Improved antialiasing in blender internal!

Finally blender internal have improved AA :slight_smile:

Holly **** this sound and looks really outstanding! I hope the final of Blender 2.40 is coming asap!

CMB02 :Z

nice… Gauss still looks best :wink:

gauss, for sure.
any measurement of rendertime penalties?

but it makes everything look blurry. if you want nice crispy images, render with mitchell-netravali [my favourite] or catmull-rohm [which has the pixar flavour!] :slight_smile:

by the way… when ton asked the ILM peeps this year at siggraph how they solve their AA problems; they just replied “oh we render everything twice the sice and scale it down!”
so i guess everyone has problems with AA, even the big guys… stupid square pixels!

edit: next step, adaptive sampling…


I think Orange is the best what could happened to Blender :slight_smile:
So much impovements in so short time.

but it makes everything look blurry. if you want nice crispy images, render with mitchell-netravali [my favourite] or catmull-rohm [which has the pixar flavour!][/quote]

It’s a shame you can’t use a mixture. That was one of my ambitious feature requests - automatic frequency based anti-aliasing. Why isn’t it possible to use a filter like Gauss or whatever on high frequency data (usually background data) and a sharper filter on low frequency data automatically? Actually, you might prefer that the other way round with high frequency data sharper depending on z-depth. Blender should be able to tell the variation in the measured samples per pixel and pick the best filter for the job. One other thing was choosing the samples per pixel automatically but you wouldn’t know how many you’d need until you started sampling.

Yeah, there are more reasons than anti-aliasing for doing that too. One is if you want to crop the images at certain points, you can still maintain full resolution. I think compositing is another reason. I’m guessing they use both anti-aliasing to a certain level and also scale it up to double the size. Their computers can handle that kind of punishment.

I think some renderers still cope pretty well at low sampling rates though. When you render with a Renderman renderer on default setting, the aliasing is mostly zero. I guess that might have something to do with micropolygon rendering and it might be due to stochastic sampling too.

That’s what I was hoping the new improvements were but this is a good development. Maybe allowing custom filters would have been even better where you just supply your own mathematical function but like I say, it’s always good to see improvements ;).

Gotta agrree with mister EyeCandy

Love that Catmull rom! Keeps a lot of the details for postpro later on.

Gauss seemed really good for lines and such, although it was a pretty limited test setup.

however it made the textures look like arse… why wouldn’t it be possible to use a seperate AA filter for textures and polygons.

as for rendering at 2x the size… wouldn’t that be easier to do (depends if your 2x resolution is AAed or not i guess)


i agree, gaus just blurs everyting and kills all fine details in textures.

great we have other settings now as well.