Trigger word is: anomaly
Description
This version trained on SDXL Base 1.0
FAQ
Comments (5)
Looks amazing! I love glitch art, but there is so much falling under that category with so much just not being the right kind of glitching I find aesthetically pleasing. But this LoRA looks promising in all regards! Thank you already in advance!
Did you train it unconventionally btw? The file size seems a little off for usual LoRAs.
You didn't glitch it by actually cutting out code, did you? :D
I trained the LoRA with onetrainer on a mix of glitch, psychedelic and surreal art. Then I used the block merge node in comfy to find the blocks of the LoRA that have an effect on the output. Then I merge into the base model, and use kohya to extract the final LoRA at a lower rank that was used in training. This way you only get the blocks needed for the effect and the final file is smaller.
The output from the LoRA changes depending on the base model and how strong the tokens in the prompt are compared to what it was trained on. Anything from grainy analog film with chromatic aberration to kaleidoscope effect to just complete noise with glitch artifacts. So you might need to turn the weight up or down depending on your prompt.
@humblemikey Wow, thank you very much for the comprehensive answer! I had no idea it was "distilled" like that. Very curious to testing it with that in mind, good range of interesting glitch-art-adjacent styles, too. Thank you!
Mikey, this is such a fun glitch effect. Even if no glitch lines/blocks appear, the LoRA adds really interesting effects. Fantastic work, as always!
Awesome style!