<h2 id="heading-710">SDXL PEPE LoRA</h2>Building on the previous and most excellent work of Dr. Diffusion's Pepe SDXL LoRA (<a target="_blank" rel="ugc" href="https://civarchive.com/models/161611/pepe-xl-lora">https://civarchive.com/models/161611/pepe-xl-lora</a>), I decided to try an expanded SDXL Pepe LoRA. I added 469 new dalle3 Pepe images/captions to Dr. Diffusions original dataset for a total of 569 (nice) images. Trained for 50 epochs using the prodigy optimizer with a batch size of 4. Images were captioned in a similar style to the original DD model and all of the original tags should work but I added some new tags including Lego Minifig, CCTV, digital art, family tree, print, game cartridge, anime, girl pepe and N64 render.

BS-Pepe-V1

Version one of the Blaire Silver SDXL Pepe model. Sample images were generated in comfy with automaticCFG and perturbated attention guidance using the tempest artistic model. One issue with this model is that i used horizontal flip augmentation so text gets generated poorly.

Files with SHA256: 139c8668c4b37af3297b9affe26899f93eefed7f9250eed008e0cfee471c2537