Thanks a lot!
Hello using V1.9.0 to upscale rare TV shows.
I believe this copy is ripped from Betamax. Format is in MPG, pal, 25fps and 720x576 if that helps. First episode I went with DTD V1 at 100% then GCG V5 to 1080p and turned out amazing. 2nd episode I am getting âshakyâ camera/jitter throughout. Reinstall windows, re-tried few times and same result. Can anyone help
Thank you very much for the detailed workflow input! Just to be clear, you use original file the first time, and the 2nd time for AA you use the original file or the 1st Topaz pass one, and then use the 2nd AA export to Gaia?
I love the sharpness of the Dione models, but as someone mentioned here, I get colored combing lines in high motion areas with them Any advice how to tackle that? I know QTGMC removes then when deinterlacing but the image is a bit soft. Does QTgmc has an option to NOT deinterlace when processing an image?
Thanks!
With that being said I am getting some insane gains with this program!
https://drive.google.com/drive/folders/17QIt6GuCGwIDN0gzgWexU8CYH4XkkaFH?usp=sharing
Here is a 30 sec clip. Before and after DTD v1
Looks like wrong field order.
Would someone or perhaps a developer mind elaborating what this option exactly does?
I had no idea this was even an option until now and was probably the reason why I was losing so much SSD free space.
It specifies that it doesnât affect output quality which is good. Just curious as to what I would benefit from having it enabled.
Does anybody have a benchmark on 3090?
It will be good if we all will take one video and start using it as a benchmark
By default, VEAI only have 288x288 for each model.
When VEAI process a video file, it will check to see whether other size model can benefit the process. If yes, VEAI will download the needed size model automatically. You can check the log file to see what happening.
You may also download all the models by yourself and put it in the VEAIâs models folder. The folder location may vary depend on your VEAI version.
If you have all models downloaded, the folder size should over 3xGB.
There is no option on DTD for field order. With DTV same stuff occurs.
I am actually reworking my workflow, since some issues have yet to be resolved. I will report back once I have solved them. At the moment I am experimenting with changing the order in which I apply certain worksteps in order to get better quality.
It is with GaiaHQ, Artemis LQ and so on. Zoom in to 100% and spot dark patterns every few pixels. This increases with % of rescale.
Thank you. I appreciate the explanation. Iâm very picky about my SSD space so I will keep it disabled.
Anyone managed to get variable frame rate DVDs to work with Dione? Trying to upscale Buffy the Vampire Slayer and variable frame rate episodes are a jerky mess.
You donât want to do that. Those downloads are literally your key to the quality of the output you get from this program. I donât know where you see that it âdoesnât affect qualityâ but that is a ridiculously wrong statement.
And yes, the downloaded models absolutely affect quality. If they didnât, they literally would have no reason to exist. If the program downloaded it, it was needed. This should never be disabled unless you are in a situation where you cannot download for quota reasons.
So do we think the tensor cores on the nvidia 20 or 30 series are being utilized yet? I donât know how useful they could be. They do a different type of AI upscaling in real time.
Iâm a little bit disappointed in the performance difference between my new card, an RTX 3070, and my old card, a 1060 3GB. Itâs maybe 2 to 2.5 times faster in VEAI? I kind of figured it would be a little better than that. Iâve verified downloadable models are coming in, but Iâve never seen the card use more than about half of its available VRAM, regardless of what model Iâm using and what resolutions Iâm dealing with. I donât know if thatâs significant.
720p to 1080p in Gaia HQ takes about .95 seconds per frame. I know there are faster models, but itâs still often the best choice when using a high quality source, as it can add realistic detail while retaining all original detail in ways that the other models cannot.
Yes, they have said it only affects speed, not quality.
If you want to believe that, feel free. It wonât be the first bit of text in the UI that beta testers have pointed out as being incorrect, and it wonât be the last.