Uploaded by request.
Pony:
Best at 0.75 using T-ponynai3 v6.5 & v5.1 (Weight Optimized)
Trained with original line art from Shon Howell at 1024x1024
*Adetailer recommended for distant faces*
--
SD15:
Best at .75 - .80 weights.
The dataset was of rough sketches but I managed to clean them up enough to give pleasing results.
Trained with a variety of buckets surrounding 768x768.
512x768 works well in both portrait and landscape
Description
Much higher resolution and more artifact removal.
FAQ
Comments (4)
For XL models it would of been better to train them at 1024 resolutions otherwise you get odd generations. In my tests with this model and different checkpoints it ended up looking distorted when rendering in sizes larger than 768 resolutions.
SD15? Yeah, that was a while ago. I could hardly train a 512x512 back then. The Pony is trained at 1024x1024 and I don't see these issues.
@UncleJert ahh I see I misread the description, though I am not sure why it is being distorted for pony, weird..
@LindezaBlue 🤷♀️