It’s simple, if you don’t want to double the framerate use Dione-TD. Dione-TD can even process progressive inputs and in some cases with a very good source it produces better results than Artemis V10 or even GAIA-HQ.
But Dione isn’t ready, there’s ghosting, anomolies and quirks that aren’t in the QTGMC deinterlacer.
Dione-TD is the one that has the most artifacts and noise of all the 3 (and it’s the one that should be used as last resource), DV being the best and most accurate 
Also, I have an interlaced footage that also has these coloured interlaced red-yellowish lines in between, I’d like to know if anyone has any advice? If I use the Bob Weaver de-interlace on FFMPEG, it removes it but then the picture quality is blurrier giving not so good results in VEIA (when compared to work directly on DV, which looks SO clear!).
Maybe for next update of the Dione, a deinterlace logarithm can be added to tackle also those coloured interlaced lines…
I gave a fix about this issue on the beta forum. This seems to occur on some 4:2:0 chroma subsampled sources.
To fix this you have to first upsample your source to 4:4:4 (RGB or YUV)
You can use avisynth and add a line like this at the end of your script.
ConvertToRGB(interlaced=true)
use eventually the right colorspace conversion if your source is Rec709, Rec 601 is default
ConvertToRGB(interlaced=true, matrix=“Rec709”)
1 Like
Ohhh thank you very much for this information, didn’t know it was posted before. So VEAI uses Rec601, so I gotta convert my DVD source from 709 to 601?
By any chance is there a similar script for ffmpeg? I’ll do some research since I can’t use Avisynth on a Mac 
Update: Exported as ProRes YUV444 and changed from 601 to 709, still get the artifacts Is there a thread that is more appropriate for these kind of questions?
I tried deinterlace mode on a DV video, and I’ve got black screen again. This trick no helped.
Any trick for AMD card? Driver is latest.
Update1: it seems I found a solution.
Down GPU memory frequency to lowest (1500), and GPU frequency to -20, power to -15.
Set VEAI mode from Dione-DV to Dione-TD. DV was better in preview, but my PC has crashed. Under Dione-TD working good, and I will see results after end.
I used 2x zoom only, all settings leaved on default.
Update2: I encoding now with Dione-TV, and this version quality is better as TD. My PC hasn’t crashed yet with new settings.
I have encountered two problems when deinterlacing PAL DV Wide using Dione DV v1 and the preset 100% (Denoise/DeBlock) and Keep audio.
- The length of the audio is longer than the video. Se this screenshot from Vegas 18.
Observation: when the render is finished, the progress bar skips backwards from for example 205 to 196 frames.
- The colour are stretched compared to the original DV-file. Compare with Se this screenshot from Vegas 18.
I hope this can be fixed in an update.
Jøran Toresen
The new Dione model is damn impressive, although I noticed a bug. With input video of MKV format, if the “Aspect Ratio” defined in the MKV wrapper is set to 4:3, then in VEAI during processing with Dione-TV, the frames would jump around every few frames. (The original video is from an interlaced DVD, and was not transcoded, just wrapped in MKV.)
I just did an upscale of a camera video that was turned, and something I mentioned many versions ago is still a problem. Camera videos have a “Rotation” field in the metafiles that most viewing programs will observe. VEAI is failing to read that field and write it out to the enhanced video. I am able to fix it without re-encoding, but would prefer not to have to. It doesn’t bother me that the video shows sideways or upside down in the VEAI preview, but if you would just read that field and write out the same value (when present) it would be nice.
All PAL content is being resized to 768 x 576 when using Dione but that should only apply to digital content like DV which uses the full 720 horizontal width. The horizontal width for analogue PAL video is 704 x 576 within a 720 frame. Therefore, if no cropping is applied before processing, there should be an option to set the output frame at 1x within Dione to be either 768 or 786 based on the content.
Digital, full 720 width video should be multiplied by 1.0667 to reach 768 wide, 4:3 ratio
Analogue, 704 width (with a 720 frame) video should be multiplied by 1.0926 to reach 786 wide, 4:3 ratio
Are you sure this is just limited to Dione? VEAI has always ignored horizontal frame width, and produced a value based on vertical frame and aspect. So if the file (as seen in mediainfo) says 720x576 4:3 VEAI will produce 768x576. Conversely, if a file says 768x576 5:4 then VEAI will produce 720x576.
You might be right, I usually try to avoid working with videos with inbuilt ratios and just import everything as pure video with no wrappers and eyeball whether it’s 704 or 720.
Shouldn’t it even better to adjust the AR based on the height ? I mean PAL 720X576 4:3 would produce a 720X540 output and so when upscaling 2X we would have a 1080p file. because 1152p in video is really uncommon and even unplayable with some external players.
Analogue PAL (704) in that case would ideally be set to 528 to be precisely 4:3 whereas digital PAL is fine at 540. But you’re downscaling from 576 to 528 or 540, losing resolution.
I see a lot of NTSC videos with the same issue, where it’s been set to 4:3 based on the 720 width and rendered as 720 x 540 when it should be 720 x 528. NTSC works a lot better in that regard as you’re upscaling from 480 to 528 or 540.
It ultimately depends on the source.
To be extremely nerdy, the exact width is 702.91525423728813559322033898305 for analogue PAL.
So the height should be 527.18644067796610169491525423729
Those decimal places keep me awake at night, as NTSC is a clean 704.
I am very impressed with the new Dione model. Before I had to use another app to deinterlace, but Dione handles it better than even the big-buck video editors and I appreciate the doubled frames, again, even the best interpolators fail to produce an image this clear. As this is only V1 of the Dione model, I imagine sharpness can be improved upon, but still near on par with the Artemis models using Digital8 and Super8 video imports. However, with VHS imports the clarity isn’t quite there yet, as the frames still remain blurry, albeit with noise removed and corners sharpened. I’m not even sure VHS imports have the detail required to do a proper upscale, but in all very useful models.
The new Artemis models are noticeably sharper than the previous versions, so amazed with the work you’re all doing. Frames do indeed look less plastic and even the motion appears smoother. In fact, if the footage is not interlaced, the Artemis models are my first go-to as they do such a great job with sharpening jagged edges. It’s hard to imagine how these models could even be improved upon, other than making the frames appear photo-perfect like it was shot on a high-end 4k camera, but they are closer than I imagined they would be.
3 Likes
I’ve got a few issues with this app for a while now and will be interested to know if anyone else has this and how you fixed it.
If I don’t run Video Enhance AI as Administrator, I don’t see any models in the AI Models list on the right. The list is blank and clicking it does nothing. As soon as I run it as Admin, I can now choose a model from the list and run a video through it.
Another issue is if I run the program as Administrator (which I must otherwise it doesn’t work), I can’t drag and drop files. I have to manually browse the folder by clicking on the center of the window or File > Add video. This issue happens only with Video Enhance AI and with none of the other Topaz software I own.
The can’t drag issue is normal for any program running as administrator. The other problem is not as obvious. Can you confirm that the models folder is in the default location C:\Users\[user]\AppData\Roaming\Topaz Labs LLC\Video Enhance AI?
Thanks for the quick reply, @deeImgGuy. The only folders I see in this directory are “Logs” and “userPresets”. The whole “Topaz Labs LLC” folder in C:\Users[user]\AppData\Roaming\Topaz Labs LLC\ is 11.1 kb large.
I suspected that it might not be there. Most likely it got put in your Program Files or some other “protected” folder, which a program can only access with Admin privileges. If you are comfortable looking in the registry you can look at HKEY_CURRENT_USER\SOFTWARE\Topaz Labs LLC\Video Enhance AI and the value modelDirPath. If that is anything but the proper user folder it is your problem.
It was always my understanding that the NTSC/PAL pixel aspect ratios were the same for both analog and digital signals. (Actually, if we’re being entirely pedantic, analog video doesn’t have pixel aspect ratios since pixels are a digital concept, so if we’re talking about pixel aspect ratios it’s all digital video, regardless of whether it originated in the analog or digital domain.)
Rec.601 was the standard which defined how analog video should be represented in the digital domain. Since in the analog video domain the vertical resolution was fixed and the horizontal resolution depended on the sampling frequency, what Rec.601 standardized was the sampling frequency (13.5 MHz) for digitizing PAL and NTSC video. That resulted in an 702x576 active image for 625-line 50 Hz video (aka PAL), and 714x486 active image for 525-line 60 Hz video (aka NTSC). Since many early codecs used 16x16 macroblocks the decision was made to align those digital resolutions to their nearest mod-16 boundaries, which is how they came up with 704x480 and 704x576. At the same time, the folks working on those standards wanted to allow some room for horizontal overscan (in case of sampling misalignments), so the maximum allowed resolution of digital SD video was increased to 720x480 and 720x576. That didn’t actually change the pixel aspect ratio though, it just horizontally enlarged the image area.
The standard Rec.601 pixel aspect ratios for SD digital video were therefore:
- 10:11 for 704x480 and 720x480 intended for 4:3 displays
- 40:33 for 704x480 and 720x480 intended for 16:9 displays
- 12:11 for 704x576 and 720x576 intended for 4:3 displays
- 16:11 for 704x576 and 720x576 intended for 16:9 displays
When you apply those PARs to 704x480 and 704x576, you get the familiar square pixel SD resolutions of 640x480, 853x480, 768x576 and 1024x576.
So then what about 720x480 & 720x576? Well, the pixel aspect ratios for those resolutions should be the same as for 704px, and when creating a full-frame 720x480/720x576 video one is supposed to either blank out the 8 pixels on each side to keep a 704px active image, or resize the image to have slightly higher vertical resolution (720x491 and 720x589) and then crop off the extra lines from top and bottom.
Now, as anyone who has managed to make it this far reading this post is probably well aware of - all this pixel ratio talk is some highly technical, super pedantic, obsessive compulsive stuff that makes no sense to people unless they’ve got a decade or two of experience in video signal processing. So naturally when DV cameras and DVDs became popular, a lot of times this math got completely ignored. Everyone wanted to squeeze out as much resolution out of digital SD video as possible, so authoring digital video as 704x480/704x576 didn’t make sense to most folks when they knew DV cameras recorded 720px of video and DVDs allowed all 720px to be used. In the end some DVDs ended up being made by strictly following the Rec.601 standard and using the exact PARs I listed above, but many (most?) ended up being made by just scaling 4:3 or 16:9 video to 720x480/720x576. The DVD standard unfortunately didn’t provide any metadata to distinguish between those too, so at this point trying to figure out if a 720x480 widescreen DVD should be resized to 853x480 or 873x480 to preserve the intended display aspect ratio is anyone’s guess. In the end, the difference is so subtle (2%) that I truly doubt anybody really cares. 
