- Jan 19, 2022
- 327
- 656
Ya … I figured it was a first pass down and dirty proof of feasibility exercise.You have to consider that I took the images from the game files. These were 400x400 low res icon images.
I upscaled these and then ran some standart stable diffusion converter over them.
I also converted these as a batch. So no afterward editing.
Just used "smile" and "looking at viewer" promts.
It all takes just a couple of minutes.
For that little effort, the results are great, imho.
Don't get me wrong, I don't want to show of or something. I just find it fascinating how good it works and how fast this can be done, with this little effort.
I said 60%-80% were improved.
Which is a huge amount.
Sorry if you took the positive feedback as a neg.
I figured that with my upvotes and the improved comment I could share where I saw that the AI “improvement” … wasn’t.
Specifically if Ash looses her sometimes crazy eyes. (Which was DigDug69’s comment)
Also the AI seem to add bunny teeth when it sees bunny ears.
And OFC the hands but whatever.
Consistency could also be an issue … don’t want to have 12 “different” versions of a character.
It raises lots of great possibilities… and it will only get better with time.
I tend to try and identify the broken parts to correct or to maintain consistency on a project.
Last edited: