I don’t know if this would or would not require some form of AI like Real-ESRGAN that programs like upscayl use.
One thing keeping me from even thinking of davinci resolve or after effects is how incredibly powerful the upscaling capabilities are. The ‘standard’ if one was to call it that (such as what ffmpeg provides), that just use a math algorithm to upcale is often very ineffective. Say you want to take 720p or even 1080p (Which I often have because I have a variety of older video files or I use older software to make stuff) to upscale to 2.5k or even 4k, it most likely will leave halos and other distortions that is very noticeable on a larger monitor.
So how would kdenlive go about implementing such a feature. It be a killer feature if it was integrated because it’s something I would see myself using often.
Maybe I’m missing something about what you actually want to do - but I don’t really see much sense in that at first blush. Upscaling isn’t something that you want to be doing every time you render your original source.
Iff it’s something that you want to do at all, it’s something that you want to do only once. To take your original source and try to fluff it into something that uses more pixels, but doesn’t actually contain more genuine information about what was originally imaged - which you then use as new source material for whatever you want to do with it next.
And the best or fastest or most optimal way to do that fluffing is very much going to depend on what your original source material is.
So this seems much more like a preprocessing job, that you do with whatever tool from a choice of many that is most appropriate for the effect you want to achieve - to create the newly generated source clips which you then want to edit. Possibly even a long running job that you could be doing on some other system while you do you initial editing layout with proxy clips.
The best of the current generative upscalers, which don’t just interpolate the missing neighbours but try to ‘imagine’ something pleasing to fill that missing space with based on their training set, tend to be extremely slow to run - and which is the actually best at it is an ongoing competition that we are nowhere near the end of yet - so what would we gain from picking a ‘winner’ ourselves, from the limited set with a licence that is compatible to ours, and forcing that long wait to be part of the editing pipeline instead of just being a cleanly separate job like it actually really is, that you can do with whatever tool is most appropriate on the day you want to do it for the source that you have?
@northlean AI tools like Real-ESRGAN and frameboost/motion interpolation are on Kdenlive’s midterm roadmap, and there is an open feature request for AI effects. The issue hasn’t received much attention, but you could always donate to Kdenlive while mentioning these tools, which could likely increase the devs’ interest in adding the features to Kdenlive.
P.S.
I am not affiliated with Kdenlive in any way other than being a user, but like you, I really want these AI effects (particularly motion interpolation) integrated with Kdenlive.
slomoVideo was created while Simon was active with the Kdenlive team - but it likewise remained an external tool instead of being completely integrated.
It might be nice to have something like that more tightly integrated with the time remapping - but again, it’s a fairly computationally intensive process that you probably don’t really want to re-run for every change - so it’s an open question as to whether it too should remain an “external tool” which you use to boost the framerate of selected clips just once before using them as input source to the editor.
Looking at integrating slomoVideo more tightly had been on my One Day list - but if there’s a proliferation of new tools all competing to do this job better than the others - then it seems like we should be looking at better ways to interoperate with Any Of Them, rather than putting bets on which one to integrate. At least until, or if, a very clear winner emerges.