Killso

@opinali @K0kush0 @costantinismo btw even with motion vectors, it fails to keep consistency, note the floor near the couch, the shadowing completely changes (utter shock, the AI doesn't interpret the scene in the same way when things move even a bit). Or that it hallucinates details in their own promo shots

@opinali @K0kush0 @costantinismo to then run a super heavy ai model at the same time that 'll smear over 2sec cause it takes 2sec to fully regenerate a full quality image. I see very well what is the intent of nvidia and where this is heading, don't worry.

Daniel got important clarifications from NVIDIA. TLDR: the "DLSS5 skeptical" were right about *everything*. 1⃣ It's a 2D AI Filter. Input is only color buffer & motion vectors. The model doesn't see geometry, lights, PBR properties, normals, anything🧵 youtu.be/D0EM1vKt36s

@opinali @K0kush0 @costantinismo And secondly, I don't see how nvidia choosing the "easiest" way to integrate an argument, is this supposed to be a tool for rtx remix? No, this is supposed to be a tool to integrated in modern games as far as I know, this is not an argument to make the tech dumb.

@opinali @K0kush0 @costantinismo Obviously I also think this should be integrated more tightly at an engine level instead of a kind of post process, but this is not even what I was thinking of when saying to use more passes, if it already uses vector passes, it can use more, it's dumb not to.

@K0kush0 @costantinismo @opinali Saying no to using more passes is just dumb, yes, you can do it, we've seen it, but it make no sense on a technical level, why would you say no, to knowing what materials are and how they're setup? The engine CAN output this already, it's nothing new

@K0kush0 @costantinismo @opinali disagreed, not using more passes have profound issue, for one, it heavily relies on the original final image quality to interpret correctly, and for two, even if it somewhat works, it's much more prone to error, and it's one of the root issue that makes the final image deviates

@costantinismo @opinali Same question, I don't understand, and even if for some reasons you thought that was a good idea, the BARE minimum is it should require specific buffers that details the material params, that's BARE MINIMUM and they didn't do it, smh

@opinali how can an actual engineer with years of experience come to the conclusion that the best way to implement AI in rendering is putting it as a filter after everything is done. mind boggling stuff happening at nvidia

@stats_feed The best one (by far) is Pussy, attached to "léchère" that sounds like licker (but wrongly, cause lécher = licking), alongside its torrent of pussy.

🇫🇷 French towns and villages that sound like something funny: • Condom • Ars • Herpes • Lorgies • Trécon (‘très con’ - very stupid) • Bitche • Messein (‘mes seins’ - my breasts) • Pabu (‘pas bu’ - haven’t drunk) • Corps-Nuts • Folles (‘fou’ - crazy) • Chatte (female cat) • Le Saix • Duranus (hard anus) • Saint-Arnac (‘arnaque’ - scam) • Largeasse • Pinas • Villeperdue • Vinsobres • Couille (testicle) • La Conne (the bitch) • Monteton (homophone for “my nipple”) • Montcuq (homophone for “my ass”)

@iamcs2kitchen So this is where this idea comes from actually of the sens changing with stretching, it's technically false but consequently true if we look at the ratio distance on screen vs mouse movement.

@iamcs2kitchen reach the same spot ofc, but visually it'll be more than 3cm apart now, whereas vertically, it'll still stay 3cm, that means, if you play without knowing it's stretched, it basically "feels" like horizontal sens is actually higher, even if it's not.

There is a wrong misconception that your sensitivity in 4:3 fullscreen windowed is different than native. This is not true. CS2 reads raw mouse deltas, not screen pixels. Your sensitivity is same.

@NOTimothyLottes the same scene in an offline renderer, which uh, isn't exactly done in 2 clicks, you can't just upscale and compare here...

@NOTimothyLottes About dlss5? I agree and worst is that I've seen at least 1 person working at fucking epic talk about dlss 5 producing "correct lighting", mind you, compared to something that was path traced. However in good faith, comparing here is harder cause you're better off rendering

Likewise notice that the key theme in DLSS has been a complete disregard for "ground truth". For example no tech press or marketing shows the game super-sampled and then down-sampled (A=ground truth) -vs- DLSS upsampled (B=slop) -- to the same resolution to compare A vs B.

@playko777 @duck585c 1. Cause you think it changed anything regarding anything that you list? Says who exactly, your source at valve? 2. Why not? Why changing maps that have been the same for 15 years too then 3. Literal point of the update

@duck585c 1. show me one person who asked for this, and not animgraph2, ac, cl_bob or 128 tick 2. why would you change something like this after 15 years of game existing? 3. try spamming smokes now with this amazing update

“Who asked for this” “bad update” when valve updates it’s bad and when they don’t it’s also bad, this isn’t a bad update people are just being retarded rn

@GamersNexus there is no meme in this one

Here's another from someone. You people are insane. Thanks for the entertainment!

on second thought, DLSS5 has a point