Using AI to manipulate Sauerbraten screenshots

[Salatiel]

In 2021 OpenAI published DALL-E, a neural network capable of manipulating and generating images from text prompts (https://openai.com/dall-e-2/), it is still only available through a waitlist, but since then our dear open source people have come up with alternatives that we can use for free, and what do you do when you get your hands on one of these tools? you find a way to put Sauer in it of course :P
below I'll leave some experiments I did modifying Sauer screenshots using only text:

An even more futuristic Xenon base


There are other variables you can apply in addition to the text prompt, so some results may look weirder.
Here is a landed spaceship:


Venezified venice

if the canals became paved roads:



Turbined turbine




Snow and rain in urban_c




Modern architecture in urban_c




Rainy alley in kopenhagen


Police car in ghetto


You can also use the result as input multiple times and get something that is more distinct from the original image, here's a screenshot of a race map made by TristamK shaped like a subway with people walking in it:


Some more random tests:
Bedroom in CpHills (Cooper)


Cooper's motorcycle


Synderf's tank



Galaxy's restaurant kitchen



Razgriz's triforts




If you want to try it on your own without a super computer, there are some "Image Generation Models" available on Google Colab and Kaggle, some require you to have a pro account on these platforms, others like the one I'm using (Lite's Latent Diffusion v9) just need a verified phone number.

Bonus mm-auggiecat.jpg