Toy Story versus... *spawn

[milk]

The comparison is relevant as it allows us to see what makes the most differenfe. The original toystory has models tessellated to subpixel levels and perfectly highres textures, yet, for the avarege public it would look better if made with the KH engine. Goes to show how lighting and shaders make a difference.

 

[zed]

The original toystory has models tessellated to subpixel levels and perfectly highres textures

But thats the thing, some of the textures are low res and you can see the polygons in some objects.
also 'models tessellated to subpixel levels' doesnt really make sense as this is true for any renderer (some LOD stuff excepted), as it depends on the distance from viewer to object.
The only thing I can think that toystory beats todays stuff is the reflections and even this I don't know as can't see any good screenshots of non flat reflections, hell for all I know they may just be using spheremaps. Had a google couldn't see any good reflection screenshots
did find this from toy story 2 though (which is a marked step up qualitywise from toy story 1)

 

[Shifty Geezer]

also 'models tessellated to subpixel levels' doesnt really make sense as this is true for any renderer (some LOD stuff excepted), as it depends on the distance from viewer to object.

Toy Story's objects are defined as NURBS which are tessellated to triangles at the sub-pixel level. In typical renderers, and virtually every game, models are defined as triangular meshes and only have so many triangles as the object has. Tessellation and displacement mapping could solve that but isn't use much because of issues. Thus games have triangles far big than per pixel resulting in visible corners, where Toy Story doesn't (unless there was a glitch in the renderer).

 

[milk]

The only thing I can think that toystory beats todays stuff is the reflections and even this I don't know as can't see any good screenshots of non flat reflections, hell for all I know they may just be using spheremaps.

Toy Story was rendered with a rasterizer, not a raytracer. Reflections were achieved with scondary cameras assigned to planes or with sphere maps as you mentioned. All hand placed by the lighting artists. Pixar only moved into a globally ray-traced solution very recently (late compared to other AAA animation studios) with monster's U.

 

[zed]

Toy Story's objects are defined as NURBS which are tessellated to triangles at the sub-pixel level

All of them?, eg the stirups in the above picture, also part of the bedspread looks like displacement mapping (though perhaps it isnt), the big difference in a film vs game, is with the film you will always know how far away the camera will be.
Also did renderman render with nurbs (perhaps it did), but whenever I have used nurbs, the engine would first convert them into triangles. then again 1995, some cards still rendered polygons back then (i.e. not triangles only)
Yes perhaps there are more triangles in the above toystory render, than if you rendered it on todays PC/console (but that would be more due to an in-optimal implementation, just like I could render a sheet of flat paper on todays PC, with more polygons than a typical toystory scene)

I'm surprised no one has grabbed the assets from toy story 1 (converted them into a 'nice' format), imported them into one of todays engines, and done an A->B comparison (image quality and framerate)

 

[Laa-Yosh]

As far as I remember, Renderman's geometry primitive at the time of Toy Story 1 was called a "grid" and it was some sort of quadratic B-spline patch. Even polygons were converted to grids for shading and rendering IIRC, and of course NURBS surfaces were diced to grids as well.

Most of the geometry pipeline was basically a "divide and conquer" loop, where all the geometry primitives were split and re-processed again until their size went under a predefined threshold.

This had a few consequences for the rendering:
- displacement mapping was almost free because everything was tessellated to subpixel size anyway
- shading was only calculated for the grid points and interpolated from there (so it was sort of a vertex shading approach )
- shader antialiasing was linked to the level of tessellation (it was called "shading rate")

Rendering was also done by "buckets" or what we call tiles in realtime rendering. This was to keep memory use manageable, as only the grids contained in the bucket had to be kept in RAM.


The above is also why there's no point in making any comparisons using the original assets. Except for PRMan, no current renderer would be able to process them the same way. You can of course do any kinds of custom tessellations, and render them as triangles, but it wouldn't be a close enough match; and tessellating to the levels that was used in the movie is useless anyway, as all renderers do shading in different ways.

However, Pixar has implemented Catmull-Clark subdivision surfaces by the time they started on Toy Story 2 (after all Ed Catmull was and is their CTO...), and thus they converted all the NURBS assets to subdivs for TS2. The control cage of a subdiv model is not really different from a polygon mesh, and every software renderer today has support for subdivs; even many realtime engines should be able to support them.
PRMan still tessellated subdivs to grids (and does so today as far as I know) so what it actually renders was just as different, but at least it would be a closer comparison.

 

[London Geezer]

You knew it was coming...

 

[Nesh]

You knew it was coming...

Its like Eurogamer stuff are also reading our forums

Another interesting comparision would be games today vs Final Fantasy Spirits Within

 

[milk]

Its like Eurogamer stuff are also reading our forums

Another interesting comparision would be games today vs Final Fantasy Spirits Within

By the amount of technical inacuracies they tend to comit (albeit minor) they are not reading enough.

 

[London Geezer]

Still, they pretty much agree with what I and a few others are saying here. Compared to the first Toy Story, you can see the improvements and how the results can look pretty similar considering the imperfections of real time rendering. But from the second one, CGI just went far ahead to what we can even fake these days.
Still, considering a rendering time of 33ms compared to unlimited rendering time, current real time is very impressive.

 

[Karamazov]

tech aside, if i showed my girlfriend games like the order or UC4 she would say they all look better than the old toy story movie.

 

[Jupiter]

The results are good but STAR WARS BATTLEFRONT (2015) on the PC in UHD, Ultra settings and 60fps looks much better. And that even works in Endor below 16.6ms with one GTX 1080Ti.

On current hardware with one graphics card STAR WARS BATTLEFRONT comes the closest to CGI in my point of view. With even faster hardware it would be Wildlands or The Division. The latter also has Hybdrid raytracing for shadows. In general it is always important to keep artifacts to a minimum.

I personally dont find 30fps beautiful to look at.

 

[Clukos]

I'm starting to think that you are a troll Jupiter

 

[Jupiter]

People from many forums are asking if games have reached the quality of old CGI movies. Some say for example Ryse, The Order 1886, Ratchet and Clank or Uncharted 4 have done it the best so far. And that's where I say that STAR WARS BATTLEFRONT stands out. Overall, the picture in BATTLFRONT appears very quiet. Despite the partial very good graphics and effects there are only a few elements which distract. I don't know no one game with such a good performance optics rate. Even such a dense forest as Endor with 39 other players/AIs/vehicles plus falling trees, lots f explosions etc. runs with an exeptional high performance. Unfortunately there are no softshadows. Too few games have these.

On the one hand STAR WARS BATTLEFRONT was released in 2015 while on the other hand Kingdome Hearts III will appear sometime in 2018. Nevertheless STAR WARS BATTLEFRONT looks more impressive which is why I do not understand the hype surrounding the graphics of Kingdom Hearts III. Still the visuals are good because of the appropriate use of the Unreal Engine.

 

[Karamazov]

because of the ps2 era and the claims of "toy story graphics in real time". Since then we have been waiting for such a comparison.
KH3 is not even the best looking game on ps4, but here we can directly compare it to the Movie, and that's what is interesting.
Achieving an overall similar look with realtime constraints and tricks.

 

[Shifty Geezer]

because of the ps2 era and the claims of "toy story graphics in real time".

Can anyone actually find a source for that?

 

[London Geezer]

Can anyone actually find a source for that?

It was you. You said it almost 20 years ago. I remember like it was yesterday. It was the beginning of the era of fake news. And look where are now. You ruined everything.

 

[Laa-Yosh]

Toy Story definitely had no ambient occlusion. The technique wasn't even invented until Pearl Harbor and Jurassic Park 3, in 2001.

I could go on and nitpick the video for a while longer, but it can be summed up quickly - the DF guys have not done their homework properly

Okay just got to the motion blur part and it can get worse. PRMan's motion blur is still a lot better than any post effect, as it is proper 3D motion blur. Whoever wrote this analysis, he/she clearly has very little idea about the particulars...
Okay, now the same with AA. Seriously, it's not that hard to read up on how PRMan worked, instead of trying to describe it with completely different techniques in current real time graphics...

Rendered on PC workstations?? Those were SUN machines as far as I remember, ohh this is just sooo baad...

 

[Laa-Yosh]

Okay, so a few things about post vs. 3D...

Post motion blur can not handle anything moving along a curve, it's linear (at least what I'm familiar with), and it does not affect shadows, or reflections. Pixar's 3D implementation handled it all, and it also had a 3D DOF implementation.

On top of that, the undersampling that the video points out was actually a feature - render quality settings were highly flexible and so it could be used to make motion blur almost "free". I'm quite sure noone would have noticed the grainy blur during fast movement.

And I've already mentioned how PRMan has a super high level of shading and geometry AA that no realtime post-AA or temporal stuff can even dream to match.

 

[Nesh]

because of the ps2 era and the claims of "toy story graphics in real time". Since then we have been waiting for such a comparison.
KH3 is not even the best looking game on ps4, but here we can directly compare it to the Movie, and that's what is interesting.
Achieving an overall similar look with realtime constraints and tricks.

To be honest Toy Story graphics were our expectations. Actually anything close to the CG quality of PS1 games could have been missinterpreted as Toy Story by excited video game fans. Me and my friends back then where these people. Hell I was calling the Dreamcast a Toy Story graphics powerhouse the very first time it was announced. Then when the PS2 was showcased with real time tech demos of GT, Ridge Racer, Final Fantasy (FF8 + the old man from the Spirits Within), the Bouncer, Tekken etc, for many teens that was superb CG quality graphics. Sony literally used assets from the CG scenes
The term Toy Story graphics officially used by a company at a press conference was not by Sony but by Microsoft when they showcased the XBOX for the first time using CG target renders of the dancing robot and Malice. Nintendo was also using the Rique Squadron footage to hype the GC by making direct comparisons scene by scene with Episode 1. And indeed for many of us the death star dog fight was very close to almost indistinguishable (not by today's standards)