DLSS 5 preview

[CAT-THE-FIFTH]

The top comment is very interesting. If Nvidia gives devs ultimate control on the strength of this tech, i suspect that for AAA games (that aren't sponsored by Nvidia) this will be the ultimate fate of DLSS5. A post processing affect that you need to pixel peep to spot the difference. Simply just a different look to the final image.

The ultimate goal might be for developers to integrate this directly into the game design process.

 

[mrk]

They won't want to do that as it will give a lower visual experience to console players which makes up for the biggest average player base.

 

[CAT-THE-FIFTH]

They won't want to do that as it will give a lower visual experience to console players which makes up for the biggest average player base.

Good point.

 

[mrk]

 

[eddyr]

That's just mental gymnastics being employed in order to handwave away the opinions of the developers and artists who've actually worked with it.

"Most People Bashing DLSS 5 Are on the Peak of Ignorance" - Veteran Game Artist Shows Exactly How Much of a Difference Lighting Can Make

A veteran artist behind Dishonored 2 and Hogwarts Legacy fires back at DLSS 5 critics, showing exactly how much lighting can change a face.

wccftech.com

"If DLSS 5 Was Shown as a Next-Gen Hardware Reveal and not AI, You Guys Would Be Going Nuts" - Game Dev Hits Back at Anti-AI Crowd

Veteran developer JP Kellams and analyst Ryan Shrout push back against DLSS 5's social media backlash for being a mere "AI filter".

wccftech.com

If you have some evidence to support the idea that the only developers and artists to have actually seen and used it in person, are tweeting about it positively because they're being paid to, then show us.

Otherwise, it just sounds a bit silly:

"Wahhh the artists and developers who've actually used something disagree with me about it, they must be shills...."

I don't care for the argument about payment personally. But have you read your first link.

That guy has grasped the wrong end of the stick completely, or is misrepresenting and recharacterizing the objection. Which is rather amusing.
He perhaps even shows he himself doesn't quite understand the fundamental tech at work. His instinctual baseline comprehension is to compare to specific generative AI end-products, and say "it's not a prompt based image generator", bizarrely focusing on the front-end user interface and text input of other related but different products. The expression "no _ sherlock" springs to mind.

He is focused on and likes the result. Great and understandable. He appreciates the work/effort it saves for the level of final output, bridging a gap which while shrinking still exists and that he and others struggle with. Great and understandable. Choosing to close himself off to the source of the poor reception.


Your second link too, wrong end of the stick.
"Its the delta", No, its the how and nature of the result.
Oh and the insight that AI is here to stay. Again "no _ sherlock" springs to mind.

 

[GordyR]

I don't care for the argument about payment personally. But have you read your first link.

That guy has grasped the wrong end of the stick completely, or is misrepresenting and recharacterizing the objection. Which is rather amusing.

While I appreciate that they may not be your objections, they absolutely are the objections that have been so widely spread and regurgitited across the internet, in a manner not at all unlike tabloid sloganeering.

That's what he's complaining about, and it's precisely the way that DLSS 5 has been characterised by such a large mob of ragebait YouTubers, and huge swathes of their audiences.

If they're not your objections personally, then he's not talking about you; but he's not mischaracterising the complaint, it's just not a response to yours.

He's taken umbrage to precisely the kind of response that I've taken umbrage with myself.

He perhaps even shows he himself doesn't quite understand the fundamental tech at work. His instinctual baseline comprehension is to compare to specific generative AI end-products, and say "it's not a prompt based image generator", bizarrely focusing on the front-end user interface and text input of other related but different products.

Again, that's truly how it's been framed by so many with their "AI slop, Instagram filter" wails.

It's a technical rebuttal to that, which I expect is out of sheer frustration at other (largely unqualified) people's mischaracterisation of the technology, currently flooding the internet with comments, based upon their playground level understandings and their experiences with AI chatbots.

There's plenty of good criticism out there, and discussion about aesthetics and what this means for the industry as a whole, that I fully expect the two devs in question would be entirely on board with, but again, that's not what they're responding to.

He is focused on and likes the result. Great and understandable. He appreciates the work/effort it saves for the level of final output, bridging a gap which while shrinking still exists and that he and others struggle with. Great and understandable. Choosing to close himself off to the source of the poor reception.

I appreciate all that you're saying, but we've got 10+ different types of people, with probably 10+ different types of criticism, some ranging from careful analysis and balanced consideration such us in the reddit thread that Tinek just shared, right up to the mindless "AI slop" screaming bandwagon jumpers who likely don't know their arse from their elbows when it comes to technology. Your argument or perspective very well might be a legitimate and carefully considered one, but that's not been the reaction of the wider community as a whole, which to me at least, seems to be what these devs are responding to.

 

[keyser van someone]

I've played around with image/vidoe gen and LoRA gen in Pinokio and ComfyUI. Good times were had by all.

Since playing around with that I get a feel for how it works and I think there is some DNA in that for DLSS 4.5 and 5.

It's better at the lighting part because lighting follows the same rules regardless of any external ai model training data.

The issue squarely lies in using non-game assets to produce a new face image and you end up with a mismatch of real women images fighting with the look of the game art style (not real looking, still CGI).

If you restrict the upscaling/lighting to only use game assets, similar to how a lora can change an ai image/video gen and stick to the game art style then that could work and shouldn't end up in uncanny valley. imo

It's moot until it's released, but it feels like we're arguing over DLSS v1 right now, yeah it sucks, but one day it might not.