Hopefully Upcoming Ultimate Video Upscaler Almost Ready - Better than TOPAZ AI - Industry Level - Capable of Automatically Upscaling Entire Movies - Please Check All Images and Leave Feedback of Features

After the recently published Wan22 Lora tutorial, for which I am very very grateful, the next item on my wish list would be the ultimate video upscaler. Any progress there? Thanks in advance for all your efforts!

Is there a time estimate for when it is expected to be published? The result looks very good in the examples.

This is probably a rediculous request, but it's for an old video I edited that I have been trying to restore. It was transferred a few times over the years and is very dicey now, with wobbly video and edits that can pull the video ever so slightly out of sync for a moment. Can the program help with this by detecting a shot (starting with the first frame in a shot, going through the end frame of that shot) and only do the one shot? I know I would then have to move the playhead and do hundreds of individual shots to clean up the video, but at least by automatically detecting a shot it wouldn't be necessary to also move the in and out put again and again, too. Probaby too niche a feature for most people to need, through.

Can you add SeedVR2: One-Step Video Restoration via Diffusion Adversarial Post-Training to your video enhancer code is released better than STAR? https://iceclear.github.io/projects/seedvr2/ https://www.youtube.com/watch?v=stdVncVDQyA demo please

Hi Furkan. app looks very promising!. Do you plan to include face restoration feature ? (codeformer / gfp gan)

Looking forther for the app to test it

I would like to see examples in a more dynamic video

Can you add a very basic editing tool to cut sections of video and export them either directly to the upscale section or as a saved file?

can Star upscale 720p to 720p that's what i do with Topaz Starlight Mini

Based on example I see that banding of colors on eyes and cheeks is stronger on upscaled. I think frames should be "restored" or smoother before upscaling. The best would be to see example with generated realistic video with people etc. Because there is a lack of good upscaling video with realistic scenes. Some day I tried your SUPIR upscaler for frames of videos, all was ok but small details like hair was generated with randomly seed so they didn't move but blinked.

I have RTX 3090 24GB what is the maximum upscale resolution?

topaz sucks

The thing is, in the end of the day, you want a diffusion based upscaler for the final output. I've been using Starlight Mini on my PC, it runs locally, it's slow but result is superior. It smoothens out the frames, makes them more cohesive and corrects inconsistencies. When you apply the same technique for each frame, it will be inconsistent, because longer AI videos are made out of many separate clips, which require different handling. That's why diffusion model shines on AI videos. If you have non AI videos, then this upscaler is faster and more reasonable to use. But on AI videos, diffusion is the way to go. I stopped using other Topaz models and only use their Starlight mini. Result is like night and day.

that's a lot of upscalers, can't wait to test :-) I don't know if it is something that can be done, but using teachache creates bad or blurry hands for people, could an upscaler do something about that ? (like there are upscalers that try to make faces look better or for photo we have ai enhancers that create new details in photo)