Intriguing and nice writeup. This however tries to encode all the hidden painter's knowledge. Would it be easier to generate realistic images from paintings and use these pairs to finetune an image generation model?

From the article

> In the days of generative image models, it's still fun to see what can be achieved using traditional image transformation techniques.

> But the really fun part of the project was in combining a number of techniques, such as Perlin noise, wavelets, and edge detection, that I played around with previously in isolation. All these different tricks came together for this project, making it possible to build something greater than the sum of its parts. I find these are the most rewarding types of experiments where you can build on things you've previously learned and combine them in novel ways to make something new and unexpected. I hope you enjoyed the journey as much as I did working on the project.

You can already use genAI to do similar things. The whole point, however, was to see what I could do using traditional techniques. The other benefit is that it's something you can run locally without needing a lot of resources, and have full control over the shape of the output. You can also combine this with generative models, and have a model generate a picture, then have this repaint it. Ultimately the point was to have some fun and play around with combining these techniques.