- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
I’m completely speechless. This looks so terrible I thought it was a joke, but apparently Nvidia released these demos to impress people. DLSS 5 runs the entire game through an AI filter, making every character look like it’s running through an ultra realistic beauty filter.
The photo above is used as the promo image for the official blog post by the way. It completely ignores artistic intent and makes Grace’s face look “sexier” because apparently that’s what realism looks like now.
I wouldn’t be so baffled if this was some experimental setting they were testing, but they’re advertising this as the next gen DLSS. As in, this is their image of what the future of gaming should be. A massive F U to every artist in the industry. Well done, Nvidia.
Lighting in video games encompasses a lot of things these days. I’m not a graphic designer but, to my untrained eye, it looks like this tech is pushing hard on things like subsurface scattering to make materials appear more realistic. It doesn’t appear to be changing the actual polygons. This picture further down seems like a more realistic example of what it would end up looking like when the technology is complete. I think they picked that particular first image as an example BECAUSE it’s such a stark change.
Ok but the included picture added eyeliner and a red lip stain.
It Hollywood glossified it.
Zoom in on the before; the textures are fairly low res so it’s hard to see at a distance but she clearly has eyeliner along her top eyelid. I can’t really tell if the intent was for the dark undereyes to be eyeshadow or bags from being exhausted. But you can tell the AI is pulling from source coloration, not just adding things willy nilly like a beauty filter. The lip coloration, idk, that could be an artifact of the early technology or a hallucination. But to me it appears that she had some sort of lipstick on because it doesnt extend all the way to the corners of her mouth. I don’t think contouring is the right word but maybe someone here who wears makeup could add some detail?
edit actually looking even more closely, i can’t tell if that’s eyeliner or just her eyelashes being dense. Either way you can see what the model is trying to replicate. I don’t think it looks good but it isnt doing a “beauty filter”. If the model is adding eyeliner, it appears to be confused based on the thick black line in the source image. Again, these are really low textures compared to what it’s trying to output so hopefully this improves over time like every other DLSS tech has.
It’s not contouring, it’s adding muscle lines, that I interpret as pure scaling up. But yeah as someone who wears makeup, I stand by my position especially on the lip color. It just reminds me far too much of the incel redesign of Aloy from Horizon Zero Dawn. It also plumps the lips and gives them more variance from center to edge.
And I want to be clear here, I don’t think they did this on purpose. I think that the training data is more likely to use more conventionally attractive women’s faces including ones with makeup and so it accidentally does some light beautification while in there.
Yeah, I dont know what the word is. My point is mostly that you can clearly see what it’s trying to recreate from a low-quality source. It seems to be keeping a very close adherance to the original looks? If you lay the images on top of eachother you can see the same creases around her eyes, the same nose shape, the same jawline, the same skin-colored makeup at the corner of her mouth to make her lips appear more defined. I think you’re spot on about the “training on conventionally attractive models” which is the part that I’m hoping improves with time. I don’t think applying this to existing works is appropriate at all but apparently it will use nvidia streamline so the mods will come whether we want them to or not sadly.
I am so not sure about your points, if they are all the same, and AI just tries to add more details, then should AI disregard the source material and make a more plain interpretation to appease certain viewpoints?