Honest Feedback and Linux Support

Topaz Video General
1

After using Video AI for a couple of months, I need to share my thoughts. It has been a fully frustrating experience, and the developers need to take some serious feedback from the community. I’m not only speaking for myself, I’ve seen numerous comments on the forums sharing the same thoughts.

The Good

Proteus is the best video restoration and upscaling solution I’ve found. I’ve found it to be superior to any open source option, AI or traditional, and I even prefer it over Starlight Mini. I find it wins in most quality aspects over Starlight Mini, is up to 50x faster in my testing, and works on AMD cards. It blows open-source upscalers such as Real-ESRGAN out of the water. (Note that I have not tested the full Starlight model, as I want to keep my rendering fully local.)

The Bad

The UI has a significant number of bugs, including ones that impact significant features such as pause/resume and retry failures from checkpoint. The Chronos models also fail randomly on longer videos (30 min+), which combined with retry from checkpoint being broken, makes them near unusable. I’ve reported all the bugs I’ve found via the forums. Unfortunately, none of the reports I’ve filed has received a response from a Topaz team member. I have noticed there are other issues on the forums being marked as “Investigating” status, so perhaps there’s something I’m missing, but I’ve tried to include as complete reproduction steps as possible on each issue.

I think a lot of users are confused as well by the release of the new UI, as it doesn’t feel different from the old one in a meaningful way, it still has all the bugs that the old UI had, but it also has several new bugs. So from our point of view, we feel like the dev team is not taking enough care to ship a fully functioning and properly tested product.

The Ugly

Linux support. As I’m sure the devs know, there was a Linux beta in the works as of a year ago. The beta was quietly cancelled, a piece of information that I only found in the comments of the latest beta release thread, from a user who had emailed support, not even from Topaz staff. Despite the large number of users who have requested that Linux support be added, instead of addressing our concerns, the staff recently locked that thread, with no comment, to prevent further discussion. To me, that feels like a big middle finger to us as users of the software. Topaz is the last single piece of software that requires me to keep Windows on my system, as Microsoft pushes harder on forcing users to have an online account, adds more spyware integrated into their operating system, and continues to break various features with every major update. This is really the biggest issue, combined with how the staff have responded (or rather, not responded, as we’ve gotten no direct communication other than the locked thread signaling the staff wants us to shut up about the issue), that would move me away from Topaz. Every other piece of software that I use either works on Linux or has an open-source alternative which does and is of similar quality. Unfortunately, Topaz lacks that, and its Proteus model lacks an equivalent, which is strong praise for Proteus, but frustrating as a user.

10 Likes
2

I second the Linux support, this should be prioritized! I believe there is potential for performance improvement under Linux compared to Windows as well as it has more advanced up to date file system for moving big video files compared to the ancient NTFS…

3 Likes
3

From an enterprise perspective, Linux support is crucial. We have massive GPU compute farms available. All exclusively Linux. All completely unavailable to Topaz workloads. We will be moving away from Topaz for this reason, despite the superiority of their models for this work.

4 Likes
4

Are there any discussions of bringing back the Linux builds? I have moved all of my PCs over to Linux over the past few months, and quite literally the only thing I have not been able to port over are the Topaz products.

I looked and there used to be Linux builds but they seem to have stopped. I’m on Ubuntu 24.04 which is newer than the 22.04 which I guess was the last build and there were too many dependency issues to try to get it installed without breaking everything else on the box.

I know a lot of people have been making the shift over to Linux this year and it only appears to be accelerating. Looks like 2026 may be a big pivot with Microsoft wanting to inject a lot of AI directly into the OS which has people annoyed. (I’m a huge proponent of AI - but NOT in the OS itself.)

As it stands now, my subscription is set to renew next month and if there aren’t any plans to start supporting Linux I’m probably going to just let it lapse because I won’t be able to use the product without creating a dual boot into Windows just for Topaz alone. That in itself creates a real nuisance to my workflow and would be a huge time waster flipping between OSes and I’d just look elsewhere for a Topaz replacement rather than having to deal with that.

6 Likes
5

There have not been any serious talks or discussions about investing time back into a Linux build as the previous releases were lightly used by users.

This could change in the future though, if there is a large shift to the Linux OS and demand.

2 Likes
6

Thank you for the response, although that is unfortunate news. I’m afraid I’ll have to join the others saying I won’t be renewing my subscription as long as there’s no Linux support, and will stick to the fallback license on the Linux beta version 5. I’ve migrated away from Windows as much as possible, and don’t have a desire to use it just for Topaz.

I’m wondering how accurate the metrics are for Linux usage, given that Linux users may be more likely to go in and disable telemetry than Windows users. From the comments on the Linux beta threads, it definitely didn’t seem to me that it was only lightly used. There’s also the fact that the Linux beta only works on Nvidia, so any AMD GPU users were excluded from trying it (unless they wanted the slow, slow CPU rendering).

2 Likes
7

In terms of speed or quality (or both) ?
I just got a M4 Mac (12-core cpu, 16 core gpu, 24GB um) , just wondering if I should try Real-ESRGAN.

8

In terms of quality. I haven’t compared the two for speed, although Proteus is a relatively fast model (I’m getting 11 fps upscaling from 1080p to 1440p on a 9950x + 4080 Super). As far as open source models go, Foolhardy-Remacri is the best upscaler I’ve found, as it avoids the oversharpening issues I’ve found in other models like Real-ESRGAN. However, Proteus has two big advantages over those models, which is 1) it is temporally aware, it’s designed for video so it actually takes surrounding frames into account to make better decisions about how to reconstruct the video, and 2) it performs enhancement as well as upscaling, while the open models will only perform upscaling, and for things like deblocking, I’ve found Proteus to do a better job than any traditional deblock filters as well.

9

This is what I heard about Real-ESRGAN, that it wasn’t temporally aware, so there might be a ‘shimmer’ between frames.
I’m looking to get my M4 Mac Pro I just bought into my workflow somehow to take the burden off my 5090 PC.

F-Remacri also appears to be temporally unaware. How do you improve that?
I did a little digging myself but don’t know how accurate this is (below). Comments?

What does address FR temporal consistency is switching to a video super-resolution model that explicitly uses information from multiple frames:

  • BasicVSR++ is designed to aggregate info across frames and is reported to produce smoother temporal transitions.
  • RealBasicVSR is another real-world VSR line that emphasizes balancing detail vs artifacts using a cleaning+propagation approach.

(I tried Proteus vs Starlight Mini, and SLM won out in terms of quality… sadly because Proteus was quicker.)

10

All those upscales aren’t really temporally aware.

E.g. in low res faces they change from sunglasses on to off to on again just when the angle changes a bit.

Unfortunately this is still true for Starlight which otherwise is the best for low resolution inputs (for higher quality you might have better results in fact by using Proteus or Iris).

2 Likes
11

IMO not supporting Linux is betting against AI and especially local AI. The large amount of people picking up powerful GPUs to do local AI media work is observable. Most of the tools for this work best on Linux.

2 Likes
12

I wonder if I could do a 2x upscale with SLM, and then feed that into Proteus for a further 2x, since Proteus is good with cleaner footage, and SLM can get my video to a quality level that P can work well with?

Just looking for a workflow that’s less time consuming than a 4x upscale with SLM, because this is giving me 0.2fps on my 480p > 1920p video. I’ve got a few hours of 59.94fps footage to convert so its many days of rendering.

I’ll do my own testing of course, just wondering what your opinion was.

13

Some people here have reported doing exactly that.

1 Like
14

I tried a Proteus (and Iris) 2x upscale on video that SLM had already 2x’d.

I was impressed, for about 85% of the video it looks very adequate. But there are scenes where I have very poor sand texture. SLM 2x and 4x deal with this OK, but Proteus and Iris make a bad looking texture out of the sand noise. See image (if it shows below)

I am experimenting with Recover Detail and Grain settings in Proteus and Iris to see if I can mitigate this. Also on faces which are in the distance it produces hallucinations on the face, and morphs them over time.
Any advice on how I can get it to deal with textures and noise in the sand ?