Howdy everyone, this is it, one of the projects i really wanted to work on, a custom 1.5 lora using ALOT of images (some from very high frame counted) gifs. This lora is made to make the yoshi's you make in 1.5 models look MUCH more akin to the ones of a Komdog/Komponi
suggest you put komdog and yoshi as the activators.
Description
Comments (5)
I tested with Kotosmix and SweetMint. Pics are.....'okay' in quality, but it might be over trained (the style is a bit bad) and only back poses.
mmm...i see i see. i guess it makes sense. sorry i did a shit job. im gonna try and remake it with diffrent settings, i hope it helps. Thanks for the review, im very new at this lol.
But I love where this is going, keep up the good work!
A tip is to get some models here in Civitai where you can download training images (some creators put it below the lora, like a separate download. So you can study how to create a good dataset. You can even generate your own yoshi LORA (say, 1000 pics), then select the best ones, in different background and poses, and then create a dataset based on your own creations, plus the best ones you used in your current LORA, to create a NEW LORA (Say, Yoshi version 2). I create my datasets using GIMP and Notepad. Yes, simple as that. You have to put good description on each image. If yoshi is facing back, you have to put 'rear view'. If yoshi is closing his eyes, 'eyes closed', if testicles are not shown, dont put it. As we are in AI world, the 'brains' learn from that, that's why we create newer versions until its good. Oh, and please try to use default checkpoints (i.e. if SD1.5, use default 1.5. If Ponyxl, Use default Pony, not AutismPony and other derived checkpoints), so your model can be used for other brains without breaking.
needs more cock