Perpetual coffee request: Keep me motivated and caffeinated:
Description
FAQ
Comments (10)
of course, SDXL version more powerful
Hey! Nice to see you're back!
Will you be converting all your Loras to SDXL?
I sincerely hope you do! Since you had said you retired, I was trying to figure out how to convert some (or possibly all) of your Loras to SDXL myself. I must say it's much more difficult than I expected. My respect for your efforts and devotion increased by an order of magnitude!
Any chance of you going over what changes to your training method were necessary to get SDXL Loras?
yes and yes
@konyconi I'd kill for you to redo the bird and muppet lora asap tbh!
@konyconi thank you! Can I just bow down and worship you? (joking ofc)
Thing is that for now the LORA's IMO will be inferior quality to SD 1.5. But in some months this will change for sure. We need a custom super finetunned model to align models there. For now the only way to "convert" (which is impossible without THE SAME datasetS U^^) is to get good AI generative images of that model and train those after nice resolution to 1024x. And the training is inferior IMO but that's just me. Some are telling is better... I don't know. I trust my eyes.
@LDWorksDavid I actually like the results for XL more than those for 1.5
@konyconi @LDWorksDavid I'd say it depends on the subject, For some, like the coffee machine, Iliked SD1.5 better. For most though, I feel like SDXL gives a much nicer result.
I need to figure out why my A1111 stopped working with SDXL before I can explore this correctly. While I like ComfyUI for some tasks, I much prefer A1111 for exploration of models and multiple tests. The diverse extensions are just so useful.
@konyconi I think gives more detail but if you look something very specific goes a bit far away. Matter of tastes. More detailed for sure, IDK if better at least in my case. BTW, wecome again!
Tuyệt vời!