The future of 3D creation: on the web?
fxguide.com
fxguide.com
I saw a demo of Lagoa[1] this week which would interface very well with this type of product (clara.io). You could have both the creation and materials / rendering without the burden of your own render farm and without the cost of some of the high-end software normally required.
In the demo I saw, it was pretty amazing to see an accurately rendered sneaker being moved in real-time with its colors being changed at the click of the button. Even over the slow wi-fi, results streamed immediately and progressively got more and more detailed.
We've implemented streaming progressive renders using V-Ray (a premium and very popular renderer in the visual effects and architecture world) in Clara.io in a similar fashion to Lagoa. Here is a video of how it works in Clara.io:
http://www.chaosgroup.com/en/2/news.html?single=544
We haven't launched this feature officially, the above is a four month old tech preview, but that will happen soon.
Make a change? Cancel the job in the cloud and send a new one. Use a map-reduce to build output movies an just have them available when done. "Render" becomes "play when ready"
I think the 3d content creation world is ripe for disruption.
When in interactive rendering mode Clara.io does constantly render after each batch of changes, but it is more than 20x as expensive as not rendering in terms of server costs.
We want to make Clara.io as low cost as possible and one way we can do is to not have it rendering all the time. :) But if you have deep pockets go for it!
The same concerns about cost apply to automatically creating the movie on each change. Yes it is possible, but the cost for computing that may not be required is pretty high.
Things like, progressive enhancement rendering -- either quality or resolution -- like wavelet compression for rendering. Massively parallelized rendering pipelines -- borrowing from the high performance password cracking on GPUs, and real-time game inspired rendering pipelines. Static scene analysis/dynamic compositing; rather than require the artist to think about layering and compositing, let the computer algorithmically break down a scene and maximize rendering speed.
Perhaps it will eventually be determined to not be cost effective -- and I'm sure people much smarter than I are working on it -- but I think we will see some really interesting and novel changes in this space in the next 5 years after being relatively unchanged for the past 20.
And just for the record, I think Clara.io is fantastic, not saying you guys are doing anything wrong, just talking. :)
For now we just have high quality viewport that is similar to game quality graphics, that is super cheap because it leverages the local GPU rather than cloud resources.
Turns out, if you want high-quality(All quality is relative) rendering that doesn't take much time at all, we can do it any modern game engine can output decent image in about 30 ms. And an even better one still quicker than a human can react. And there are tons of people which that is quite sufficient.
Videos from source filmmaker show that: http://www.youtube.com/watch?v=WUhOnX8qt3I http://www.sourcefilmmaker.com/
Also see UE4 infiltrator: http://www.youtube.com/watch?v=-bLOi3mo9NE
* I use always lightly here, that is of course not true on certain time scales.
> Things like, progressive enhancement rendering
This is a very common option in a huge number of renderers. Particularly physically-based renderers It particularly useful when you want interactive feedback, less so for 'final' renders.
> Massively parallelized rendering pipelines.
Rendering is one of the original problems that generally fit into the "embarrassingly parallel" category, hence why render farms are even a things. There are very many existing offline renderers that take great advantage of GPU computation. Paired with progressive enhancement, certain renderers use this particularly well for interactive editing. But it's also used in very many a render farm.
> Static scene analysis/dynamic compositing ... let the computer algorithmically break down a scene and maximize rendering speed
This has been done since likely the day rendering was invented (not actually). The artist only thinks about layering and composting when the artist wants specific composting control in post. And there is no way you could do that as an automated process, just as you can't do the rest of the art as an automated process.
> Perhaps it will eventually be determined to not be cost effective
What won't be cost effective? Those techniques? They already are! Or cloud-computing for offline rendering? It already is as well! Some of the first clouds as a service were and are put to great use for offline rendering from the very start. And a 'render farm' is really just a 'private cloud' for rendering.
> after being relatively unchanged for the past 20.
This is confusing to me. Rendering is a field with one of the most progress every year. You only have to go to Siggraph once to see that as self-evident.
Anyways, I'd love to talk more or answer any other questions you have. Rendering has been a passion of mine for over half my life. And anything that has the ability to transform how we do things is very interesting. Just as Clara.io very much is.
Granted, Verold does not handle 3D _creation_ on the web. Instead we are building a platform for creating interactive experiences using 3D content. Eventually it will lead to a full fledged game development platform. Check it out!
I know the solution is to install 64-bit browser, but that doesn't seem to be the default for most of the popular browser on 64-bit Windows machines (or maybe other os's too)