How is your experience with re-seeding images to the system going from small resolution to a larger one? When I tried it, results were not as good as starting a high res image from scratch.
Thanks for pointing this out. Actually NeuralBlender.com was not to be released to the public yet, but the topic in this thread kind of invited it. I will clarify that and improve the UX.
It is based on VQGAN (trained on imagenet) and Open AI's CLIP.
To make it more similar to art you would need to do some "prompt engineering. Try: "painting of ..." or "... painted by Van Gogh"
Part of the #NeurIPS2018 Creative Art gallery currently in Montreal.
This machine artist creates a new piece of art every few seconds - indefinitely.
Inside, the particular neural network developed an affinity for depicting portraits of people with one blind eye. Once in a while this is compensated with a face having a third eye.