaqruf/kohya-trainer
• Code Assistant
• OpenAI ChatGPT
• Original cyber-meow repo credits
• This is a collection of many resources found on internet (credit to the orignal authors), and some python code written by myself and ChatGPT.

Whats new?

• (02/27/23):
• Add WANDB_MODE=disabled to [7.2] Begin Training
• Fix cc_training_test_set error in [7.2] Begin Training
• Fix fa_move_file error in [8.1] Only keep images with faces and resize
• Move [6.3] Data Cleansing to [5.2] Data Cleansing and change to n_faces instead n_poeple
• (02/26/23):
• Initial First Release 🎉

Google Colab

Tested on Google Colab Free.

• GPU: Tesla T4
• Python: 3.9.16
• Pip: 22.0.4
• Torch: 1.13.1+cu116
• Torchvision: 0.14.1+cu116

Notebook Name	Link
Anime Screenshot Pipeline

Runpod.io

Tested on Runpod.io [TSR Pytorch docker image].

• theskinnyrat/pytorch:3.10-1.13.1-117
• https://runpod.io/gsc?template=97cv87yo27&ref=8rg64ew2
• Python: 3.10.10
• Pip: 23.0.1
• Torch: 1.13.1+cu117
• Torchvision: 0.14.1+cu117

Link