Mary Moody (Born in Aug/23/1992) is a camgirl from US, Los Angeles, CA
Preview images created using Dreamshaper SDXL Turbo and TurboVisionXL
Add the lora and also "m4rym00dy woman" like in the example prompts.
Try LoRa weights between 0.6 and 1
SDXL is far from photorealism compared to SD1.5 fine tuned models, but we are getting there!
Description
FAQ
Comments (5)
SDXL is far from photorealism compared to SD1.5 LOL Oh Ok.
Sad but true. SDXL has more detail, definition and base resolution, but train LoRAs is a pain, the result is far from SD1.5, where we have also lots of fine tuned models, plugins and works good with many different samplers. SDXL still needs some more time to mature. I learned that the hard way, trying to training models and Loras and failing a lot.
@Skullkid Thats fair, there is a much more robust selection of lora's and options with 1.5, it's had a lot of time that xl hasnt.
@glitter_fart I think the main issue with SDXL is hardware requirements, then we should see some improvements soon, as people discover new optimization options. SDXL turbo for example is one thing that is really impressive, generating 1024x images in 6 steps. Also options to train models online like here in CivitAI with an accessible price may help to speed up SDXL fine tuning.
@Skullkid yeah most deff, the extra ram/time required for it are something people who are already comfortable with 1.5 don't want to bother with, and have to agree that with turbo and some time things will improve. already using a "turbo" merge model that handles 1024x natively at 2 steps with dull but good results and more steps for details/colors/higher cfg. I expect some turbo models will be pushing 1536 or even 2048 soon but some things will need to be adjusted for that to be feasible/timely