Photorealism Tricks of the Trade


#1

Hey guys and gals. I thought this might be a cool thread to start. What are some of the tricks of photorealism in visual effects that you’ve learned over your studies? Here we can share this information. I’ll list a few.

1. Using energy conserving, therefore physically correct shaders.
2. Linear Workflow
3. Not screening or adding fog/mist, but composite as an occluding element.
4. Setting motion blur to 0.5 or 50% to mimic the 180 degree shutter of a motion picture camera.
5. Adding grain, and other optical phenomena such as lightwrap, bloom, vignetting, lens distrotion, etc. to "imperfect" the final composite.

I know these are pretty basic, but some people miss them in their studies.  Any other tips you guys and gals could throw out there would be appreciated.

#2

*I am still learning like everyone here, so I dont want to sound like the “authority” on photo real, but here’s is my 2c on the techniques i use in my own work:

For me it simply helps having an understanding of how things work in the real world. This can be a combination of visual analyses (eg staring a photo), and academic research(eg reading a paper how light works, and then working out how to apply it in 3d).

There is a big difference between realism(what we see with our eyes), and photo-realism(what a camera captures). I strive for photo-realism, it’s the imperfections you get that generally make an image interesting(cg chromatic abrasion, lens distortion, film grain).

Couple things that i think help sell photo-realism:

  • Correct light temperatures(eg sun is blue, tungsten(common studio lights) are orange ect).
  • Depth of field, especially those funky bokeh hi-lights.
  • Motion-blur (so many people seem to still skip over this)
  • Proper physical Light fall offs
  • Shaders that obey the laws of physics (i love mental rays mia for this).
  • Always use SSS for skin
  • Use physically correct lights (eg area lights)
  • A good GI solution
  • Where possible, render out everything in float (32 bit)
  • Lens distortion
  • Chromatic aberration
  • Lens flares (i like stirring up a bit of controversy)
  • Film grain
  • Agree with above, adding lights blooms/flare in post can help heaps.
    … well thats a start at least!

edit: Read some further posts, and remembered some stuff:
> Black is never black, it’s always slightly off
> The blue channel is almost always the grainiest when adding grain


#3

Working with a team of insanely talented people who know what they’re doing.

:wink:


#4

The best way to get your stuff photoreal is to get a photographic image in the appropriate format (ie the one you’re trying to match) and really, really look hard at the thing to see what it is you’re supposed to recreate and then work it out from there. Every format has its own idiosyncracies - there’s no hard and fast formula for matching them - the only guarantee of success is diligent observation of the source. At the end of the day, everything you do in a VFX image is an approximate imitation of reality so creative judgement as to deciding what matters the most in selling a picture is just as (if not more) important than the technical route that you choose.

btw: rendering at reduced res and then interpolating up to soften the image is fine for low res formats like HDTV, but I wouldn’t recommend it for film res work as it just introduces a whole bunch of additional filtering artefacts.


#5

Awesome thread!
Not long ago people here would of been up in arms over this thread.
Also be careful about mentioning chrommatic abrevation around here some people tend to think they are the authority on what effects can be used or not. j/k :rolleyes:


#6

You can also run your images through an unbiased renderer (indigo, maxwell, fryrender, busyray, sunflow etc) to get an idea of what the physically correct solution for your image is and then recreate that in whatever your actual renderer is.


#7

also, it’s chromatic aberration


#8

Leigh’s spot on, I learned a lot from working with matte painters. A couple of things really helped me out.

First was finding the right bits of reference. (flickr.com is your friend…) Sometimes it’s good just for sussing out what the finish line is, sometimes it’s awesome for generating textures. Not having good reference is like building a dog house without a blue print. It can be done, but with lots and lots of needless piddling.

The second thing they taught me was using data in image channels to cut out elements. Imagine you’ve got a picture of a tree with a blue sky behind it. If you look at the color channels, you’ll find an almost perfect b&w image that’d make nearly a perfect cut-out of the tree. Then you use the dodge and burn tools to perfect the alpha and use it as a mask. Often that technique is not only faster, but more reliable in terms of getting a nice cut-out.

Fun stuff. :slight_smile:


#9

cheat the audience eye for all cost! takes lots of shortcut :stuck_out_tongue:


#10

**I am interested in hearing more on this, or useful links on the subject.
I have just started rendering with 3d motion blur, and I left it at a default of 1, full 3d blur in mental ray, even though the moving object was going from the background to foreground.
It was a dark scene and looked ok to me, but would one normally adjust settings depending on the distance of the object from the camera?

I am concerned about having too much blur in my future scenes and how to determine the right setting.


#11

Thanks for the comments thus far. Keep them coming. :slight_smile:

Motion blur in a [motion picture camera](http://en.wikipedia.org/wiki/Movie_camera) is usually produced using a 180 shutter, and therefore producing "half" the motion blur of a full frame time.  Since film is usually recorded at 24 frames per second, the resulting shutter speed of a 180 degree shutter is therefore 1/48th of a second.

Taking that into account, the proper motion blur setting in mental ray or other renderers to emulate this, is 0.5 or 50% motion blur. 1.0 or 100% motion blur is not realisitic in regards to film cameras, because the frame is advanced when the shutter is "closed" during the next 1/48th of a second.  In other words, a 360 degree shutter or 1.0 motion blur is not possible, because there would be no time for the frame to advance.

EDIT: More useful info…

http://en.wikipedia.org/wiki/Shutter_angle


#12

Ok cool.
I will try it out.
Maybe it will reduce my render time too(i hope!).


#13

Hey Kelgy,

Motion blur in Mental Ray is a bit of a world pain (as in slow and noisy).

Nick


#14

**It took me 36 hours to render a 1000 frame animation with 3d motion blur.
But it was REALLY cool when I played it for the first time. It was my Jurassic Park test footage OMIGOD! Its got motion blur!!! moment.

edit: and its something I wanted to do since looking at adding motion blur in stop motion animation using vaseline on glass. I’d definitely something I’d rather leave to the computer.
:slight_smile:
I am re-rendering with the new settings and will compare the frames.


#15

Add a photoshop lensflare :cool:


#16

Hm, I’ve never heard of this one and I don’t think I’d see a happy supe if I handed anything over like that. Always render at the best quality you can, then downgrade, dirty up and blur from that.
Instead, try out some different AA filters and filter ranges. Some can add in a slight blur without mushing up the picture like uprezing might. Or blur in post.

Wiro


#17

I used to do it in TV where resources were scarce - on old analogue 4:3 PAL systems it produced acceptable results.

Working in film I did it across several shots on one show where it was the only solution to an otherwise impossible deadline. It didn’t look that great and I’ve been very careful ever since not to let things get to that position.


#18

From Christopher Nichols:
"Modeling:

  • bevel all the corners of buildings to catch speculars on the edge
  • even a perfect building is never perfect
  • only model what you see
  • model to a pixel accuracy not a 1/8th of an inch

Texturing:

  • UV unwrap everything
  • Provide high res textures for everything
  • Provide at a MINIMUM color, bump, and specular for every surface
  • dirt is pretty, dirt is detail, dirt is scale

Rendering:

  • Use high quality AA and texture filtering
  • use motion blur on every shot no matter what
  • Render at LEAST 10bit log cineon, if not full HDR or EXR format
  • Render LOTS of passes for compositing

Compositing:

  • Match the focus with a circle convolve (not just a blur)
  • Carefully match the film grain in all channels (blue is the hardest)
  • color shift all the blacks to match the color timed plate. Remember there is no such thing as 100% black.
  • Edgeblur all the CG elements to make the less CG in your scene. If you used proper fresnel reflection with your plates, you can reduce the amount that is needed."

#19

Being someone that would also strive to have photo realism and a keen photographer, this is my 2c’s.

I believe that you can have all the elements listed, and sure there are a lot of people out there that think that having all these elements would give you perfect results, but if we are honest, then we know that having them in the right proportions is ultimately key, i dont want to sound all philisophical about it, but its true.

You have to study life, take pictures, examine the real world and how pictures differ to get it right, thats where the talent lies.

You can have a tree and set of tools next to you, but you wont have a beautiful table and chairs if you dont have talent.


#20

Good post Daniel!

It’s not really a photo-realistic tip (well maybe it is), but one of the most common flaws I see, especially in student work is the dreaded “CG camera”. This goes hand in hand with “Editing, what the heck is that?” It just screams “This is 3D”

By this I mean shots where the camera is moving in a non-filmic way, doing crazy moves or never cutting (one camera for a lot of cuts).

Again, visualisation is key, watch some movies and notice how much the camera often doesn’t move, or moves in certain constrained ways.