After, 2 days (slow I know). I now have a fully funtioning e621, danbooru, and rule34.us tag list downloader. Which after, the tags are downloaded. About 5-15 samples based on said tag is downloaded as reference. The tagger than self updated itself which with my gpu will be very slow lol
After, the training is done (which should mostly just be comparing images for similarities) to create a custom checkpoint model. Than, the checkpoint model would be used along the .py and .bat to accurately tag images and hopefully character characters correctly tagged which smilingwolf and a few others mix up from time to time.
Be aware this is just a private project and not much worth sharing. Since, this project is for testing and personal research. That being said I might test the new model on requested tagging commissions to see how well the tagger holds up.
I'd try to create a captioning model. But, I don't have enough vram to even begin a 8-12vram program on 2gb of vram -.-