Thanks for checking out my first LoRA. If anyone has any tips on on-site training, please leave a comment. Enjoy!

10•26•24

Updated sample images to provide a better apples-to-apples comparison between styles. Each is weighted higher than I would typically use them to exaggerate some of the differences, so I’d recommend tuning them down as you mix and match with other LoRAs.

Expressive 2D

Use it for flat, expressive anime characters. Generates everything from sketches, hand-painted illustrations, and individual anime stills.

Liquid Love

This is my second style. Use this style to generate semi-realistic anime images with a more emotionally expressive style.

Another use case for this style is to enhance hi-resolution fixes with Real Dream to turn cartoon, anime, or semi-realistic images into photorealistic images. To do this, find an image in your gallery (for best results, choose an image that has the most realistic proportions), tap on the magic wand icon in the lower left corner of the thumbnail image, replace your checkpoint with Real Dream, and replace all LoRAs with Liquid Love.

I also recommend maxing out your steps to 60 (the difference between 30 and 60 steps in buzz is only 1-2 buzz per image), and I’ll usually generate 2 copies and post the best one. Let me know how it works out for you.

etheREAL

This is my first LoRA, and it plays really well generating semi-realistic images. Use it to create emotionally expressive characters from different angles. It also plays really nicely with hi-res fix if you want to add a layer of photo realism to it. It has a bias towards dramatic lighting, which really makes images pop.

Teknikolor

I put together 60 images of cyberpunk noir images for this LoRA. I recommend tuning this down a bit in the 0.10-0.30 range to start and see what you come up with. It’s meant to work well with neon and dark backgrounds.