Here's another single-image Lora. Her in-game portrait shows her with red hair, and there's only one illustration showing her full body. I think illustrious is truly remarkable in that it somehow managed to learn, but there are inevitably some inaccuracies in the details.
Default outfit : parole_zenonia, blue hair, purple hair, gradient hair, long hair, black eyepatch, black eyes, earrings, red demon horn, black demon wing, ragged_leather_string_bikini_top, leather_highleg_bikini_bottom, leather bracer, black boots, leg wrap, ragged_leather_thigh_guard,
Description
FAQ
Comments (4)
You can make a SINGLE IMAGE LORA?!
How does that even work? Great OP images btw
Since most retro games only have a few illustrations, I often try to increase the number of images by cropping and rotating them here and there. By lowering the DIM as much as possible and adding a lot of noise, I can somehow create a LORA, but honestly, the quality isn't great.
Still, LORA training has become much easier than it was a year or two ago, so I expect that soon we'll see a new model that can sufficiently understand a character with just a single illustration.
@maomo314 Thanks for sharing that. I was wondering what other people are doing for characters with very little art.
I'm working on a Growlanser 2 character LORA and she only has like 3-4 images out there. I made a crappy LORA to make tons of ai images of her. I'm going to use the results that are good enough to make a new LORA.
@FondantViper Isn't that an illustration by Satoshi Urushihara? A character with such a distinct style and intricate details would be difficult to work with using my clumsy methods. I, too, constantly struggle with insufficient datasets. I hope you find a good solution.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.