For several concert videos I’m remastering/upscaling I need an AI model like Iris MQ with face recovery but considerably stronger.
These are pro-shot indoor gigs from like 2015, often dark and the wide angle zoomed-out shots don’t have enough detail to properly enhance faces. The source bitrates are good but the videocameras used may not have enough low light sensitivity.
Rhea does better but loses faces when the lights dim.
Thus I’m looking at other upscaling apps and projects. Gigapixel’s Recover in v7.3.1 recovers the details except the eyes are wrong. The singer should have dark brown eyes but Gigapixel covered the eyes with skin and put white points for the pupils… looks really spooky.
I found the Stable Diffusion project and the main 3 GUIs: A1111, ComfyUI, Fooocus. I tried reading the various install and usage docs but my brain turned to mush. Then I read they all produce flickering video because the models only work on one frame at a time, not several frames like TVAI.
I did read someone is able to use Resolve Studio to get usable video. But better would be a diffusion model that works over multiple frames and keeps the enhancements consistent. Then I found the Upscale-A-Video AI model which claims to do exactly that (flicker-free upscaling).
When I tried to install it I got stuck at the Conda part. This is some installer app that I don’t have. I googled it and there are 3 binaries (Miniconda, Anaconda, Miniforge what?) linked here:
https://docs.conda.io/projects/conda/en/latest/user-guide/install/windows.html
Which one do I download and install?
I have Windows 10 on the TVAI box. I have Sandboxie Plus available if that would help prevent breaking the box. I have the boot drive imaged with Acronis TI.
My 11GB GTX 1080Ti GPUs only support FP32. I don’t know if that may be an issue.
Any tips would be appreciated.
info on the model:
