Topaz Video AI 5.2.1

[alanhusband-677420]

The GPU selection is still reverting to AUTO after creating a single preview, and running on AUTO causes the Rhea model to consume all vram on my GTX 1060 and then throw an error. When selecting the GTX 1060, the preview is created correctly without consuming all vram.

Iris consumes more vram on AUTO but not enough to cause an error. Iris uses less vram with the GTX 1060 selected, but then reverts to AUTO after a single preview.

EDIT: It seems like all exports are carried out in AUTO mode. A Rhea export just consumes all vram and throws an error. Reselecting the GTX 1060 doesn’t help.

EDIT2: Just checked the FFmpeg commands. Confirmed that with the GTX 1060 selected I can get one preview with “device=0” before it reverts to “device=-2”. But all exports revert to “device=-2” when the export begins.

logsForSupport.zip (208.7 KB)

EDIT3: Was able to force a Rhea export with “device=0” through the command line, which seemed to process ok.

EDIT4: I don’t know why the GPU selection is reverting to AUTO, but I’ve worked out why models are consuming more vram, which causes Rhea to throw an error. The AUTO mode is not respecting the “Max Processes” setting. I have it set to 2 which should keep it in “low power mode” (1 process) for a single enhancement, but AUTO is ignoring that and trying to run 2 processes on a single enhancement (yes I know that sounds counter-intuitive). Running 2 processes for a single enhancement consumes more vram which is a problem for Rhea unless your GPU has alot of vram.

[Harald.De.Luca]

Have you even downloaded the template? If you didn’t download it, it’s normal that it doesn’t appear.
For my part, the model is very present

[karol.guzik (ZoDDto)]

Pushing out beta versions as a finished release.
Classic Topaz. :>

[dariusz.zawadzki84 (dariusz)]

THX

[frayoshi (Francesco Yoshi Gobbo)]

Their salaries have to be paid with the extra income, leaving them the only ones to gain.

Well, they lost me… Luckily, I was still in the 30 days refund period.
Kudos for keeping their word, at least on that!

[Harald.De.Luca]

The new Rhea model looks promising. When I see the quality of the previews.
But still has a little trouble with certain small typographical characters, such as the letter “z” here, which is transformed into an “i”. For the word “zone”.

Otherwise, in terms of overall detail, it looks pretty good. Maybe a little weird on the skin around the forehead.

Here, you can almost see the details of the character’s eyes. A good point. In addition to having the grain of the film removed but without damaging the small details.

A lot of the hair details are coming back.

Seems perhaps better than Proteus 4 or iris.

Even the grain of this energy discharge is suppressed and smoothed out.

The stars are well recovered. There’s no loss and it’s a good sharpen.

Detail is better, even in this blurred sequence.

Hum perhaps subtle details still disappear in the brown stains of the garment.

Note that the details of the ramp grid have been greatly enhanced.

I think that after export and further enhancement with just a focus, it will be quite sharp.

Interesting. On a close-up, I can almost make out the symbols on the DHD.

But the only downside is the rendering time of this model. With my RTX 3070 and a minimum power limit setting, I’m at 0.20 - 0,21 seconds per frame. That’s a lot. If I increase the power limit to 100%, I’m down to 0.18 seconds per frame, but the temperature rises to between 55 and 56° with the fans on full blast, so as not to exceed 60°.

And an enormous final rendering time.

I hope the model will be improved in terms of speed and quality, although it’s already pretty good in terms of image quality. But for the moment, graphics cards are pretty expensive, and mine is barely 2 years old.

In conclusion, the model is great for image quality, but its major drawback is its rendering time.

I forgot, thanks for correcting the interface problems that have been bothering a lot of people since the release of v3. At last, an interface as good as v2. Especially when you minimize the taskbar window and restore it.

[newborn2010]

RTX3090ti SLI. Need All GPUs option back I don’t think that multi-card users are less than 2%. If you want to conduct a rigorous investigation, please initiate a public poll on the forum.

[jptm.van_melis]

I stopped using Rhea for the time being.
I can see where it’s better than Iris, but the most important part of each image are the faces of the people and they lack expression after Rhea is done with it.

I also do not know what I need to do to tone things down.
Maybe it’s the “recover detail” which Rhea is lacking.

It will probably improve, but even if it does I will not use it if the speed stays so slow.

[PeteVideoEdit]

Even if like Iris very much it denoises often too much since implementation of Iris.

I use therefore the manual settings of Iris and denoise is at “0”. This doesn’t help really much, cause
the denoising is always for some footage too much.
I hoped with Rhea, that I can adjust by myself how much denoise I would have on minimum with manual settings. Also here - as in Iris itselfes - is the minimum denoising always too much.

[danfoerster]

Topaz is not longer working with the new AMD Graphic-card update from June, for the AMD Radeon 6600M series. It sends a resolution error - “Try changing the output resolution, bitrate or encoder.” If you do, you get a new graphic error.
edit: it is not a Topaz Problem, each version of topaz isnt working anymore with the AMD Update.

[gene-8240 (GeneK)]

The 24.Q2 update to the PRO edition drivers for Radeon workstation cards took down my entire system when I installed it. I had to use system restore to get back to 24.Q1.

[danfoerster]

If you have davinci resolve it is not working anymore by going back to the old version. That was the problem I habe right now.

[gene-8240 (GeneK)]

I have Resolve, but so far haven’t worked up the courage to try the plugin. I still see too many problems being reported.

[PVan23]

I can only say that they better make the standard license a LOT more stable and performant BEFORE coming out with a PRO ‘multi GPU’ version…

The standard user license owner will be left ‘solo GPU’ and the PRO version will be FAR to expensive to BUY (min. 1000$) AND what semipro user has the money to add 2 NVIDIA 40xx cards (if the power supply can handle it) into their systems?
Not even speaking about the ENERGY that this will cost to run 2 40xx cards in a tower setup…

I for one do not understand why Topaz is choosing this way of business to come out with a PRO version…

Their standard model still has performance AND lots of bugs and we are already at 5,2,1…

Not even speaking about the MAC user that is now confronted (even with ULTRA Mac Studio) with a LIMIT in the PRO version what he/she can do…the Mac studio Ultra cannot handle multi GPU support (cannot add cards…) …

With the RHEA model (that I like!) i takes my Macbook Pro M1 18 Mins. (!!!) to have a 30 secs render preview). …and this for a 1280 x 720 HD movie convert project. …18 Mins. for 30 sec render !!!

I simply do not get the logic at Topaz…

[mattpebworth]

Been here long enough to remember when they went back on the lifetime license…

[matteo.villa]

Bring back multiGPU for VideoAI.

It’s clearly coded in the software.
Was there before you greyed out and blocked on purpose within the latest update.
It’s an existing feature removed purely to “create” a new “Pro version”.

There are amateur and professional with small business who can’t afford 1.000+ € yearly or more just to have an already existing feature removed.

if you want to increase the value of the Pro version, make it possible using multiple GPU on multiple projects, or improved batch rendering workflow, while keeping multiple gpu usable for single project within Video AI

[tjalve.aarflot]

Please handle Pro as a license, not a software limit.

Pro users that needs full commercial license, team seats, admin and permission settings, render farms etc. need to purchase a license for that.

Enthusiasts shouldn’t be punished by removing the multi-gpu option. We’re not in your market for Pro in any case.

[moheiddinedabbech67]

GTX 1060 Unfortunately Rhea is Very Slow

[moheiddinedabbech67]

How do you do it?

[moheiddinedabbech67]

I could just run Proteus and Iris with the two-pass option