Collada plugin for After Effects

More info here
Completed Projects > Jutsu

Fusion Compositing

Images of Fusion flows created for the exercises in the book "Digital Compositing for Film and Video" written by Steve Wright

Luminance = 0.30*Red + 0.59*Green + 0.11*Blue
Luma key on black, white and gray
Custom chroma keyer
value = square root(sqr(r1-r2) + sqr(g1-g2) + sqr(b1-b2))
(r1,g1,b1) - normalized target color
(r2,g2,b2) - color from image
Bump map
get lum, blur, subtract small amount and then subtract from original lum
Using the built in keyers
raw matte = Green - Red
raw matte = Green - max(Red, Blue)
raw matte = g-max(r,b)
with color curve to scale it, gamma slam (not shown) to check for excessive black or white pixel noise, preprocess green screen
Screen leveling
use gradients to adjust for uneven lighting, gradient starts at color (0,0,0) and is added to the green screen matte is pulled using the custom distance map
Screen correction using a GS, a clean GS and solid GS
clean matte is the final matte but the correction matte will often also work in many cases
Garbage matte
combines blue only(none of the girl) and core(only the girl no blue at all) to get resulting garbage matte, loose matte around girl preserve details and then core taken out
Filter the matte
blur raw matte grain suppression then scale with color curve to reduce noise
Adjust matte size and control edge softness
expand contact via blur, color curves and built in erode/dilate node
soft edge details and spill, hard edge no spill and lose of detail
Despill - green limited by red,
new green = if green > red then red value otherwise green or
spill map is green minus red, and despill is green minus the spill map, subtracted values don't go negative is key
Despill with green limited by blue instead of red
Despill - blue limited by average of red and green,
green limited by avg red and blue not shown
Unspill - GS add red and blue subtract green,
for BS add red and green and subtract blue,
all values scaled by a color curve, subtract a lot then add little of the other 2 colors back in
Composite = matte*foreground + (1-matte)*background,
scale the matte for transparent effects
Processed Foreground compositing(Ultimatte)
not as easy to understand as the classic method, but it "works"
Add-Mix composite
classic composite with matte attached to color curves to scale them
Edge blending
use edge detect on matte use as a mask and blur final result within the edge itself
Light wrapping
create an internal edge mask and screen the background color over the foreground, may need a blurred version of background for more even effect
Soft comp-Hard comp
soft matte preserves detail, hard matte for core of the foreground, first soft composite with some transparency then hard comp with no transparency and soft comp as background
Premultiplied CGI images
images that are already scaled by their alpha channels, color correction needs to be done on images without premultiplication, un-premultiply by dividing image by it's alpha channel(not perfect solution but if color correction is minor is not an issue), highlights can be a problem if color exceeds alpha channel values- best solution is separate highlight(specular) render
Screen operation
for forground light emitting elements that don't block background light, 1 - (1-imageA)*(1-imageB)
Weighted screen - create matte from foreground to suppress the background values, this is the scaled background
Image blending
multiply is like projecting an image onto another image, maximum and minimum
Photoshop blending modes - blend is over the base
Overlay - if (base < 0.5) then 2*base*blend otherwise 1-2*(1-blend)*(1-base)
Hard Light - same as overlay but base and blend are reversed if (blend < 0.5 ) then 2*base*blend otherwise 1-2*(1-blend)*(1-base)
Soft Light - if (blend<0.5) then 2*base*blend + (base*base*(1-2*blend)) otherwise sqrt(base)*(2*blend-1)+2*base*(1-blend)
soft light equation in book is incorrect and when corrected is close to the photoshop version of soft light
Slot gags
2 high contrast objects, animated and one masking the other
Brightness and Contrast
use color curve set black and white points then curve line to adjust midtones, can use an s shaped color curve to increase contrast without clipping values
Color matching - sample what should be light and dark gray regions in background, find values that add/subtract rgb to make true gray and use to find percentages, the percentages +/- can then be applied to the foreground to make it match the background
Shadows - shadow density, faux shadows
Haze - adjust brightness, contrast and blur to match results of haze
Nonlinear gradient - regular gradient passed through a nonlinear color curve
Glow - make heavy blur and screen with original
fake camera defocus expand/dilate with blur or pixel shifts with a final blur
if focus changes must change both blur and shift amounts
Sharpening - unsharp mask = (1 + brightness)*(image - brightness*blurred image)
Lens flare - screen operation, chage colors by swapping rgb channels
blue is grainiest of all, use noise with blur to simulate grain, grain is combined by adding and then reduce the bring brightness back down
Adding grain - match grain practice
add grain plate to image and subtract difference to adjust brightness back to original overall value, adjust the RGB grains seperately to match greenscreen grain to the background grain
Adding grain
add grain to a grainless title to match the background grain, final blur on final text so blends in better
grain averaging(not shown) take 5 frames and average them together = no grain
2D transformations - translate, crop, rotate, scale, corner pinning
3D transformations(not shown) - similar but use DVE transform for Fusion
Filtering exercises - bicubic, bilinear, gaussian, mitchell...
Lining Up images
add inverted image to itself and transform until disappears, using offset mask to create clean plate, edge detection lineup = image and target image converted to BW images and edge detection applied, then they are combined in channel boolean, target image to red, and image to green and blue=0, now transform the image by "moving green to red", pivot point lineup procedure for moving and scaling one item to fit another
Motion Tracking
single point and muliple point motion tracking with and without obstacles, track image corners to moving monitor sequence
Stabilizing - make it still or smooth the motion by smoothing the animation path
Gamma effects on pictures
3:2 Pullup
24 fps film transferred to video will do a 3:2 pulldown, pullup undoes this to remove interlacing which is bad for filtering (transforms, blurs, sharpening)
Non-square pixels - NTSC resize 720x486 to 720x540 to make pixels square
Frame Size and Aspect Ratio - resize ntsc 720x486 to any 1.33 aspect will make it square
cropping and pillar box to fit video to film, film is "generally" 1.33 aspect(square pixels) with different frame aspects, camera guides
Film to video
calculate and add black bars then resize to video, 3:2 pulldown convert 24 fps to 30 fps
Full Aperture Film
Super 35=full aperture, common center, common top, academy, camera guides
Academy Aperture -
2X compression in width, distortions from transforms, scope extraction
Prevent banding - higher bit depth, floating point
Banding - add noise,
increase bit depth, add grain and blur,
display banding vs actual banding - scale rgb down then scale back up if got worse then got banding, if not then display is banding not the image
Color correction on Log images
add/subtract from rgb values not scale, log to linear conversion values for viewing and for compositing
Composite Log images - must convert to linear for compositing
Composite Linear over Log
linear is converted to log, color corrected, converted to linear, log background is converted to linear, they are comped and then fed to a viewer(log2lin with viewing values), the original lin2log conversion values are modified to make the linear fit the log image and rgb values add/subtract to match color
want to do matching in log space not linear
CGI and Log
same as above but un-premultiply the cgi and color correct the background