28 replies to this topic

[froop]

So Square Enix released a few screens and a lofi video of their new engine and the internet goes wild.

My problem with it is that I don't see anything exciting and especially nothing new. We had graphics like this years ago (Crysis and Alan Wake come to mind).

Yet there are responses along the lines of "I can't believe this isn't reality!".

Either I'm not seeing something everyone else does, or people just believe whatever they're told ("this is uber next next gen!").

Somehow I'm waiting for a press statement from Square-Enix like "Haha! This is actually from our shelved PS2 game! So graphics aren't important after all!! Point proven!".

Can someone shed some light on what the hype is all about and why Square Enix chose to present their new engine this way?

[cowsarenotevil]

So on one hand, I don't see anything novel about the engine. In fact, it looks more or less just like lightmapping; it the lighting could be updated in real time, but since the lights aren't moving, I see no reason to believe that this is the case.

On the other hand, I do think it looks very realistic. It's well-done lightmapping, and the camera shake looks very organic (not that that's new, either, and it would probably get unpleasant in a real game), and the art seems to be based on something that really exists. Overall, except for what appears to be bad aliasing, I wouldn't necessarily know it was fake.

I can't help but think that they are, as you suggest, choosing to present it this way to prove a point, but I think the point is more that "realistic" artistic decisions go a lot farther to make something look real than adding a ton of "next-gen" features.

[DarklyDreaming]

So, let's see...
1) They haven't shown anything but boring static indoor geometry. Hmm. OK.
2) They haven't shown anything that couldn't be done with already available technology.
3) They show a fifteen second extremely compressed video of a camera walking down a ramp with no lights moving or anything otherwise noteworthy happening. Impressive huh?

I see no reason to be excited at this point in time. Want to see something impressive in real-time? Checkout CryEngine 3. Or Samaritan (<-- though that one was being played on a ridiculous beast of a computer!)
Yes, they announced that they're using DX11 (surprise surprise!) and that they are, basically, "doing awesome stuff". That video and the accompanying screenshots aren't much to be wowed over -- not until they throw together something at least resembling interactivity! If they wanted to play up this announcement they really should have waited until they had something to show; for what they have shown is "meh".

[SiCrane]

If you're going to get worked up every time someone online gets excited about something that doesn't seem that neat, you're going to give yourself an ulcer pretty quickly.

[Antheus]

This looks like unbiased rendering. So far, no engine has used it, either precomputed or real-time (so far not possible).

Main issue with this type of rendering is that it works ok for static camera. But AFAIK, there is real option for arbitrarily moving camera. If they somehow managed to do that, it would accomplish two things:
- next step in graphics quality
- uncanny valley will go to new depths

Latter point is relevant due to computational complexity. Static geometry will look nice, but everything else will look out of place. Shadows and other interactions, as well as subtle interactions will break the quality.

We had graphics like this years ago

Nope. FryRender is capable of this type of quality, but at vast expense of storage and only for fixed number of viewpoints. Mostly it's used for static images or precomputed animations.

Want to see something impressive in real-time? Checkout CryEngine 3

Nice, but artificial.


That said, low resolution video hides some details, seeing demo in 1080p would really reveal some details. Then again, VGA is the sweet spot that can be realistically achieved with today's memory sizes. For anything more, texture sizes become a problem.

[DarklyDreaming]

This looks like unbiased rendering. So far, no engine has used it, either precomputed or real-time (so far not possible).

Main issue with this type of rendering is that it works ok for static camera. But AFAIK, there is real option for arbitrarily moving camera. If they somehow managed to do that, it would accomplish two things:
- next step in graphics quality
- uncanny valley will go to new depths

Latter point is relevant due to computational complexity. Static geometry will look nice, but everything else will look out of place. Shadows and other interactions, as well as subtle interactions will break the quality.

[--snip--]

Yes, all true -- not that it matters.

My point was that they haven't done/shown that yet. So, regarding as to what they've shown, it's pretty, sure, but not really "revolutionary" -- not yet. When they've shown anything interactive, complete with a free flowing camera, then I'll reconsider how impressive this new tech of theirs is. By the look of things, it'll be another year or so before they do a proper showcase...

I'll hang back and see what they come up with whenever they come around to showing it off. For now I retain my position that this isn't news worthy and doesn't exactly scream impressive all over -- and the fact that they decided not to release a high-res video seems telling that they probably couldn't. Otherwise, it'd've been an obvious idiot move to not show things of in HD - after all, aren't they the ones touting how next-gen, HD-enabled, awesomesauce it is?

[cowsarenotevil]

unbiased rendering (...) no engine (...) precomputed (...) not possible

?

Unless you're working under a different definition of "unbiased" or "precomputed" than I am (or "engine"?), I don't understand.

[Hodgman]

From what I can see in the video, they've got a nice GI solution (baked or dynamic), and a physically accurate BRDF for their materials. I'd guess the most impressive feature here is that they've matched the real-world physical data very closely for their scene, such as index-of-refraction (~= 'specular mask'), reflection coefficients (albedo), etc.


Getting materials to match photo-reference like that isn't simple, especially with most engine's addiction to lambert/blinn/phong BRDFs...

[way2lazy2care]

unbiased rendering (...) no engine (...) precomputed (...) not possible

Unless you're working under a different definition of "unbiased" or "precomputed" than I am (or "engine"?), I don't understand.

There aren't that many unbiased renderers in existence and to my knowledge no game engine uses it o.O

What is your definition of the terms?

That said, I was more interested in their press release when they said that the engine was easily scalable to large and small projects. I'm interested to see more of what they mean by that. It seems like it's largest piece of interest could be in cutting costs with high performance rather than it having the highest performance of any engine.

[froop]

If you're going to get worked up every time someone online gets excited about something that doesn't seem that neat, you're going to give yourself an ulcer pretty quickly.

I get worked up over a lot of internet-BS. I need to stop reading it

This looks almost realistic to me at a first glance:




The Square-Enix screens do look almost identical to the photos, but to me the reference photos they used look rather surreal in the first place.

[cowsarenotevil]

There aren't that many unbiased renderers in existence and to my knowledge no game engine uses it o.O

I was of the impression that path tracing was a common method of pre-rendering static lighting for game environments. Even if it's not common, it's certainly not "impossible" since there are a fair number of unbiased renderers out there and it's by no means impossible to use precomputed data from such a renderer in a game engine.

[Ravyne]

The new battlefield looks *good*, but it hardly looks 'realistic' -- take a still from battlefield and a real scene with similar content and see how easy it is to tell the difference. Take a scene from the square engine (preferably one without that aliasing artifact) and put it next to a photo of the reference material. Now what?

Photo-realistic lighting has been around for awhile, but not in real-time, and much less in game engines. Perhaps in that way the technique isn't 'revolutionary', but putting whatever techniques together in the way they have certainly has some fine-looking results. They haven't shared much about it, but real-time, dynamic GI -- or something approximating it with such good results -- would be a rather big achievement, especially if its fast enough to handle large, highly-detailed scenes.

Also, Square isn't really in the business of graphics technology -- They're not iD or Epic who have to keep showing new things off because its their business -- that fact alone makes me think more than anything else that they are doing something novel with this, otherwise I can't think of a reason they'd bother showing it off.

[MJP]

Their GI looks great, and I don't see why anyone would downplay that. Getting a realistic GI bake and coupling it with balanced materials is a lot harder than it looks.

[Sirisian]

I just got the uncanny valley feeling from that video. They must be doing something really good.

[cowsarenotevil]

Their GI looks great, and I don't see why anyone would downplay that. Getting a realistic GI bake and coupling it with balanced materials is a lot harder than it looks.

Yes, absolutely, but like I said earlier, that doesn't represent any new technology, just good artistic decisions.

[Hodgman]

Their GI looks great, and I don't see why anyone would downplay that. Getting a realistic GI bake and coupling it with balanced materials is a lot harder than it looks.

Yes, absolutely, but like I said earlier, that doesn't represent any new technology, just good artistic decisions.

Simply having good artistic directions isn't enough to get realistically balanced materials -- it requires R&D on the tech side too.

For example, if you're using lambertian diffuse in your BRDF, your angular falloff is going to be completely wrong (something art cannot fix) due to (e.g.) not taking microfacet inter-reflections into account, among other things. If you don't even know what BRDF you're using, then no matter how good your art is, you'll never be able to perfectly match photo-reference for all viewing/lighting situations. Or if you're using blinn-phong for specular (or worse, plain phong), then your highlights are probably just guesswork, which will never look quite right. Does your specular function take a micro-normal distribution into account? Is your art based on real-world refraction-indices/reflection-coefficient measurements? Are dielectrics and conductors shaded in the same way? Is your BRDF normalized and energy conserving?

If these kinds of questions are unanswered in your renderer, then even with the best artists, you won't be able to reproduce the real-world physical interactions of light and material.

[cowsarenotevil]

I'm not downplaying the amount of work that a renderer needs to do, and if you don't have a renderer that gives you the flexibility to represent a range of real materials then of course you'll be limited artistically. I am saying that there's no evidence that there's anything new here with respect to baked lighting, nor does there appear to be specular lighting of any sort whatsoever.

[Hodgman]

nor does there appear to be specular lighting of any sort whatsoever.

There's balanced specular on everything in their video, and it's what makes it look realistic. Pick any pixel on the wall in the video and track it's motion - the colour will change.

[cowsarenotevil]

There's balanced specular on everything in their video, and it's what makes it look realistic. Pick any pixel on the wall in the video and track it's motion - the colour will change.

I'm not at all convinced that the video shows this; I tried your suggestion (admittedly fairly briefly) on the regions I could find with the sharpest angle changes and I didn't really find anything that I could definitively say was not caused by the video compression. Additionally, I would expect there to be some kind of simulated exposure adjustment as well; you'd have to sort that out from actual specular lighting as well.

Maybe you could point out more specifically what you're seeing in the video (or reading about the video?) that demonstrates this?

[way2lazy2care]

I'm not at all convinced that the video shows this; I tried your suggestion (admittedly fairly briefly) on the regions I could find with the sharpest angle changes and I didn't really find anything that I could definitively say was not caused by the video compression. Additionally, I would expect there to be some kind of simulated exposure adjustment as well; you'd have to sort that out from actual specular lighting as well.

How did you not see it? If you watch just the right wall when there are angle changes you can see it pretty clearly.