Milky Way

Started by Dune, December 17, 2019, 01:49:06 AM

Previous topic - Next topic

Dune

Trying to get a nice Milky Way... WIP's for now. Not at all satisfied yet.

René

Are the stars procedural on the background shader? I tried this and the problem I encountered was that the look of the stars depends on the render resolution; the stars seem to get smaller at a higher resolution. That might have something to do with the distance of the background shader, I have no clue at what distance it is.. I'm curious what you'll come up with.

Dune

Yes, and you are right, I encountered the same. Adjustment may be needed per render resolution. I tried Milky Way and stars on a very high thin 2D cloud layer too, but that looked like cloud. Doing my best now not to have the MW look like cloud.

René

Maybe a large sphere instead of the background object could do the trick, I have never tried that.

WAS

#4
When I did my Terragen galaxy (at angle we would see the Milky Way) that was the biggest concern was star scales. They need to be fairly large, actually. Even then, how those shapes are handled by AA is always going to be a bit diff. Especially though MPD is going to effect the "roundness", and you need to use Voronoi Billows, with no variations.

Are you using my old share as a base? I notice star clumping which was occurring for some reason opening my old project in TG 4.4 The noise settings are interpreted differently, and the colour map got messed up.

bobbystahr

something borrowed,
something Blue.
Ring out the Old.
Bring in the New
Bobby Stahr, Paracosmologist

Dune

The background object is a large sphere, so I don't think that's the problem. It would seem easy to have some masked fractals cover the northern hemisphere, just like you would on the planet (but the with huge values). I do use voronoi with settings that make it perfect spheres, but also another setup with some blue nodes that makes perfect spheres (adding two Voronoi 3D A vector nodes). Stars seem to cluster somehow, which might be due to 256 variations limitation, but with some fiddling you can get a nice spread. Even using 100m range of starsize, though it seems that if you extend that to 1k or bigger, you don't see much difference. Shifting color offset just a wee bit will get either a very dense dotty galaxy or occasional small dots, hard to subtly increase number of stars. I will experiment further.
The MW itself is the problem; it often looks like a cloud, getting it to fade out is pretty hard, even using smoothing and soft SSS for masks.
And indeed, Bobby, that's where I looked for ideas. Some great images to be found!

Dune

Spent some more time on this, but should also spend more time on a believable landscape some day. But this is it for now.

KyL

Really cool experiments. I'm going to have to dive into this at some point too!

The great thing when using a fractal is that you get a correct behavior as the stars should be a tiny tiny point of bright light in any given resolution. Even with the highest maps I could find (16k), it doesn't really work.
But that means you will have to be sampling things like crazy and it is very likely going to flicker in animation.
Still a very interesting challenge, I am following this closely!

WAS

#9
Quote from: KyL on December 19, 2019, 12:09:33 PMReally cool experiments. I'm going to have to dive into this at some point too!

The great thing when using a fractal is that you get a correct behavior as the stars should be a tiny tiny point of bright light in any given resolution. Even with the highest maps I could find (16k), it doesn't really work.
But that means you will have to be sampling things like crazy and it is very likely going to flicker in animation.
Still a very interesting challenge, I am following this closely!


If your stars are actually big enough (you should visibly be able to see roundness form the voronoi billow) flicker should be minimized. The flickering is from having far to small of stars and what you're seeing is only there because TG is trying to sample "something" of whats there.  Additionally of course, stars twinkle and flicker with spectrums of light reaching earth inherently. One of the most basic principles lacking from many Scifi. Likely cause static maps save time, and are easier, and people (viewers) shouldn't be nit-picking a pixel (will likely blow right over them). xD

Further more, panning stars even in films on HD TVs look like crap in general. Lol They flicker, streak, interlace, etc.

Oshyan

These kinds of procedural starfields unfortunately don't work well with adaptive sampling (including and perhaps especially the new Robust Adaptive) because they rely on consistent sampling of extremely tiny and bright areas of a noise function, and sampling inherently has randomness, so unless the sampling algorithm can recognize that this tiny, bright area is *desirable* rather than an outlier result, then it will not adequately sample it. This is very difficult for any adaptive AA algorithm to handle because such bright points are generally undesirable in normal renders, often called "fireflies". For the same reason procedural starfields also tend to be resolution-dependent even without adaptive sampling, i.e. the results will change if you change render resolution.

- Oshyan

WAS

Quote from: Oshyan on December 19, 2019, 04:54:51 PMFor the same reason procedural starfields also tend to be resolution-dependent even without adaptive sampling, i.e. the results will change if you change render resolution.

- Oshyan

I think this is just inherent to what we're dealing with, especially with faint stars no larger than a pixel even on HUBBLE sensor or otherwise. Or even say, downsizing a star field at a large resolution (always a issue when dealing with compositions in the past with PS being pixel based). Like i was getting at before in my previous post. You can see this for yourself in various formats. VHS 240-480p to HD resolutions. 35mm releases to Blu-Ray/DVD, and again, trying to pan such images on different mediums/screens. Even TV's have issues panning these across LEDs and the like when they're dealing with pixels, especially smart TVs and their always defaulting to noise reduction, resampling, interpolation, etc. It's not just rendering. The result is always going to vary across formats and mediums. And again, even filming stars, we get twinkling, dashed stars, vanishing stars, etc in overexposures, etc.

Dune

Not satisfied about the trees. First of all it's only one type, ugly roots and snow is not the same node set as other snow. Also have to mask out the highlights on tree snow from further away.
MW is not good enough either yet. I set the central interstellar gasclouds blocking the starlights to -0.05, but that's too much.
Defer MPD=0.5 AA 6 default.

WAS

#13
Quote from: Dune on December 22, 2019, 02:13:50 AMNot satisfied about the trees. First of all it's only one type, ugly roots and snow is not the same node set as other snow. Also have to mask out the highlights on tree snow from further away.
MW is not good enough either yet. I set the central interstellar gasclouds blocking the starlights to -0.05, but that's too much.
Defer MPD=0.5 AA 6 default.

Seems to be coming along very nicely though. I see what you're doing with the snow and it looks good.

One thing I'm finding hard to do with the milky way, is getting a good balance between blurry (less noise) shapes and more defined shapes like seen many images

Additionally I notice going from a soft 1.5 noise to 5 area, coupled with Robust at 1/4 jumps render time significantly. Probably better to work with default 1/64 anyway.

sboerner

QuoteNot satisfied about the trees. First of all it's only one type, ugly roots and snow is not the same node set as other snow. Also have to mask out the highlights on tree snow from further away.

MW is not good enough either yet. I set the central interstellar gasclouds blocking the starlights to -0.05, but that's too much.
Perhaps, but steady progress all the same. Impressive.


Maybe reduce the glow on the horizon? A light source like that would overwhelm our perception of the MW.