Video Enhance AI v2.2.0

Would you mind elaborating on that process in more detail?

First of all Topaz video enhance AI is a ground breaking product. FANTASTIC!

I mostly upres HDV footage and have to do a double run to get a sharp UHD result.

I first clean with Artemis MQ
Then upres with Artemis MQ

Is it possible that you can look at a setting within Artemis that will allow the user to manually increase or decrease sharpness detail.

Artemis dehalo strong does a fair job but makes the footage look artificial.

Thanks again for a fantastic product.

John

By setting these simple defaults, it only takes a few seconds to pick the the upscaled video and the original with the sound track and sub titles I need.

Sorry for the baby steps. Ok, I’ve got my settings set the same. How exactly would I go about batch converting several videos at this point? What’s your step by step process for this? Thanks in advance, I’m trying.

Video Enhance AI v2.2.0, model Dione Interlaced Dehalo v1.
Original video mpeg2 dvb-s 720x576.
After improvement in in Topaz — black horizontal borders - shaking
Example on 2 latest videos:
youtube .com/watch?v=kd891QfB6vU
youtube .com/watch?v=Won2h1KNEk4

Gaia GPU model quality is much improved to match Gaia CPU model quality in this version, but it seems it is at the expense of speed. Gaia has been on a rocky road, previous 2.1 and 2.0 version of Gaia was fast but geewiz the GPU model looked bad when compared to Gaia CPU model. I can’t remember which Gaia was using CUDA but that was fast and looked good with either GPU and CPU. I really like V2.2 Gaia GPU quality now… so much that I don’t need to used the Gaia CPU model (under recent previous versions I did). I have a 2080 TI and it takes 1.4 secs per frame to scale from 1080 to 4K output to mp4.

You can’t batch process MKVTooLNix, like you can in VEAI, but setting up the the basic defaults, makes the processing go very quickly. For me, I pick the upscaled video, uncheck the VEAI audio. Then I load the original video file, uncheck the SD video, then check the audio track and subtitles I want. From there, I start multiplexing. For a single television episode, I can do all these steps, plus multiplexing in less then 60 seconds. Of course, this is using PCIe NVMe SSD, which is very fast. So yes, it’s only one television epsiode at a time, but goes very quickly.

Ok, that’s what I originally thought. Thanks for the reply. So that’s exactly my issue, I’d really rather not go through an entire TV series, one episode at a time, like that. I believe that is unnecessarily time consuming when an indexing/copy function could be implemented by Topaz.

Topaz has not updated the Gaia model recently, so it must be something else affecting the quality and speed difference.

Today i updated my WIndows to 21h1 and for test enabled hardware accelerated GPU-scheduling. This time the Gaia models work normally, with the newest Nvidia game ready drivers. I don’t know which of these did it, or whether it was something else.

Topaz updated the backend libraries in v2.2.0 so I guess it might affect the speed and quality.

You’re indeed correct. Artemis models are being downgraded heavily. They were my favorite models. Now, the only model that I use is Gaia CG.

1 Like

I agree about Artemis models, but for me Gaia CG v5 hardly ever works. It doesn’t clean well and it doesn’t add detail well. On the other hand Gaia CG v6 works ok but it adds a visible matrix pattern over everything. What am I doing wrong?

Gaia CG means for movies. It means you don’t destroy all the details like Artemis. Film grain is what I personally really like and I don’t want to sacrify film grain for plastic look of Artemis. Gaia HQ also did the same thing like Artemis, but less intensive, and used on movies/footages with no film grain. I always use Gaia CG v5 for movies no matter DVD or Blu-Ray. It deblocks really well without denoise (I don’t like denoise). Sometimes, I prefer having film grain, but not the static grain like VEAI does. I’m talking about dynamic color film grain. I enjoy no artifact movies than no noise/grain artifact ones.

Gaia CG v5 is near perfect I can say. With some adjustments in precision, Gaia CG will definitely become my most favorite model in VEAI.

1 Like

Thank you, I will try it for movies, however most of my sources are videos, so that’s probably why it doesn’t work for me.

When you used to like Artemis, do you remember which version it was?

When I add GAIA CG v5 model to upscale my 640x480 clips, then the whole footage is characterised by some ā€œmoving surfaceā€ (ā€ždancing pixel/artefacts), just look at it: The whole scene is moving so I think GAIA CG is not ok for my clips.
Another clip is prepared by a method proposed by JCP29 according to the following path: 480p to 2560x1920p (400% AAA9) and then AHQ11. To be precise, once JCP29 proposed 480p to 2560x1920p (400% AAA9) and then down to Full HD (1080p) using AHQ11. I have modified it a little bit, namely I downscale using DaVinci Resolve to 1080p , much faster and the apply on 1080p just AHQ11 Denoise/Deblock 100%. It gives sharper output. So I do not download using Artemis HQ11 but DR and then use 100% denoise/deblock AHQ11.
Is there any to even further improve the footage? The method using AAA9 and AHQ11 does not introduce such strange behaviour as CG. Please check Victor if you observe the same problems.
PS Can you write me your upscale time from 640x480 ProRes422HQ to Prores HQ 2560x1920 by AAA9 400% and another one from 640x480 ProRes422HQ to 2560x1920 by GAIA CG v5 400% Prores HQ.
Thank you.
https://we.tl/t-yS25XJb3Ve

Exactly, and my 480p clips are recorded using very strange codec, since they show dancing pixels, artefacts, flickering even in daylight. Bitrate is high, AVI clips but strange footage…

v9/v10 is the sweetspot.

I used v11 before. Used to work okay for me with high quality 1080p low bitrate movies.

I usually export as TIF and I tell everyone to do the same for the best quality output.
480p => 4K = 0.17 s/f with 450% upscaling.