I am sitting here thinking about how it is basically impossible to relight a person, but it seems like it should be at least somewhat easier than it is.
Does anyone have any tricks for removing a light on someone’s face, or just in general removing light?
(It’s of course way easier to add a light than to remove.) I’ve done a good bit of relighting using normals, etc. — but it hasn’t been removing an in-camera light.
To clarify: there is no clip, and I’m not working on anything that requires it at the moment.
I’m looking for info that will broaden my mind. It’s the sort of thing people might have some tricks for, and it isn’t something I have tricks for, so I’m asking about it.
The flat demo footage makes me worry it will cause even more jobs to be shot in the “marvel style” of ambiguous lighting in support of backgrounds-to-be-designed-later. A huge win for all of us virtual re-lighters, a huge loss for aesthetics. Haha.
Actually, there are increasingly better tools. We’ve done tests in Resolves and there are multiple tutorials.
It comes down to getting a normals map and depth map of the face with AI tools. Then use a light that honors the maps model the face properly including AOs. Unlightining maybe possible as it’s essentially a negative light.
I was just in a position where I had to regretfully explain to a client that I was unable to significantly relight 2 humans walking through bright sunshine without spending major time and energy, and they perhaps wouldn’t like the results.
Someone made a remark about a film, utilizing a certain technique, and I explained that they won an Academy Award for it.
Those people that want Oscar level results on Instagram budgets…
Oh wait, this is a great business strategy. Don’t use your own money, leverage the VC funding of big data centers in the AZ desert to do this. Bingo! I knew AI was brilliant.
Done this in the past to a certain extent by patching in footage that was not over exposed.
And a lot of paint.
Time consuming.
The Beeble thing looks like it might need some 3d modelling to get the normals…?
Saw a demo of something using ai last night to do face replacements. I wonder if it could be leveraged to cut and paste in under exposed footage? However, you still need good tracking and as we all know, light changes are the bane of trackers.