Tip: Three step tuning method, to recover bad quality videos

Did you tried the proteus Approach?

Proteus 4 wasn’t available in TVAI 3, right? So you would need to change it to use Proteus v3 in TVAI 4.

1 Like

@Mayday Have you found a general 3-step method using Proteus. This is how I summarized it.
Step 1: Proteus
Step 2: Proteus
Step 3: Artemis

Step 1: Denoising

  • Model: Proteus (Manual mode)
  • Denoise: Increase until noise is removed, being careful to preserve shadow detail.
  • Recovery Original Detail: Set to 10-20.
  • Output Resolution: Keep at original size.

Step 2: Sharpening and Detail Recovery

  • Model: Proteus (Relative to Auto mode)
  • Sharpen: Adjust for clarity without noticeable artifacts.
  • Recover Details & Anti-Alias/Deblur: Start with both at a high value (10-30) and gradually lower until the image looks natural.
  • Recover Original Detail: Set to 0.
  • Output Resolution: Keep at original size.

Step 3: Upscaling

  • Model: Artemis (Strong Halo)
  • Output Resolution: Choose your desired resolution.
  • Recover Original Detail: Set to 10-20.

How do you change it?


1 Like

Thanks, I got it. Do you know how to preserve the Auto settings from Prot3. I tried another video and does not have same outcome. I compared both video the older i did when i had v3.0.

I am guessing it has something to do with the Add Noise and Recover Detail.

@markwaltermd You do only a single run?

Assuming you are using the same model and model version, if used in manual mode with the same parameter values, then yes, it should be the same. but in Auto mode there are differences between versions as the Auto evaluation detection curve is changing to optimize auto detection process.

1 Like

You can check the ffmpeg command in both 3 and 4. I think in 3 they have estimate=8 in 3 and in 4 they changed it to estimate=20. That will make a difference and is not something you can change outside of changing the command and running it in your terminal opened from within TVAI.

1 Like

No. I don’t like the results Proteus generates.

I’ve changed my method since I last posted to only 1 Step:



I know, looking at just a screen-cap it doesn’t look like much of an improvement, but in motion it actually is.

Was I supposed to use Proteus 3?
I tried these steps. I probably overdid all of them. It was hard for me to decide what looked right on the smaller image.
Compared to what I was getting with just Proteus 4, the results seem like they have a smear-blur in parts and less fine details. Of course that could be from me overdoing the denoise step. I probably also did too much sharpen. I left them at your starting values since I thought they looked okay.
Proteus 4:

Your steps:

I guess these get compressed too much on here to see that the ridges in the socks stay in the Proteus 4 version, but are smoothed out in the other one. You can still get the idea though by looking at the knob in the fireplace versus the fork on the far right.

1 Like

Which one is better? Are you using Auto for Proteus 4. Yes indeed it worked for me on Proteus 3. have you tried on Instead of Using Dehalo, using Artemis High Quality, Or Medium Quality Halo?

Manual or Relative to Auto?

1 Like

I used Auto in this one.

1 Like

Basically, First you go to x4, then to 720p
How did you resize to 720p?
Have you tried going to 4K resolution at first pass instead of x4?

Which one is the definite method, using Proteus or Iris?

Which version do you have?

Have you tried HQ Gaia as the last step instead?

How do you sharp the video, before scaling up. Which method?

Hi… I have changed my workflow with the new versions of models. First of all if I have to deinterlace, denoise and sharp, I use this:

QTGMC(TR0=2, TR1=2, TR2=1, Rep0=1, Rep1=0, Rep2=4, DCT=5, ThSCD1=300, ThSCD2=110, Sharpness=2.0, Sbb=3, NoiseProcess=2, EZDenoise=1.0, NoisePreset=“Slower”, StabilizeNoise=true, NoiseTR=2, NoiseDeint=“Generate”, ChromaNoise=true, denoiser=“KNLMeansCL”, DenoiseMC=True, Precise=True, ChromaMotion=True, TrueMotion=true, SMode=2, Subpel=4, PelSearch=4, Overlap=8, ediThreads=16)

If the video is very noisy, after that, I use nyx3 at original resolution with reduce noise 100% and recover detail 100%. After that I use Prob4 at 122% with recover detail 100%. After that I go with Gaia HQ to 1920x1080.

If the video is not interlaced, but with noise, I use Nyx3 and after that Prob4 and gaia HQ, with the same settings as above…