there is workaround until they fix this, it’s adding a network drive instead of workig with network share. working here with my Nas and network drive.
1 Like
notice any slowdown doing it over network?
Same file, same setting. Artemis High. 720p->1080p. 59.94 fps. Mp4 H264.
speed in seconds per frame. Gpu : 1050Ti
10 seconds preview
local Drive :
3.0.1 : 1.30 spf
Network drive
3.0.1 : 1.31 spf
i uninstalled 2.6.4 on this computer so can’t compare.
Also, convenient.
As a telecommunications programmer, (a very long time ago,) having obscure bugs showing themselves was very beneficial. - Being the only programmer on an individual telecom project, It was up to me to open a data session (via modem) and using that particular mainframe’s protocol, I had to emulate the type of terminal(s) or punch card readers being used by that client bank and respond to the mainframe as if I was doing manual data entry from the bank. - There were also batch mode runs from those terminals, the slightest mistake in either data entry or the ‘secure’ communications protocol being used would cause an error and the whole thing would be rejected by the servicer mainframe’s front end.
We had a device know as a Datascope, which could actively record both side of these communications sessions. This made it possible to “play back” the conversation between the two computers and see, bit by bit where the problem occurred.
The actual conversation between the two computers was so fast, that our little emulation machine had only one or two instruction times between bits or bytes int the blocks being sent and received, so we had written an OS that could multitask between getting the data blocked and ready and also manage the peer-to-peer communications session going on in real time.
Today, a lot of that is handled at the chip level and/or the dinosaur mainframes we were talking to finally went obsolete. But, there are still many parallels in today’s video processing software. The programmers are just less aware of it.
2 Likes
When I looked up the difference a few weeks ago, I got the impression that the studio drivers are just more stable and released less often—but ultimately the same as the game ready.
Hi ! I’m testing my videos with the gaia hight model, I currently have a gtx 1660. And I find that it’s really slow anyway. Video (720 x576) occurs at 0.4 fps. (without upscaling) With version 3.0.0.11b and version 3.0.1 it is the same speed. Anyone have any advice to give? 
What a difference a couple months makes.
The AI imaging landscape is entirely different than it was just two months ago.
With the open-source release of Stable Diffusion, there are tons of programmers now working on over 1000 different forks. Many with fantastic GUIs…but wait…Isn’t this AI image generation, not upscaling? The big thing here is that new features are appearing basically hourly…many of these features include new and innovative upscaling algos. It is only a matter of time that things like ‘cupscale’ (which itself is dead I believe) proliferate…and get really good. There are just too many excited programmers working now!
Topaz had no choice. To keep their foothold they had to release this now.
Too bad they had to rush it out and release incomplete software with bugs. But, I get it. Hopefully they can build the plane while they fly it.
I’ll be sticking with 2.x for a while… I can’t use 3.x on paid work yet. It is just too wonky. Hopefully with the influx of real customer issues they will get 3.x up to snuff quickly. I like to see them remain a leader in this space.
What about TOPAZ training AI models. Did they discontinue to develop/Improve models while they focus on GUI and ffmpeg?
6 Likes
Obviously not. Apollo was released at the same time. Maybe the AI model team was mainly working on that. From what I know about AI, data collection to train models, training models, and programming the software that uses the trained models can all be done independent of each other.
Did you vote for it?
no need to vote 
you said the answer is what you wrote “continuous model development is the core of What VEAI does” and it’s the same with every based Ai software.
unless we’re able to train our own materials and have a crazy amount of Gpu to do this (we can rent training gpu computer on specific cloud machine offer), as soon as Veai will be stable, we’ll see 3 to 4 news model per year like before. pretty sure of that 
Then what is the point of the feature request list?
You see this in your crystal ball? Can I borrow it?
2 Likes
You see this in your crystal ball? Can I borrow it?
yes, i launch it in few seconds be ready to catch it ! it’s fragile ;).
2 Likes
I bought VEAI in order to improve video quality by AI. There are enough products on the market to manipulate (trim, crop, stabilize, upscale, convert) videos. For my workflow I need professionally trained AI models to clean and modernize old, low resolution videos. I would prefer to use a good TOPAZ VEAI Plugin to an existing video processing product rather than a standalone system under development.
4 Likes
that’s what will be available certainly in the future. the choice of ffmpeg has been done because they plan to use Video Ai as a plugin in major video editors. it has been said during the "back to the future 3.0 thread, if i remember.
they are a lot of multiple possibility, unfortunatly, the software must be stable at least and usuable with any major bugs fixed which “forbidd” the use of the software before further developement.
the resume/processing must work too.
there is no good video software doing acceptable upscallng on the market today except this software.
it will maybe change in the future, but for now, there is none. the only ones available are opensource Ai stuff which are only doing pictures upscalling, and even with that, they are not as good as what offer Topaz with their pictures Ai suite.
the other stuff, trim, crop, etc were requests from people here (including myself). “we want mouse to crop !!” etc …
it’s hard to know when a company ask to the users what they want in a future software, that they listen them, and after getting messages that “filling these requests was not that important”. it’s not easy.
anyway no matter what people want , the software must work and be stable, unfortunatly there is no other way to do it, except being patient.
if they do an exclusive cli command software, they’ll get a crazy amount of people who will complaint because “i use only your software in standalone, i don’t want to mess with premiere or davinci resolve etc…”
If they’d have stuck initially to developing the core product in version 3 rather than trying to add what many will consider to be unecessary frills (initially), I think they’d have stood a better chance of being able to release an essentially bug-free version rather then the bug-ridden (albeit mostly useable) one we still have.
Video stabilization and cropping for example, any decent video editor has them now and top end products right down to the level of Vegas Pro can use pro stabilization plugins like Mocha Pro. That kind of thing should have been left out of the initial releases and added in halfway through the first release year, as a free upgrade.
4 Likes
I totally agree with that !
It’s interesting but irrelevant for me what people request TOPAZ to do. I appreciate very much what they achieved in photo AI imaging. That was the reason why I spent money for VEAI also.
Might be that my expectations are too ambitious for AI enhancement for video edition, but I see no real progress in those AI models I need for my purposes (Artemis, Proteus).
When I upscale a 90 minutes video downloaded from Mediathek with a traditional video editor (like Magix, or Pinnacle) it takes about 2 hours processing, while VEAI takes 1.5 days, and the output result is different but not remarkable better.
TOPAZ has proven in photo imaging that they are competent in AI. That’s what I want to see also in video AI enhancement: ongoing effort to improve video quality by well trained AI models (especially face refinement).
4 Likes
you are right topaz is not better
You are right about that:
I have purchased all products from Topaz Labs, and for upscaling photos and especially noise reduction at high ISOs in RAWs, I find the software truly outstanding, better certainly than almost all the competition.
On the other hand, on the video upscaling side, I have noticed that there is a lot of potential, but to date the results are not commensurate with the time it takes to encode video: on my 4K OLED TV on which I am using my mini desktop PC, I have installed as a video player the fantastic MPC-HC version 1.9. 23, together with the MadVR video renderer, with internal upscaling set to “NGU Standard”, the results are incredible, considering then that this process is done in real time, without any encoding (and comparing on the fly the result with the MPC-HC + madVR player of the original SD video, with the upscaled video from TVAI, both full screen in 4K, the final result is not so different, but on one side I have the real time rendering of the player, on the other side I have hours encoding to get a not too much better result).
Let’s say Topaz definitely needs to work on improving upscaling quality and especially encoding speed, which does not take full advantage of the power of modern GPUs!
2 Likes
