Some python scripts (written with the help of chatgpt) that I use to prepare LoRA (Low Rank Adaptation) training for anime characters.

• 1_rename_mkv.py : it renames the titles of mkv episode to "01", "02" and so forth
• 2_ffmpeg_extr_jpg.py : extract frames from mkv with ffmpeg tool
• 3_delete_duplicate_images.py : pip install imagehash pillow psutil
• 4_detect_person_or_delete.py : it loads yolov8x6_animeface model in order to check if AI can find a "face" inside loop of images in a folder. Deleting any image that doesn't contain "anime face" in the frame to save disk space.
• 5_character_categorisation.py : https://github.com/KichangKim/DeepDanbooru/releases
• 6_batch_rename_resize_convertojpg.py
• 7_Comfyui_CAPTION_TXT.json : workflow for comfyui to caption every image from folder

GPU ACCELERATION : For gpu acceleration, use miniconda : https://docs.anaconda.com/miniconda/ PYTORCH CONDA GUIDE : conda create -n ptorch_env python=3.9 conda activate ptorch_env conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia pip install numpy opencv-python ultralytics

TENSORFLOW CONDA GUIDE : conda create -n tflow_env python=3.10 conda activate tflow_env conda install cudatoolkit=11.8 cudnn=8.6 -c conda-forge pip install tensorflow==2.10.0 pip install numpy pillow tensorflow-io deepdanbooru psutil

check TensorFlow GPU : python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"