VaM 1.x DLSS 5 in VAM2 + DLSS 5 Experiments in VAM1

[Curves3E]

I think you should ban the use of computers also..they are the devil's work. Smoke signals are the way to go.

On a serious note, banning a dlss technology that's especially beneficial to your product is equal to shooting your feet. I am guessing that your thought process is that dlss=AI and AI=slop. AI is slop but dlss does not equal generative AI.

Your users are really exciting about the new technology, searching the forum for the capabilities and implementation of dlss and really soon they are going to find out about the "ban". I really don't think its going to go the way your team is thinking..Just my 2 cents

I think they are smart enough to understand that the technology doesn't add polygons to the models; it only takes what it sees and reshades the 3D model based on cues it finds in the scene. This changes nothing in terms of liability; otherwise, developers wouldn't have access to tools like this at all.

I think VaM 2 may actually have DLSS 5 one way or the other. Forum moderation, as always, will get harder to enforce because how can you tell which screenshots are genuine in-game screenshots and which are forgeries generated entirely by AI?

 

[EvilBox]

I did a 3 female person atom scene with DLSS 5 on in VR, works, averages around 18FPS (jumping between 16-22.)

That was fast! Great job!

I think you should ban the use of computers also..they are the devil's work. Smoke signals are the way to go.
On a serious note, banning a dlss technology that's especially beneficial to your product is equal to shooting your feet. I am guessing that your thought process is that dlss=AI and AI=slop. AI is slop but dlss does not equal generative AI.
Your users are really exciting about the new technology, searching the forum for the capabilities and implementation of dlss and really soon they are going to find out about the "ban". I really don't think its going to go the way your team is thinking..Just my 2 cents

I’m excited about it too, but I can understand why screenshots like that aren’t allowed in the content previews. Just think about people using AMD cards. A new user signs up, sees a beautiful screenshot follows the link and pays without really reading the description then launches it on their own machine and gets something completely different from what they saw in the preview. Ta-da! That’s not supported by default - go buy another GPU xD

 

[julianfx]

33fps single person atom, so yeah, pretty big FPS hit. 75fps with it turned off. To be clear, my VR is set to limit it that way. I am on a 5090 with a 9800x3D. Will say that when in desktop mode it's around 60-65fps. So, it's literally half, which makes sense because it has to re-render for each eye. I'm trying to optimize and find some FPS gains, but I'm not holding my breath, we'll probably need Nvidia to improve the actual neural engine to improve FPS, which I think they will.

Thats great news. If you find the time please make a guide.

 

[BillTheButcher]

That was fast! Great job!


I’m excited about it too, but I can understand why screenshots like that aren’t allowed in the content previews. Just think about people using AMD cards. A new user signs up, sees a beautiful screenshot follows the link and pays without really reading the description then launches it on their own machine and gets something completely different from what they saw in the preview. Ta-da! That’s not supported by default - go buy another GPU xD

To be fair that's the new users fault. You don't buy a car without looking under the hood do you?

 

[hazmhox]

I think you should ban the use of computers also..they are the devil's work. Smoke signals are the way to go.

On a serious note, banning a dlss technology that's especially beneficial to your product is equal to shooting your feet. I am guessing that your thought process is that dlss=AI and AI=slop. AI is slop but dlss does not equal generative AI.

Your users are really exciting about the new technology, searching the forum for the capabilities and implementation of dlss and really soon they are going to find out about the "ban". I really don't think its going to go the way your team is thinking..Just my 2 cents

Vocal minority / Actual real market added value... funny how you interpret the reality when again... 6% of the market (at the moment) will be able to use the final api.
That's not because a couple dozen dude posted screenshots over discord that it means everyone wants that... nor CAN use it.

Neural Rendering is litterally generative AI. Don't do revisionism because it suits your argument. It contains the keyword "neural" in the frikkin' title. That's litterally the base of generative AI. And you just need a couple of demos to litterally see characters are getting slopified with it... saying otherwise is just intellectual dishonesty. You like that stuff? Great, good for you. But at least be honest with yourself.

Our thought process is not about preventing AI slop, it's about protecting our community. Most of the decision process behind all the spectrum of post processing was because everytime a mod/system appeared to massively post process the output image of VAM... Every. Time... some creators immediately jumped on the occasion and abused it to present a content that is widly different from what you would get without these tools.

You're making this look like it's a new thing... it's litterally in the guidelines since a long time. When the very advanced Stable Diffusion tools arrived, the moderation immediately took a stance on it. When some Reshade or massive photoshop modifications were used, the hub took a stance on it... this is not different.

And finally, you probably won't see this because you apparently did not read the thread and subsequent answer... we don't ban the technology. We are refusing at the moment, full DLSS5 screenshots and previews in the release documentation of a resource. But I guess nowadays on the internet it's easier to polarize a single answer taken out of context from a massive thread on a complex discussion and moderation process.

 

[hazmhox]

To be fair that's the new users fault. You don't buy a car without looking under the hood do you?

Hahaha

You really think anyone (like the average random joe) look under the hood when buying a car? I don't think so.

EvilBox is right. We're talking about the internet. People don't read, people don't look at instructions, people get baited. Even tho I somewhat agree with you that users got overall lazy and that sort of thing might fall under their responsibilities... these sort of practices (showing shots that do not reflect the vast majority of hardware and needs a very specific tech and api) fall somewhat under the "dark patterns" category... or said differently: click bait.

One thing that I will say tho: if as an artist on the hub... you can't do very nice shots of your content and being happy of them without NR, you probably need to work more and improve your art. VAM could and can be beautiful without that thing that came out 3 days ago

 

[rabbit1qaz]

Hahaha

You really think anyone (like the average random joe) look under the hood when buying a car? I don't think so.

EvilBox is right. We're talking about the internet. People don't read, people don't look at instructions, people get baited. Even tho I somewhat agree with you that users got overall lazy and that sort of thing might fall under their responsibilities... these sort of practices (showing shots that do not reflect the vast majority of hardware and needs a very specific tech and api) fall somewhat under the "dark patterns" category... or said differently: click bait.

One thing that I will say tho: if as an artist on the hub... you can't do very nice shots of your content and being happy of them without NR, you probably need to work more and improve your art. VAM could and can be beautiful without that thing that came out 3 days ago

in the case of items preview using NR should be forbiden, overview should show item as it is withoug any AI enhancment, hazmox has right here i hate when eg when buying some 3d models "creators" put AI created picutres and its nowhere compared to real product.

 

[EvilBox]

Think about this for a moment. We’re no longer talking about just another technology we’re talking about a real divide, not only within our community, but among consumers in general.
Ideally, regulators in the US should force Jensen and his company to share this technology with competitors, especially AMD. This isn’t just another gimmick that nobody really needs, like HairWorks back in the day. This is a genuinely powerful tool, and it creates a very real divide between users. There can’t really be healthy competition in a situation like this, and in the end it’s the consumer who suffers from the monopoly. That’s the first issue. I think pretty much every gaming community is going to get shaken up by this, and the arguments around it will only get louder.

The second issue is development itself.
I’ve worked in development teams before, and I know that some decisions are genuinely difficult to make. Right now, the VaM developers are in a very uncertain position. Nobody really knows where this technology is going, whether it’s worth investing development time into it, how long integration would take, what technical problems might appear, or how many new risks and maintenance headaches it could introduce. Those are already difficult organizational questions, and then you multiply them by the technical ones aspects, as well as the ethical side of it (how another artist might view the issue). But there’s also another risk: if you ignore this moment and put everything into traditional processing and rasterization, you might end up falling behind later if neural rendering becomes widespread. If it becomes widespread of course. So there’s another distinction to make here: there’s the game the developers are working on, and then there’s a different game — the one the users actually experience

Personally, I don’t think we should be overly hostile toward AI. It’s already an incredibly useful tool. Remember how people used to make fun of AI images because they couldn’t draw fingers properly? Or how everyone constantly complained about hallucinations? Those problems haven’t disappeared completely, but they’ve already been reduced enormously in just a year. Give it a few more years and these tools are going to become extremely powerful and naturally, probably quite expensive too.

And now for my confession: here are some things I used AI for.

Spoiler: 1

https://i.ibb.co/FLf59Lx2/1.png

1 cake was generated with AI, although the result wasn’t good enough, so I reworked it manually.
2 decorative pattern was also AI.
3 confetti AI too.

Do I feel like I’ve lost any creative freedom because of that? Not at all. It simply makes me faster. And I’m not ashamed of using it in the slightest, because I’m getting a good result. Ignoring tools that can already save you time and help you create things today even if you have to pay for them just seems silly to me.
One more thing. I didn’t ask the author of this screenshot for permission to use it, so I hope he’ll forgive me for that!

Spoiler: 2

At the last moment I decided not to include the image I respect him as a creator so I just feel like it would be wrong to use his image without asking.

But what do you think he told me after trying this? “DUDE I’m never launching the game without this thing again” qnd the point isn’t that his character in the screenshot was somehow generatively improved. The point is the overall image quality. It no longer looks like the same plastic mannequin we were used to seeing before. It starts to look much closer to modern graphics. I think that’s my last two cents on this topic. I don’t really have anything else to add

 

[XRWizard]

I have a 4090.. totally going to try and push NR even if it's at a lower rez...

I think I've taken textured polygons as far as I can... Sure vam 2 hdrp will look better but it's still going sit in the uncanny valley for me, hoping NR will take us past it.. it's annoying NVIDIA have stitched it up (same deal with cuda) but maybe in time there will be an open source implimentation.

Some official guide or implimentation would be awesome.

 

[hazmhox]

That’s the first issue. I think pretty much every gaming community is going to get shaken up by this, and the arguments around it will only get louder.

True! One thing too: let's assume it's not only limited to 50XX (which it will be)... the fraction of people being able to buy a GPU running that somewhat properly ( at least 60fps, 120fps being the ideal situation ) is ridiculous. GPUs being able to output proper framerates are top of the line... at least in the 70Ti tier. These GPU cost from 1K to 2K+.

The question is not "who wants it?" but more "who the frak can afford it?"

if you ignore this moment and put everything into traditional processing and rasterization, you might end up falling behind later if neural rendering becomes widespread

Modders gotta mod tho. 95% of VAM is community content and mods. VAM 1 does not have DLSS or NR, yet, it's there. I think people are just whining because it's good to have a polarized opinion and a strong view on any subject... when in fact "man, just mod your vam and get on with your life".

It changes nothing that the hub would not allow the NR previews or the game not having the tech integrated... when you can enable it on your end by other means.

Personally, I don’t think we should be overly hostile toward AI.

Not everyone is fully hostile towards AI. I'm gonna contextualize this: since I know you, you seem like a genuilely passionate creator who works so hard on his craft that he releases probably some of the best enviros on the hub. Your artistic vision and your approach of the AI toolset is... well the "toolset" approach.

What people are hostile toward is the full on generative approach and the IP/license theft that these techs induce.

And I'm 100% both of these profiles. I find neural nets technologies extremely powerful for toolsets and I frikkin loathe generative AI as a way to produce massive amount of low quality content in a very short period of time. (and I'm specifically not diverging on the ecological approach of that debate).

The thing is, people advocate or defend AI because they like shiny stuffs, easy content, and in the case of the NR for VAM... horny gamers who wants "dem photorealistic boobz". It's not a question of is it good or is it bad... it's a question of "it somewhat looks like dem cool generative babes". It's not hard, if you look at the DLSS5 channel on discord (without pointing people out), you have very very bad shots slightly lifted by the NR and people react on this like it's good... it's not.

So yes you're right, I think we should understand what is tool and be supportive of anything that can help artists with neural nets technologies. But we should be way more hostile to anything that is lazy generative bullshit driving the overall artistry down the drain.

It no longer looks like the same plastic mannequin we were used to seeing before. It starts to look much closer to modern graphics.

If you say that in the "VAM 1 visuals" context, true.

But, we still have a problem (VAM 1 or other games). NR has a massive bias which comes from generative models. You see all the tropes and bad things emerging from face generation being trained on mostly... porn: heavy transformation that tends to create more caucasian looks, for any semi-dark skins a tendency to white-wash them, intense conversion of "slight skin defects" to freckles for no reason, highlights and tones lifted to look like instagram and tiktok beauty trends... and that's just the generative part.

If you account for the exaggerated tone mapping (even if you reduce the local tone to 0) that feels like ACE tone mapping multiplied by 3 (and I'm not a big fan of ACE in the first place). The loss of depth and contrast half the time... we're gonna get games that looks all the same and will feel like a frikkin' instagram feed.

Give me a NR that does local AO, some slight tone lifting to improve the "lighting feel"... just don't frikkin modify the image so much that everything ends up looking the same.

It's a good thing that this is an API and cuts the framerate in half... 'coz that means it will be a non-mandatory pass in games. If they are pushing this without the ability to opt-out. That's a no.

 

[Ladiesman217]

That was fast! Great job!


I’m excited about it too, but I can understand why screenshots like that aren’t allowed in the content previews. Just think about people using AMD cards. A new user signs up, sees a beautiful screenshot follows the link and pays without really reading the description then launches it on their own machine and gets something completely different from what they saw in the preview. Ta-da! That’s not supported by default - go buy another GPU xD

DLSS 5 can be on or off and have different settings as well. If you allowed people to post screenshots of DLSS 5 based images it makes it impossible to compare things fairly. Also does anyone even know if DLSSR 5 can generate 2 imags properly 1 for each eye? So that one is for the left the other for the right and the end result is consistent and works as VR is supposed to. Because I wasn't expecting that to ever necessarily be possible with AI let alone already without Nvidia seemingly even promoting that as something it can do. Which if it can't of course DLSS5 should be off because that makes it about as much use as chocolate teapot as far as VAM is concerned. I'm hearing people say it works for me but I didn't spot anyone saying it works for them and specifically claiming it works IN VR which is the entire point of VAM.

 

[Ladiesman217]

Man, you’re expecting something pretty quickly that is actually quite difficult to do properly. DLSS 5 is a lot more complicated than it looks. Right now we don’t have proper engine-level integration. For full control you need the SDK and an actual connection to the scene itself. Semantic masking can automatically separate different types of objects, while engine-level masking lets developers or artists explicitly tell the renderer where and how neural rendering should be applied. VR without that foundation is basically impossible to do properly. I wouldn’t expect a well-working VR implementation anytime soon.

I can expand a little on what I meant because I’ve looked into the technology more closely now.
For example those weird faces people show in anime-style games don’t mean the technology is somehow “destroying creativity.” Without proper integration, the model is simply running without the artistic controls it’s supposed to have basically following the default behavior it was trained and configured for. And after looking through the SDK this is really the important part: the artist decides what can be changed, where it can be changed and how strongly. There are controls for structure andtone, semantic masking, engine-level masking and local intensity masks/textures. So DLSS 5 doesn’t take creative freedom away. With proper integration, the final control still stays with the creator.

For example, if I look at a glass cup in my scene and think it looks bad, I can tell the neural renderer that this particular object should be heavily reworked into more natural-looking glass, while leaving the other objects untouched or only changing them very slightly. And in VaM right now we basically don’t have that foundation yet. Getting neural rendering to run is one thing but properly controlling it at the scene and individual object level is a completely different problem

Is it actually working for anyone in VR? Because if not I've thought about this alot and I think it's possible VR becomes a sort of last stronghold for 3d artists until AI can just literally do it all in 3D because DLSS 5 just might never work in VR. I don't know if DLSS5 or any other similar tech can ever be applied to VR because it's rendering 2 cameras in 1 3d scene. When you apply DLSS5 it's making changes to the final image which are not true 3D and i don't know of any way you could tell the AI to work in a way that changed everything to work exactly like 2 3d cameras in the same single scene for example if the pore details or hair strand positions and shadows didnt match it doesn't work. If it works it works and would work emmidiatly but if it doesn't it doesn't and I can't even conceive of how you could fix that so IDK if it is something you fix or if it doesn't work emmidiately it just can't work in VR. But even if it doesnt work in this build of DLSS 5 because 1 or both eyes are just a black screen it may still be it just works but also be because they havent switched it on. But id it works but the images dont match I'd give it les than 50% chance of ever working in VR. In the sense that if it ever did work by the time it was working you'd already have photoreal visuals in VR and wouldn't need it.

 

[EvilBox]

True! One thing too: let's assume it's not only limited to 50XX (which it will be)... the fraction of people being able to buy a GPU running that somewhat properly ( at least 60fps, 120fps being the ideal situation ) is ridiculous. GPUs being able to output proper framerates are top of the line... at least in the 70Ti tier. These GPU cost from 1K to 2K+.

Now I think I understand your point of view 100%. I think we’re just looking at NR from slightly different angles. I tend to look at the whole frame and composition first, and think about how this technology could help me with materials, AO, reflections, lighting, and so on — I’ve mentioned that more than once already. I don’t really see NR as a “make it pretty” button at all. NR can definitely go too far, and that’s exactly why I’m so interested in proper controls and masking through the SDK. Honestly I’m already prepared to start painting body masks by hand
From what I understand your main priority is a little more focused on preserving the character itself. I can only speculate here, but there may be some kind of model data inside the DLLs that could eventually be modified to affect things like face generation and partially restrict what gets changed. Or you could hard-code a mask that only allows changes to certain areas hands, fingers, toes, the tips of the ears, maybe a little bit of the nose, etc. That feels more like a half-measure, and I can already imagine someone from the community removing that restriction with a mod almost immediately

The Instagram argument is probably the only part I don’t completely agree with. Let’s be honest: people were turning photos into idealized versions of themselves long before generative AI existed. Photoshop, retouching, beauty filters, and entire “make me look better” services. AI mostly accelerated and simplified a trend that already existed rather than creating it from scratch. People have always chased some idea of perfection, so I wouldn’t be surprised if, eventually, the whole site ends up flooded with this kind of imagery and video.
Pandora’s box is open now. As a moderator, you’re basically doomed to spend your time looking at and moderating generative boobs
My condolences

 

[hazmhox]

The Instagram argument is probably the only part I don’t completely agree with. Let’s be honest: people were turning photos into idealized versions of themselves long before generative AI existed. Photoshop, retouching, beauty filters, and entire “make me look better” services. AI mostly accelerated and simplified a trend that already existed rather than creating it from scratch. People have always chased some idea of perfection, so I wouldn’t be surprised if, eventually, the whole site ends up flooded with this kind of imagery and video.

You litterally don't disagree with me in that statement. You're just saying that in the past it was done differently but I don't see where you disagreeing on that state of facts today

Pandora’s box is open now. As a moderator, you’re basically doomed to spend your time looking at and moderating generative boobs
My condolences

Haha. Thanks

 

[EvilBox]

You litterally don't disagree with me in that statement. You're just saying that in the past it was done differently but I don't see where you disagreeing on that state of facts today
Haha. Thanks

Oh God, why are you making me write a wall of text?
There’s basically nothing for me to argue with here. I agree with a lot of what you wrote. As for copyright violations… hm. I know firsthand that companies have been training their internal models on copyrighted material for quite a while books written by very well-known programmers in their respective fields. Authors’ rights are violated everywhere, unfortunately. That’s simply how this industry works. Something similar was already happening about a year ago at a large company in my city. I know about it through friends who work there. Back then nobody was really talking about it were mostly just whispering.
You may have even noticed some strange subscribers appearing on patreon. I suspect that some of them are bots/agents automatically buying up content and other materials from creators’ pages (well at least I do get some strange transactions sometimes almost immediately after I publish a post).
And when it comes to the huge amount of AI-generated garbage, I completely agree with you there as well. I see it constantly on marketplaces for 3D models, textures and materials. Sometimes just looking at those assets makes me feel sick. So I understand what you mean by the Instagram style and the overall point you’re making. I’ve played around quite a lot with different models in comfyui myself, so I know how quickly that typical ai girl img starts to stand out