[FEATURE] Cross-Platform Desktop Video Editor

[Tophness]

--- Update: Moved to Tophness/Inline-AI-Video-Editor ---

As mentioned at #1010, I'm creating my own video editor with the ability to generate video inline with a desktop port of WAN2GP.
I've just got the basics working. You can do all the standard splitting/joining etc. that you can do with a normal video editor, but now, you can also select a region, right click to bring up a context menu and join it with WAN2GP.
This feature is a plugin in the video editor. It sends the first and last frames for that selected region and asks WAN2GP to generate video for that exact length (after converting FPS), and then when it finishes generating, you click which one you like, and it inserts it inline exactly where that selected region was.

There's still much to do to make it a fully-featured video editor, but this is gonna be crazy powerful with WAN Alpha, Controlnets etc. You can already add them as separate layers. But I'm planning to deeply integrate it with the video editor to make it like an AI-native studio.

Screenshots

[Tophness]

I have a lot of ideas for this, but I just had a crazy one:
You could use a classifier to work out if the part you've just inserted was a good generation for the frames you're joining as well as for the context of the overall project.
Who knows how well it would work for complex stuff, but at the very least, it should be trivial to have it work out if the generation had artifacts in it like hue shifts and garbling. And if it detects artifacts in it, you could have an option to automatically regenerate (maybe with different parameters) until it gets one without artifacts.
Still keep the old attempts until you decide to delete them obviously, but it would mean 1st frame/last framing your way to an hour or more of content seems more realistic than if you had to manually experiment and test everything 30 times just to get 5s of content that works.
If everyone (optionally) contributed like 1% of the resources they use for generating to train the classifier on their own diverse content, everyone could have one that works really well