Category Archives: Python

Nuke and Shotgun Integration

molecule-style

In my years at the Molecule in NYC, first as a freelancer and then as a perma-lancer, I had lots of time to observe how a mid-sized studio manages the sometimes huge number of shots that need to be kept track of on a daily basis. When I started, the studio was using Ftrack to manage its shot pipeline, and while poking around online I saw that ftrack had a python API, and a lightbulb went off – Nuke + Python + Ftrack = ? Awesomeness, anyway.

So I went to work on building a nuke-based shot tracking system, first for artists but then expanding to include supervisors as well. I called it “the dashboard”, and though it started small it quickly became essential to the Molecule’s pipeline. When the studio switched tracking packages to Shotgun, a lot more functionality was exposed and the whole thing just got 50% better. You can catch a quick glimpse of it in this promotional video from Autodesk – at around 0:55 in this video, the awesome Rick Shick talks about how he uses it instead of the web-based interface almost exclusively in his role as comp supervisor:


Gone were the days of manually creating contact sheets for each project; now every artist and supervisor had access to dynamic contact sheets, through which they could see and change statuses, read and post notes and images, and quickly open any shot.

Nuke multiGrad

an advanced 4-point gradient gizmo for rotoscoping

***updated 3/9/2015

I’ll be the first to admit that I’m no expert on rotoscoping, but in my experience one of the first tools I look for is a multi-gradient – something that I can use to paint out large sections of the image, and then add detail to. So I was surprised that, besides this shake-like 4-point gradient there wasn’t anything in Nuke that was what I wanted. The shake gradient is nice, but I was really looking for one that would allow me to move the points around in space, and would interpolate colors around the points as well as in between. With some patience and a lot of math (point-slope anyone?) I created a gizmo that will do just that. Below you can see the results on a racecar:

Of course, it’s not perfect, and there would need to be fine tuning around the edges, but all the rotoing in that image was done just with this tool. The user pipes in the source image to “Source” and a mask to “matte”, and the gizmo will comp it onto the source automatically, preserving the source alpha (the user can turn this feature off under the “Matte” tab. It also includes python scripting that will automatically grab the color of the source at the points, which makes painting out areas really quick and easy.

Control Panel

Because of the math involved, points 1 & 2 must be at the top, and 3 & 4 at the bottom, or it will start to act screwy. If it needs to rotate, you’re better off translating the whole result.

Download:

4pointgradient.txt
tested on Nuke 9.0v4

Terminating a child process from python

fixing a problem with calling 3rd party applications as subprocesses

I’ve been working on improving my renderfarm, and I’ve run into some trouble trying to close clients remotely. You can easily end a python process with sys.exit(), however if a rendering application (say Nuke) has been spawned by the script, it will not close. After some digging (and a bunch of great help from stackOverflow) I seem to have come up with a solution. When you call a 3rd party application with either call() or:

process = Popen(allRenderArg, env=os.environ)

Python will create a tiny, dummy process which only exists to call that application. When you exit your script, python will clean up those dummy processes, but won’t kill child processes, thereby leaving the renders going.

The solution that I’ve come up with is to get the pid of those dummy python processes, then use some unix commands to find the pids of their children, and kill them with os.kill(). If this was a linux platform I could do it in a more efficient way (possibly using pstree) but on OSX I have to use grep and some fancy code:

processId = process.pid
print "attempting to terminate "+str(processId)
command = " ps -o pid,ppid -ax | grep "+str(processId)+" | cut -f 1 -d \" \" | tail -1"
ps_command = Popen(command, shell=True, stdout=PIPE)
ps_output = ps_command.stdout.read()
retcode = ps_command.wait()
assert retcode == 0, "ps command returned %d" % retcode
print "child process pid: "+ str(ps_output)
os.kill(int(ps_output), signal.SIGTERM)
os.kill(int(processId), signal.SIGTERM)

There might be a nicer way, but I don’t know it.

Technodolly Focus, Z-Depth, and Lens Distortion

being one of the only google search results on the technodolly
techno_dolly

Any comp lives or dies on subtle qualities of a scene – color, depth of field, lens distortion, etc etc. Solving lens distortion with nuke is pretty easy once you understand the process (and immensely easy if you prepare ahead of time!); without a good lens distortion solve you’ll never get a convincing composite. Depth of field is easier to do a guess-and-check method, but of course we’d much rather get an accurate result. We’ll discuss both, but first a quick overview of the technodolly itself.

Continue Reading →