Baking the semantic keys in Timeline FX

I’m using the semantic keyer on a grade (fixing ears). One known challenge with all the ML tools that they’re 100% deterministic from one time to the next.

In this case Flame seems to have trouble on the very first frame. Half the time the key is good, the other half not. And very noticeable on the first frame. It’s actually less of a problem, if it can’t do the first frame, and I can fill that in manually. I actually did, except then it started working again on the first frame and doubled up…

So the question is - is there’s an easy way to bake those ML masks to avoid them changing without notice? I know in batch I could just render it and use that. But I’m trying to stay in TimelineFX for this project, so all the semantic keys are in selectives of the Masterkeyer in the Image node.



I found that making small adjustments to the region of interest over time makes a massive difference. Tweaking the position when it starts to look dodgy can smooth things out.
You could use the SelectiveMatteExtraction Matchbox and then render that out.

1 Like

Good tip on the RoI. @hBomb42 also made the suggestion to use bfx for that clip, to have a bit more flexibility since Timeline FX don’t have render/caching.

Ah yeah. I didn’t read that properly. BFX is the way in that instance.