[Feature Request] any hyper-resolution inference

[FNsi]

I tried to inference a tiny realsrgan model (2MB!) realsrgan-general-x4v3using ggml month ago, but I couldn't make it 😅

I have to be honestly say that's the thing I used in AI mostly with fast enough speed in PyTorch...but I still want it be faster...because ton of old movies, shows, etc...

So... forgive me do that greedy request 😂 hope that could beat python version a lot!

[fire]

I am also interested in loading ESRGAN for scaling game textures and logos. https://github.com/V-Sekai-fire/logoscale

[FNsi]

I found a extremely tiny size model (0.3m) and now it's be real time upscaling for 540p video... in my APU!