Current video quality of the RESULT is very low

[AlonDan]

Currently the end result quality of the video is much lower.
I compared to ROOP which the default video quality result is great.

It will be great to have LOSSLESS or value to change from 0 (low) to 100 (highest) quality?

[Inferencer]

can't be done as they use the old inswapper model wich is trained on 128x128 resolution with no access to data on how to retrain the model the only options would be to switch to a different face swapping method with the ability to train or what most people are doing is just upscaling using codeformer or gfpgan

[AlonDan]

I'm not talking about the FACE result which is based on the model.

Just compare to RENDERED output video quality between ROOP and REFACER you'll notice that ReFacer quality is very low, while ROOP by default is lossless quality.

[Inferencer]

apoligies for the misunderstanding
I imagine the issue is a result of a different encoding method being used, roop is constantly improving this repo might be behind on tracking & implementing their updates or is attempting to save inference time by using a faster encoding method.
This is just a theory however I havent checked/ compared the code myself so welcome further input from the owner or community.

[AlonDan]

No worries :)
Unfortunately I'm not a programmer so I can't help with the code, but I hope it will be added to ReFacer because it's REALLY cool! 💙

Actually in the latest versions ROOP added a way to control the video quality, but by default it's already very close to 1:1 if you compare from original target video to the end result (ignoring the face of course)

For face quality I believe ROOP enhance with GFPGAN but I'm not a big fan because of none interpolation of frames so things may look flackery and weird for a VIDEO:
Teeth will start change or flicker, eyes, skin, etc..