How to composite
a 3D model onto a photograph
(and get away with it)
Part 1 - Preparation and Scene building
A friend of mine asked me how am I
making those composite renders, and since it seems that I've only been doing
those lately, I decided to write a small tutorial on the subject. I'll be
showing you my method, which is slightly different than what Andrew Price
showed on his Blenderguru site. A good thing in doing CG work is that there's a
multiple ways to achieve the same result, so consider this as just another way
to do so. In the end you're free to use whatever method you find most
comfortable to work with, as long you're happy with the result.
When
compositing rendered 3D models onto a photograph there are a few things you
need to consider to get a believable result. I tend to think of all of them to
be important (more or less), so in that order, here they are:
-
lighting
-
perspective
-
shadow
-
color
match
-
noise
Lighting
This
one is pretty obvious. Set wrong light angle and/or direction and it'll stick
like a sore thumb. That's also the case with intensity and contrast, but that
can be tweaked in post-processing phase. Nevertheless, the more you can match
right away the less work you're going to need to do later.
Perspective
Perspective
is also important in a way if you don't match it correctly it'll be obvious
that the rendered object doesn't belong into the scene. While you can get away
with only close approximation it's always good to try and match the perspective
as close as possible.
I use two ways
to do that: first is EXIF information in a photo that I'm compositing into:
camera type (for sensor size) and focal length (you can also find aperture
size, ISO and shutter speed useful if you're using physical camera add-on, but
it's not crucial). Then I enter those values into camera properties in Blender.
After that, I manually position the camera where I need it to be.
Second is BLAM
camera add-on, which uses grease pencil to draw perspective lines on the photo
which are going to bee used to calculate focal length. It also positions camera
accordingly in world space (vertical and horizontal rotation), but I found it
to bee hit and miss. For best result use uncorrected photos, otherwise
calculation will not be accurate. I'll be showing you both methods in this
tutorial.
Shadow
Also, very
important, but not that hard to do. Once you get the light direction right,
it's just about catching that shadow for compositing.
Color Match
Rendered
element needs to match the rest of the photo colorwise. Discrepancy can occur
if you're using an HDRI which has slightly different lighting than the photo.
So, for instance, the photo could have reddish overall color, and the rendered
element ends up with slightly bluish tint. For more immersion you're going to
need to reduce blue color on the rendered element to match the photo. This is
easily done in post-process phase.
Noise
I
add noise for even more realism, because nothing screams CGI more than
perfectly sharp and clean photo. I found a good and simple method to add noise
that matches that in the photograph, and it can be tweaked as desired.
PREPARATION
Before
starting, obviously you need to have an idea what you want to do. Right now I'm
into airplanes so I'll be making a scene featuring one. So, first, we need to
find a model we want/like. For this tutorial (because I like it A LOT) I'll be
using Chris Kuhn's newest F-14D fighter plane model. Since it's a paid model,
you can download his free F-18 model from Blendswap (or any other you'd like).
Second,
we need a scene. Because aerial composite wouldn't feature a shadow (duh!)
we'll use a ground photo, preferably that of an airport or something. I have
found a good photo that matches my idea using www.allthefreestock.com. It's a
search engine for free stock stuff, like images, sounds, etc... that can be
used for commercial and non-commercial purposes. Check it out...
And this is the photo I found:
Photo credit: Willtron via
Visualhunt.com / CC BY-SA
It's
a Gibraltar airport runway. It's nice that there was an airport name next to
the photo, because it'll help us to match the size of the airplane to the size
of the runway. With no close objects for reference you can easily end up with
an RC model on the tarmac instead of full sized plane... ;-)
You'll also need some HDRI that matches the photo. This
one looks close enough:
http://www.hdrlabs.com/sibl/archive.html, Malibu
Overlook
It's
not perfect, there are more clouds in the sky here, but because the airplane is
not as reflective as a car for example, we can get away with it. I think the
sun's angle and the sky color are OK.
Finally
we can start creating our scene...
CREATING
THE SCENE
Now
we're coming to the real stuff. We're going to match the perspective of the
photo and set our camera first, so we can position the airplane to match our photo.
Add a background photo you downloaded and set its view to camera. Then add a
camera and switch to camera view. You can rotate the camera to be parallel with
Y-axis right now, since the one on the photo is also parallel to runway. Also,
set render resolution to match that of the photo to avoid distortions:
Now
I'll show you both methods I use for matching the perspective. I always use the
EXIF information first, whenever I can. If there's no EXIF embedded in a photo
only then I use the BLAM add-on. So first, the EXIF method. For this particular
photo the EXIF information looks like this:
We can see the photo was taken with Nikon D40 camera at 18mm focal length. Quick google search tells us D40's sensor size is 23.7 x 15.5 mm. Now we'll enter those values into camera properties:
The
perspective was OK, I just needed to rotate the camera manually so the grid
lines matched the markings on the tarmac. You can see green Y-line near the
right hand side black line. Perfect match.
NOTE: If you're using a physical camera
add-on you can also enter shutter speed, ISO and F-number values, which should
help with exposure. Without that we'll just eyeball the lighting intensity.
Now
the BLAM method. You can find the add-on here: https://github.com/stuffmatic/blam. Right now it
isn't really needed, but eventually you'll come across the image with no EXIF
info and this add-on will help a lot.
A
few things to consider: BLAM uses X and Y-axis on the photo to determine
relations to X and Y-axis in Blender scene. That's why we need to draw 2 lines
for X and 2 lines for Y-axis on the photo which BLAM will use when calculating
perspective. Because of that it's important that you find lines in the photo
that you know/can tell are perpendicular in real life, otherwise calculation
will be off.
First,
go into Movie Clip Editor in Blender and load our background image. Under
T-toolbar -> misc there are BLAM settings. In N-toolbar, scroll down to Grease
Pencil and add a new layer, change stroke color to red (for X-axis) and draw
two lines on the photo that match X-axis. Do the same for Y-axis: add new
layer, change color to green and find two lines to draw on. When you're done,
hit „Calibrate active camera“ in BLAM settings in T-toolbar.
The
add-on will automatically enter calculated focal length into camera properties
(fig. 3.). As you can see, it calculated focal length pretty accurately
(accuracy will depend on how precise you draw the lines, whether they're
perpendicular in real life and if the photo has significant lens distortion).
Obviously EXIF data will give you exact values, but without those BLAM should
get you close enough without tedious eyeballing.
Just to get the
scale right, I googled the runway size of Gibraltar airport (46m) and added a
plane with the width of the runway. Next, moving only camera, I positioned
calibration plane to sit between the runway edges on our photo.
That put blender
camera at around 1.75m height. I know that center of the runway is slightly
elevated compared to its sides, to facilitate rain drainage, so taking that
into account I could lower both the camera and the runway edges cca 10cm down
which would make things even more accurate but the difference doesn't really
show here so I left the camera as it is.
Now we are going
to import our model, so we have something to use as a reference when matching
the light. I decided to add two airplanes in really low flight above the ground
coming toward the camera. Looks better than featuring parked plane with nose or
exhaust tubes looking at the camera. Sideway shot would be out of place at this
spot on runway, IMHO.
I
made one of the airplanes break off formation, just to add some action to the
scene.
Now
we set up the lighting. In environment tab add 3k HDRI from downloaded ZIP, and
set up light intensity control in node editor like this.
After some
experimenting it was clear that with only HDRI we wouldn't be able to get
strong shadows and the overall light had too strong bluish hue, so I had to add
the sun lamp as well. I used shadows on the building on the left hand side of
the photo to determine the sun's position. The way I see it, it's coming from
behind on the left, at roughly 50° angle. Luckily, sun on the HDRI is at
similar angle, and it was just a matter of rotating HDRI to correct position.
Use „World background“ under Display properties in N-toolbar to see the
environment texture in viewport.
I adjusted HDRI
strength to be just enough to get some environmental light on the airplanes and
not to be too strong. All of this tweaking will depend on the lighting on the
photo you're compositing to, and the HDRI you're using. The sun ended up with
strength of 10. The ground will probably be overblown but it looks right on
airplanes so for now I left it that way.
Before
we finalize the lighting part, we'll add a ground plane to provide bounced
light from the ground to light the airplanes from underneath more realistically,
and later to serve as a shadow catcher. The way I like to do it is, I add a
plane large enough to cover the ground around the airplanes, UV unwrap it from
view and use our background photo as a texture in color slot to make bounced
light have color of the ground, instead of just giving it flat dark gray color.
Here, we could probably get away with it, but when you're going to have more
varied ground in terms of color and details, flat color just won't cut it,
especially when rendering highly reflective object like a car.
Now
add large plane, apply scale, in edit mode subdivide it at least 40 times (W
-> subdivide), and unwrap it from view. The reason we're subdividing ground
plane is, for some reason Blender distorts the texture if the plane isn't
subdivided. I don't know why is that happening but you can see bellow how it
looks with and without subdivision. If the texture still looks distorted
subdivide it some more and again unwrap from view. Here, I used 60 subdivisions.
It's
still a bit distorted, but it doesn't matter that much. I also added glossy
shader to make the ground some more reflective. This step is probably overkill
but I still like to do it for added realism. This is my final setup:
As
I thought, the ground was to bright, so I had to lover the texture's value to
more reasonable level. Now reflected light looks good on airplanes and the
ground plane won't be rendered anyway so I consider this done. We're ready to
render the airplanes!!
Continue to Part 2 - Rendering and Compositing
