Trained on everything from CoolerINKER (also known as Inker-Shike)'s newgrounds to capture their style. Trained on AnyLora.
If you like this model, please support the original artist. If you are the original artist and don't want this to be public, contact me please.
This is my first lora, so I'm not sure what else to say...
22.04.2024 - Updated to PonyXL. Same dataset. Used the on-site training this time and tinkered with the parameters a bit.
14.02.2025 - Updated to IllustriousXL. Different dataset, different parameters, still on-site training. Consider putting constricted pupils, sketch in the negatives if you experience the corresponding phenomena.
22.07.2025 - Note: it appears the trigger word can have a side effect on some models, where a cooler (a box that keeps food frozen) appears in the images. Add cooler to the negative if you experience that. Sorry for the inconvenience. Back when I made the dataset for this LoRA, base models didn't really know objects well enough to present this problem.
21.07.2026 - Updated to Anima. This time a separate model, please go here to find it.
If you like the style, please support the original author
Description
Illustrious update. Also changed the training method (same as my recent models) and the dataset (gelbooru with original tags instead of newgrounds with autotags). I'm quite happy with the result.
FAQ
Comments (11)
Awesome style!
Very good model, very faithful to the style represented, but a problem occurs to me. In the background of the images it generates, it includes a refrigerator/cooler, and the background is usually light blue. I think it's because of the trigger and the model name includes that word. If anyone knows how to avoid this problem, please let me know
Yes, I noticed that recently too. To me, just putting cooler in the negatives helps
any chance for anima update?
Most likely, at some point. I was working on an anima update for my Katawa Shoujo charpack, but I got caught in a heatwave (I have no AC) and it got very hard to get myself to do anything productive 😅. I'll get to it though. In the meantime, you can get okay results with the @inker comics tag
Okay, wish granted. I decided to reprioritize a bit, seeing how it's taking a while to make the perfect toolkit for the kind of dataset prep I have in mind for the charpack
@Abel1502 I am not experimented with training lora and stuff, but I wonder is easier or harder train a lora for anima unlike illus
@EzDeath_ About the same, I'd say. I guess the only difference is, people have less experience about what makes a good training config, but I'm kinda just going with the first thing that works. I might go back and retrain stuff even better at some point in the future
please anima
Literally done it already -- look at the first suggested resource 🤣