This drove me crazy a bit, I thought that running multiple instances started from the command line at the same time, each with a different model, mixed up the models somehow in the GPU, or whatever.
I have looked at it in more detail and this is what I discovered on my system. The left column below is the model used with the manual mode of VEAI that renders the expected output, the right column is the model name to give in the command line to get the same output file as in the left column:
i have no idea why anyone in the world would prefer the spartanic command line before the gui…
however, i am running multiple instances for month now, on every instance is am using different models and settings and i never noticed any problems like yours. Maybe that command line ist not just inconvinient but also buggy…
i have no idea why anyone in the world would prefer the spartanic command line before the gui…
Because you can’t automate with the GUI. Unless you just use VEAI once in a while for ad-hoc work, the CLI is crucial to not babysitting your computer.
VEIA is nice for what it does, but it has limited pre-processing options. It’s built in deinterlacers don’t work nearly as well as other stuff that’s out there, and it has no inverse telecine. Even if it did have good deinterlacing, it doesn’t make sense that the “deinterlace” step would be coupled with just one or two models; it should be something that’s independent of which model I choose.
Similarly, it’s output options are extremely limited. Almost no h.264 config; no h.265. So I just want to output to lossless and use something else to encode.
So if you’re trying to get the best results, you might be preprocessing, and you’ll definitely be using something else for encoding.
I have scripts that do pre-processing with AviSynth, then run VEAI, then finally run FFMPEG to convert the lossless PNG output to h.265 encodes. And I also have a lot of files I want to process with the same series of events. Without the CLI, I’d have to intervene at every one of these steps for every file.
however, i am running multiple instances for month now, on every instance is am using different models and settings and i never noticed any problems like yours. Maybe that command line ist not just inconvinient but also buggy…
I’ve had the exact same problem as OP. The command line is definitely buggy. But inconvenient, it is not - if it worked correctly. Not having to do stuff by hand every single time is very convenient, and the CLI is how that’s possible.
I am like you, unless I need to push hundreds of files through VEAI, as I did during the weekend, then I am choosing the command line.
I did not encounter any bugs, other than the mixing up of model names as I indicated above. In fact, I am surprised that starting 8 instances of veai.exe within the same second, and then consuming 380W of GPU for 6 hours before starting to work on the next video file with again 8 instances, works very very well, all the time (40 hours non-stop), without bugs, without crashes, and sometimes filling the GPU to 22GB of the 24GB.
You can of course also automate the GUI with things like pyautogui, as I have done with ramming 60.000 video frames through Denoise AI and Sharpen AI, which also worked fine.
I’m trying to use Gaia HQ. I was hoping it would work if I just switched it as you suggest.
For me it comes up as Gaia CG whether I put in gcg-5 or ghq-5. So I pretty much always have to stop it, reset the model, and restart it after launching from CLI
Just disregard any information in the GUI when you are starting veai from the command line. It just displays the settings you used last when starting from Windows. For me it displayed Artemis High on all instances using all different models. Yet, it definitely uses the correct model as per my list. Just check for yourself by processing a few frames from the GUI and from the command line and then compare the outputs. That’s how I found this out.
This might be due to the previous beta build that had the model mix up bug, a clean install might solve it? I had no issue with the command line on the current build (except the fact that there are no switches for grain amount and size…)
I did a clean install, and I am starting the correct non-beta veai.exe from the correct folder. However, I did not delete the 2.0.0.2b Beta. Perhaps that’s it. I will test that. Thanks!
It just displays the settings you used last when starting from Windows.
But I’ve never intentionally selected “Gaia-CG” before, I’ve only ever tried to use Gaia HQ. It always seems to come up with this now (since 2.0) no matter what i do.
It is possible that it is actually using the right model; i never really checked the output? But I need to stop it and fix the grain size anyway so doesn’t actually make much difference, I still can’t automate my workflow.
If you are a bit familiar with python you could try pyautogui to automate the mouse clicks and keyboard input into the GUI fir grain and such. Or wait for Topaz to make all features available on the command line. As I said, I am still using de Denoise AI and Sharpen AI successfully with pyautogui. It works, but is just a bit tricky when using the mouse and keyboard yourself at the same time.
I appreciate the suggestion. I’m not quite desperate enough to involve yet another tool to work around the broken CLI Right now it takes my 1050ti about a day to process each file, I can live (for now) with the downtime and just check on it around when I expect it to be done and manually start up the next file.
This seems like the kind of bug they ought to be able to fix really easily; I guess I am too new here to know whether that is likely to happen or not. Hopefully I’m not being optimistic!
Right now it takes my 1050ti about a day to process each file, I can live (for now) with the downtime and just check on it around when I expect it to be done and manually start up the next file.