Frame.io (donwload enabled). Yes, the whole take.
And thank you!!
is this camera original?
I guess you tried it but I would paint patches and track them in piece by piece.
I would freeze one frame with the original image, track it and warp it with exbicubics. Do a diff so it matched with the bg. Then replace the freeze frame with a clean paint of the shirt.
Disregard my comments if this is a practice shot to figure out a workflow with MV.
Thanks @Sinan. This isn’t a practice but a live job (trademark removal for a documentary currently in post). I’ve already tried most of the things you mention. The amount of change in shape, lighting makes it very difficult to track cleanplate patches in, considering that this 378 frames, and only one of several shots, though one of the most complex.
Certainly possible to do it painstakingly by hand, but neither calendar nor client budget supports that effort. So it has to be something that scales better. It doesn’t have to be a perfect removal of the logo with red fabric identical in movement, so some leeway exists. But we also don’t want to just do the sledgehammer blur easy out.
I’ve tried motion vectors, MochaMesh warp, Nuke InPaint, ext bicubics, Nuke CopyCat. Alan’s MLTW is the closest yet, but not quite there. I know of the link to Si InPaint, haven’t tried that yet, but suspect it largely does the same as Nuke InPaint. The tutorial for Si is great, but it’s a vastly simpler shot they show.
No, these are exports from the edited timeline (UHD ProRes so decent quality). That’s all I have to work with for the job, though with some effort I could probably get a hold of the camera original. Not certain what the original camera was. Documentary film, so it’s a a mix I’m sure.
I did this in under 2 hours with MotionVector Tracking.
It would have taken to much time to paint clean frames, so I just used the original plates as keyframe input and tinted it to indicate patch area.
There are some frames that have “crumply areas” but those would be less noticeable with a clean smooth shirt.
I used the hardest section of footage, the other parts are just repetative action.
I didn’t do any holdout reveal on the arm.
2 Likes
Thanks for that. Looks good. I’ll go back to Motion Vectors for this one instead of MLTW.
Could you increase the Download Limit?
-Ted
I’ve added another 15 downloads. This really was just for Jan, but if a few other people want to see it, OK.
3 Likes