Machine learning for relighting?

I’m about to “light” some candles in a scene and there are various objects and people in the scene in relatively close proximity to the candles I was wodnering if theres any way (or tutorial) to use machine learning to relight the objects/subjects? Or if anyone has any general tips or tricks for flickering candle light relight.

hey Dino!

There are no ML tools in flame (as of yet) for a task like this. Altho there is quite a lot of ML happening with VFX out there, it’s all custom build and quite time consuming to create models in order to define the task at hand. Unless you have a lot of time and a really robust GFX card (or two!)…

that being said i think you could use some of the face and body ML extract tools to help build custom CC maps to simulate soft illumination color correct on people and the rest is roto to isolate objects; surfaces to all get illuminated. You can combine your roto shapes with shadow casting in Action. I also have had a lot of success using sapphire rays set to black or negative to simulate the negative space from shadows in illumination.

let me know if that makes sense! cheers


You can check out this series on ML face normal and depth extraction, which, in theory, is what you’re asking about. Not sure if the results will give the quality you want. I would say that this is more useful for a subtle brightening or shaping, not a photoreal effect like you’re describing.

1 Like

I’m having major late-90s Showtime promo flashbacks where we used S_GlowDarks :heart::fire::sunglasses:


SOME TRICKS NEVER DIE :rofl: @andymilkis

1 Like