Impressed by the newest submit about some actually cool LoRAs by Araminta, I acquired again into coaching my very own LoRA which I had not accomplished since a while.
The aesthetics of that is based mostly on black and purple tones, creating excessive distinction photographs with silhouette and reflections of the primary topics. Therefore the identify Rouge Noir which is french for Pink Black.
I had collected my dataset very long time in the past and was utilizing 50+ photographs within the unique model coaching however my poor RTX4080 couldn’t deal with such dimension after which I explored Runpod to coach my photographs which was good however I didn’t prepare this mannequin as an alternative one other mannequin which I educated on Octane Render token leading to prime quality 3d render photographs.
Nonetheless seeing Araminta’s work it re-ignited my ardour to coach particular model in SDXL. I up to date my Kohya_ss and resumed coaching my LoRA mannequin. Within the remaining coaching which took a number of makes an attempt and plenty of analysis testing I solely used 26 various photographs. The ultimate run I used some ideas shared by Araminta on this submit to additional refine my coaching settings. The outcome was a a lot cleaner coach model Rouge Noir for SDXL.
I’m additionally contemplating coaching this for SD 1.5 mannequin however let’s see the way you inventive folks like my LoRA first.
Listed below are just a few extra pattern outcomes from this educated LoRA mannequin. You’ll be able to obtain it right here.
These pattern had been created utilizing Base SDXL and different fine-tuned SDXL fashions that I take advantage of in my workflow.
That is personally one in all my favorite LoRAs I’ve produced to date and I hope you’ll take pleasure in creating some enjoyable photographs like I’ve. Don’t hesitate to go away your questions or feedback under.
If you would like to help our website please think about shopping for us a Ko-fi, seize a product or subscribe. Want a quicker GPU, get entry to quickest GPUs for lower than $1 per hour with RunPod.io