I usually do LoRAs I don't find on the site that I want myself, but Cattleya already has another LoRA you can check out here! I just simply forgot to check this time around and figured I would still publish it, now that I did the work already.
If you want to support me buy me a coffee!
Updated it to fix the LoRA always turning into a similiar style after a certain weight.
Weight: 0.7-1
Triggerwords: cattleyav2 for version 2.0, cattleya for version 1.0
Notes:
If you want to see her in different outfits, you should probably negativeprompt bodysuit and swimsuit and adjust their weight. They will probably still often be purple.
The higher the weight, the more accurate it looks to the actual show in v 1.0, at least from my testing. It's mostly fixed in v 2.0. I left it available for download if anyone wants that specifically.
Feel free to share some of the pictures you create with me and post them here!
Description
FAQ
Comments (4)
Thank you for making this Lora! If possible, could you share the training data you used or your workflow when making Loras?
I think I got rid of the dataset already, sorry but as for my process I can try and outline it, but I'm not professional so there might be better ways to do this:
I usually scrape imageboards for pictures of characters, if there's not enough pictures (usually around 50 is good minimum) I pick them out manually from other sources (google, screencaps, pixiv)
Then I usually crop them on birme.net and remove elements with paintnet that might affect the quality of the images (removing text from speechbubbles, censorship, etc)
For tagging I use WD14, which tags images using imageboard tags, make sure to blacklist tags that you do not want in your lora (for example monochrome and greyscale if you train on manga images mostly and try make a lora capable of colors)
As for the actual training the dataset, you can use kohya SS locally if your GPU is good enough or use a google colab to run it with google's GPUs! (There are a few, you can look around online, one I can recommend is https://github.com/hollowstrawberry/kohya-colab )
Going into the actual trainingsettings here would be a bit too much I feel like, but there are great guides on that as well.
This is more of a basic guide and not really indepth, but I hope it still helps.
@Beelze Thank you so much for such a detailed explanation! It definitely helps!
@Beelze 💀🙏😎🤙♥