I've been experimenting with a new AI model and overall video workflow technique and I thought you all might like to see the results.
This new process is far more time consuming than the previous one. As well as the new AI model, I'm actually upscaling to 4K (2160p) then downscaling back to FullHD 1080p, on the basis that this improves overall visual fidelity (in much the same way NVIDIA Super Resolution does for games). The down side is that this doubles the AI rendering time.
On top of that, I'm also dropping the upscaled footage into Adobe After Effects and adding a simulated 32mm film grain, which tricks the eye into thinking there's more detail in play than there actually is on a pixel level (our stupid brains can be easy to fool that way). YouTube's compression actually smooths that extra shine out, so if anyone would like to download the pre-YouTube video, you can grab it here.
Overall, it's a very resource-intensive workflow, with far more manual tweaking than any of my previous efforts. To put it in context, this single one-minute video took me around 3 hours to remaster.
I'll look forward to everyone's feedback. If the consensus is that this looks like an improvement over the current WC4 HD video pack, I'll consider following this new workflow through to create a WC4 HD video pack V3.0. If you guys think that it's just kinda "meh" in comparison to what we already have, do let me know. This would be a major drain on my time!