Hacker News Reader

Comment by SamBam | Hacker News Reader

Is there a model on Fal.ai that would make it easy to sharpen blurry video footage? I have found some websites, but apparently they are mostly scammy.

View on HN

Unfortunately, this is a fairly difficult task. In my experience, even SOTA models like Nano Banana usually make little to no meaningful improvement to the image when given this kind of request.

You might be better off using a dedicated upscaler instead, since many of them naturally produce sharper images when adding details back in - especially some of the GAN-based ones.

If you’re looking for a more hands-off approach, it looks like Fal.ai provides access to the Topaz upscalers:

https://fal.ai/models/fal-ai/topaz/upscale/image

Seconding the Topaz recommendation. Although be aware that is the Image upscaler model, and the parent commenter asked about video.

Here's the Fal-hosted video endpoint: https://fal.ai/models/fal-ai/topaz/upscale/video

They also offer (multiple; confusing product lineup!) interactive apps for upscaling video on their own website - Topaz Video and Astra. And maybe more, who knows.

I have access to the interactive apps, and there are a lot of knobs that aren't exposed in the Fal API.

edit: lol I found a third offering on the Topaz site for this, "Video upscale" within the Express app. I have no idea which is the best, despite apparently having a subscription to all of them.

You want a deconvolution pipeline like https://bartwronski.com/2022/05/26/removing-blur-from-images...

Or more likely https://www.cse.cuhk.edu.hk/~leojia/projects/motion_deblurri... for video

I'm dimestore cheap, I'd be exploding to frames and sharpening and reassembling with a ffmpeg>irfanview process Lol. It would be awfully expensive to do it with an AI model and the results would be expensive. Would a photo/video editing suite do it? Google photos with a pro script, or Adobe premiere elements, or would you be able to do it in yourself in DaVinci resolve? Or are you talking hundreds of hours of video?

FYI that is an extremely challenging thing to do right. Especially if you care about accuracy and evidentiary detail. Not sure this is something that the current crop of AI tools are really tuned to do properly.

This is a good point. Some of the tools have a "creative mode" or "creativity" knob that hopefully drives this point home. But the simpler ones don't, and even with that setting dialed back it still has the same fundamental limitations/risks.

Reply on news.ycombinator.com