The Doomguy. Stoic, silent, flickering eyes at the bottom of the screen. He's not really the talkative type, is he? Who would've guessed his first words weren't "death" or "murder" or "double-barrelled", but a Russian-language explanation of complex neural network modelling techniques? This week, developer Denis Malimonov unleashed his own sort of hell - replacing his regular human face with the Doom Slayer's low-fidelity visage with machine learning.
Without further ado - I'm sorry for inflicting this on your stainless screens.
Thanks, I hate it. Oh, it's very impressive technology, but I retch when it becomes more detailed, twisting Doomguy into a working dad. One who looks real impressed with the deal he just landed on his home insurance. The way the pixellated original moves and twists in three dimensions is more horrifying than any of hell's denizens.
The model Malimonov worked from is given a brief run-down over on it's Google Collab page. That system can take a square video and an image file, merging the former's motion onto the latter through AI-powered shenanigans. Malimonov then adds interpolation frames to smooth out the video, and upscales the result.
Granted, it's the sort of AI-driven tech that immediately raises alarm bells. Yes, it's funny to make fictional characters say cuss words. It's not quite so laughable when malicious actors start faking up videos of political actors, celebrities or otherwise to spread misinformation.
The smooth, fiscally-conscious Doomguy seen above isn't a result of this process, mind - Malimonov has simply sourced this uncomfortable image from Redditor Zergling103, which was itself was created from multiple attempts to upscale the original Doom portrait with machine learning.
With Doom Eternal arriving this coming Friday, let's hope some skilled modder gets this man's face behind the Praetor mask ASAP.