keep watching the skies
Sooo .. View conference is over , i’ve had the pleasure of attending Sebastian König ‘s workshop on camera tracking , and the organization (Kino) made available some very interesting footage to experiment with tracking.
I just couldn’t resist doing experiments with some nice hd footage of Torino : the idea is simple : spaceships floating over the city.
I started today , modelling a spaceship ,and to make it quickly , i used the ” SHIPYARD v07” library (90% of mesh is made straight from those blocks) .
Great resource , some quick kit-bashing (1 hour) and there i have a ship to play with.
Then i spent some more time testing render settings : i’m loving BI with approx ao , env light and emit gi. (also did some Cycles tests .. overlayed on the BI render in this pic)
I’ve started painting textures for main pieces , and used tileable maps and vertex paint on the rest.
I’ll try the actual tracking next , but the thing i’m most curious about is the render – integrating the lighting with real footage .
I always lean to illustration and non-photographic renders , so i wonder how this will turn out..
updates below :
UPDATE 6 & 7 :
More fine tuning … added the ships’ shadow on right building , and now i fixed a ugly seam where it meets the existing shadows.
Added some breakdown shots to the video , and kept tweaking the shading and comp nodes thanks to the suggestions at Kino3d.com.
This version is a bit ..too bluish .. but anyway, i’m trying to balance showing details with trying to hide the fact is CG :using blurry effects and color grading.
Ok, making it look good and detailed could also be done by refining model and shading quality (more bevels and custom textures) but i’m focusing more on the render and what i can get out of the quick model i kit-bashed out of Shipyard with minimal custom textures…
UPDATE 5 :
Much better reconstruction ! and more ships !
Watching again Sebastian Koenig’s workshop , i found the infos i was missing for good reconstruction (setting origins and axis , undistortion in previews)
So now reconstruction is pretty decent (i used a sat. image for the plan and markers for height of the building and it worked fine , it’s no longer a distorted mesh as before , so i could also map reflections on windows )
Many thanks, for all the suggestions and infos on the thread on Kino3D ! : http://www.kino3d.com/forum/viewtopic.php?f=3&t=8927
Better lighting and reduced error in tracking markers . still having problems learning the reconstruction part.
UPDATE 3 :
A quick render of the sequence , as it is now … some ‘wobbliness’ towards the end : some is actual camera movement but there’s also some bad tracking … i’ve also used some 2d stabilization on it but it’s not perfect yet.
Also the reconstruction of markers to a 3d / mask object is very approximative , i guess i didn’t remove enough bad markers. Maybe i’ll try with different footage to get a cleaner track.
The light and color matching is not as hard as i though (but still difficult !) , i still have to try using an hdr pano and reflections to see how better it will integrate.
UPDATE 2 : i switched to another footage piece from the workshop , that was easier to track , and aptly named “spaceship lookup”
In terms of tracking i’m just getting familiar with the very basics , but the render is already more interesting : it’s BI using approximate gather only.
The approx. AO works nicely , the indirect/emitters are quite useful and very fast (total is 15 sec render +20 sec comp on a 3ghz core duo2 … great!)
No proper reflections (no raytrace and no hdr pano from the site) , but -as i hoped- the handpainted specular map improved a lot the spec. realism…
This gives me interesting stuff to think about … in terms of realism / integration with footage but keeping in mind practical issues (from renderspeed to using kit-bashing to quickly detail the model , to unwrapping only some pieces and using many low-level tricks .. and see how decent …or good the result is )
About the re-constuction and masking … i did it very quickly and cheaply , there definitely must be better, more ‘proper’ workflows for that..
UPDATE 1 : very early tracking and render test :