Works with XL 0.9 Base, XL 1.0 Base, & XL 1.0 Refiner Models.
Embrace chaos with this Low-Rank Adaptation that injects abstract mixed colors and textures to your images.
The dataset consists of 85 highly detailed photographs of my own experimental paintings that have been captioned with BLIP and corrected manually.
Description
Works without a trigger but "Abstract Painting" helps.
JKAP - Stands for "Joe Kachnic Abstract Paintings".
Joe Kachnic is the artist who created the paintings used in the data set.
FAQ
Comments (10)
i love art , im gonna check it , it seems pretty nice
So, just for clarity: this is (as listed) a LORA which happens to be almost 2GB in size?
It is in fact a LoRA safetensors file that was the 7th checkpoint about 10 hours into the training.
@doctor_diffusion Thanks. I'm just not familiar enough with things related to SDXL 0.9 to be sure what I should consider reasonable filesizes for things. If model extractions are consistently this large it's going to be interesting.
can you share kohya parameters? thx so much
Most of the same settings as the "SECourses" 0.9XL LoRA video but with 85 photos and no regulators.
@doctor_diffusion super THX
1.7 gigs for a Lora? Did you train the dataset in 8000x19000? Jfc.
Well... 1024x1024 on 256/1 Dim/Alpha, what do you expect?
Interresting: This has a tendency to produce very "circular" images 9 out of 10 images I generate with the lora either on the base model or on both base and refiner have a promintent circular feature in the center, even if no "abstract painting and/or jkap" is applied to the prompt, and the image is not rendered as an abstract painting either
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.