Yesterday I learned on Reddit that upscaling in Video AI will always double the dimensions, and then—if needed—downscale. That’s a very important understanding, for I think almost every user, and I wasn’t aware of that.
E.g. 720p >> 1080p = (720p > 1440p upscale) + (1440p > 1080p downscale). I’m giving this example since I think 720p>>1080p is a very common upscaling task for the community
I suggest a tooltip/infotip/hint that will be conditional – every time a user chooses to upscale to a res that’s not a multiplier of the original res, the software will let them know that the process will involve a two-step process. It will also suggest to go 1440p, unless they need 1080p specifically. I would include a “□ Do not show this warning in the future” tick box.
By looking at the AI model folder, it is pretty clear that there are only 1x, 2x, and 4x AI models. This is the same for TPAI, TVAI, and many other AI upscale programs.
For example, when upscaling from 720p to 4k, it will always use the 4x AI model and then resize back to 2160p. The resizing is done simultaneously and it won’t affect the processing speed.
I don’t think it is necessary to inform users about this, otherwise the forum will be flooded with complaints. Many users will request a dedicated upscale model for 720p to 1080p or 720p to 4k.
I’ve started using Photo AI and for some reason it always suggests odd numbers, with a few digits after the decimal point. I’m used to Gigapixel AI which suggests 1x 2x 4x etc. Why is it different in Photo AI? E.g.
