Kurko1234First image is after autocorrection, second image is before autocorrection.
Been training an AI model for automatic art autocorrection. The AI model does not involve diffusion but it's a simple convertor that works as a last-step automatic correction of common flaws in art, which non-invasively edits the images for improved visual quality.
I thought this looks nice, so I guess I may share it. Have been training the model for 24 hours and there is still a long way to go, and I tried many times with different models and sizes. Finally, this one starts looking promising.
Let me know which one you like the most (the one before or after)
Also please ignore the black pixel lines on the left. The AI image transformation code miscenters the image so its one pixel to the right. Gotta fix that later.
Also bear in mind this is an early stage project I am doing on my free time. Still much improvement and training to do. Hopefully, I will learn a lot from this project and gain experience for future work, aswell as having yet another nice tool for me to do art. GenAI is just too boring to use though, but having an autocorrection AI to help me out is quite nice, while not modifying the image too much or changing the sense and feel of the whole artwork.
The model however does love decimating details on other artwork, specially complex ones. Guess the network is yet not trained enough.
2024-11-10 00:34:26 +0000