The model I think is fairly self explanatory.*Special Note: Seems that Adetailer really hates to play around with this above .7 strength or so, but I found generally that it wasn't needed when using hi-res fix. This concept is an interesting one as this model I think was only 250 steps and has a pretty good capture of elements of the image, it's not a 1:1 of the style and there is a lot of room for improvement, however I wonder as to if that's due to hi-res fix smoothing out the styles more gritty nature. It's an exciting prospect if I can make this work fairly well as it takes about 6 minutes to bake a model in this method and so far seems to be 80% there. Of course this LoRA is in a good position to be used I think, and I'll try with a style that doesn't have that sort of gritty texture to confirm if it's adetailer and hi-res fix smoothing out the grit and if it's the fault of the bake or not. Onto the next experiment or so. I'll likely come back to this model and try 4 epochs instead of 5 and see if that fixes the strength issue with adetailer if I can confirm that overbaking is directly the cause. It likely is as that was a common issue when I used 20 epochs for models.*Edit after futher experimentation: Seems that I can't figure out the adetailer burn, adetailer doesn't seem mostly needed with this model so it's not the worst situation. However attempting various fixes such as using 80 repeats and 4 epochs various combinations with batch size 1 or 2 only seems to result in a much lower quality result as well as the burn still happening with adetailer. Will move on from this model and try to figure it out as I move along into the next experimental models. I'll be uploading around 15-20 models a day for a while so be sure to give a like and follow for more models uploads every day!

[mossacannibalis] Experimental 1 image 100 Repeat 5 Epochs Lora Style Experiment PonyXL

Specific image request, see preview images for assorted relevant tags or look 'inside' the lora using your stable diffusion UI for them, it's based off one image so there's not too many of them.

Files with SHA256: 5820cce93b2c6947f8b1943520cf98d395950a07075c3c70ef4aaa091ca06fe8