Time for a quick update.
The past few days I've been experimenting with StableDiffusion tools as a way to generate some artwork. This is not my first foray into this new thing as I've messed around with it previously when experimenting for some half-started project ideas in the past. The difference is that this time I've switched to SDXL after being inspired by some of the excellent fan art appearing in
I can say that I've been happily surprised with the improved output of the new Models, while being simultaneously frustrated at the lack of effective ControlNet tooling as was available for SD1.5. (Yes, I have got the SDXL versions of some of the controlnet stuff but it simply does not work as well as it did in the previous generation)
Anyway, I've found the following seems to result in fairly good reproduce-able, controllable outputs:
- Using a strong LORA,
- prompting that works with the source embeddings, not against it,
- lots of repeated generation and careful selection of images to go to the next stage,
- cleaning up problems (*fingers of course) with manual edits/combining the best parts of multiple images, and then doing another img2img pass to heal stitching seams
There's one question though... the strong LORA I spoke of may in fact be cursed....
Can you all handle Dark Cookie's style in another game?
Anyway, here's some Concept/Placeholder/Early-sample art for the heroine:
View attachment 3604950
View attachment 3604967
And then one after she's getting a little strung out... and endured a little "breast expansion"...
View attachment 3604975
What do you think? Would this style/type of artwork fly?