@dkpc69  if you aren't doing a more than about 5000 steps its pretty quick to train on ai toolkit using runpod they have a ready to launch option. I did one lora with 35,000 steps and that was brutal even with a rtx 6000 rental cost bout 16 or 17 bucks of rental time. took like 8hrs but if you set it not to generate samples except for every 1000 steps its pretty quick. could knock out 5000 steps in perhaps a few hrs.

@dkpc69 hmm that's interesting I just do my generations at 1200x1800 but I do use a workflow that runs a second pass on the image at a lower denois like only 0.2 -0.3 to pop details it really does a great job. Not super time consuming but I do use res_2s as the sampler because its just a bit better that Euler at details. I shared a few of the images I made using your other lora. About to share some more with this painting style lora. They turn out really nice and sharp! There does seem to be a problem today where the images aren't showing up in the lora galleries :(

@dkpc69  if you are interested in trying the workflow I shared it in an article , the credit for it go's to Aitrepreneur https://civitai.com/articles/23396/running-zimage-with-second-pass-on-initial-image-to-double-quality-and-refine-the-output

@dkpc69  just one tip if you do try that workflow, don't use the style lora as strongly in the second pass, or you can turn it off altogether. for the image you saw, I ran your lora at 0.71 and mine at 0.5 strength then on the second pass I ran them both at only 0.15 because it allows that base model a chance to refine more.

@dkpc69 would you mind training one for either base illustrious or noobai? looks awesome and kinda wanna see the effect on an illustrious-type model

@dkpc69 Legend <3333