2023 was the year of AI and people have been gushing over Dall-E, MidJourney and ChatGPT, but there are other useful applications of generative "AI".
I've had some spare time and were restoring some old photos for my parents. I've used tools that leverage Enhanced Super-Resolution Generative Adversarial Networks (GANs) in the past to enhance or super-resolution images. This has been useful to create memes of friends and co-workers with images that were blurry or pixelated.
One of the first Google Colab notebooks branched (had to fix some issues) were able to recover/enhance some of the images I needed.
Here are some sample comparison with the locally run networks via a GeForce RTX 4080:
The scaling can be adjusted for the super-resolution feature. Some models are better suited for text, animation, or photographs.
Original image of my sister:
Enhanced (with derpiness factor at 2.8):
You can run Xintao Wang's work through Hugging Face as well. For sensitive images, I would run the project locally and if you're super paranoids, monitor the network or disable internet.