I used VEAI to up-convert an old VHS tape containing 8mm film displayed on a wall. I went from 480p to 1080p and the results are pretty amazing. I do run into issues where people turn into like little pieces of rectangles and trees or grass scenery in the background is good, but looks very patterned instead of more randomized.
Maybe these before and after shots can show what I mean. In the end – the moving video looks awesome, but can’t seem to find a way to make people look more curvy (yet still sharp) and vegetation to not looks so patterned. I’ve been using Artemis Low Video Quality. If I go with Medium, it doesn’t make a big enough difference in quality.
Regardless - no complaints here! My customers love the quality increase.
I also convert 480p movies that I took long ago from my personal collection and upconvert them and what I found is that, version 2.6.0 is a great improvement compared to the earlier days such as 1.5.0 that I had used in the past when they had the 30 day trial without the watermark. I’m now re-converting some of what I did with 1.5.0 now with 2.6.0 and like you said, the quality of the conversion is simply amazing. Trees and vegetation are more texturized and believable in 2.6.0 and people are more defined, though need more improvement such as like you said be more curvy or natural rather being a bit blocky like Lego set pieces when the video source is poor. Still, Artemis in 2.6.0 is a vast improvement to the older AI models in 1.5.0. My observation is that as Topaz train these models to become smarter, so will the quality improvement of the conversion I suppose.
If you are running a business and need to keep up with quality improvements, you may want to consider paying the yearly updates. I held off buying the software until recently, because I didn’t think the software was polished enough. I’m glad I held off as the recent version is not only faster with my GPU, but even better in converting older 480p sources and I could justify paying its asking price.
I’m actually holding off some conversions until a newer version which is coming out in December which supposedly have some Video IO improvements.
hi, I’ve been working on 8mm hi8 sd videos digitized with a digital8 dv camcorder for 6 years. The best solution for now for me, with vegas pro, in the project I deinterlace with the gpu, on the original video I apply neatvideo (slightly) and two or three accent filters, I produce the video in uncompressed avi or yuv 10 bit in SD by re-interlacing the video correctly. Then with veai, first pass: dione dv (sd 720x576), second pass: artemis alliased and moire (SD), third pass: proteus (HD). Then re-import the video and refine the settings and filters in vegas pro (neatvideo, bbc magik sharp, bbc details, bbc dvfixer, color corrector, unsharp mask, grain). (problem at the moment, dione tv and dione dv reproduce nory flickers on the movements).
That is the problem that I also have. It looks like an oil painting with black edges and some patterns, here and there. It’s hard to find a way to avoid this problem sometimes with the settings, and its time consuming most of the time. The AI does its best, I know, but still. The rending is “OK-ish” or “Meh…” 50% of the time.
I think developers need to find and add more models and, of course, need more time to improve the rending when it’s a blurry video. I also believe it would be nice to have an option to protect the black edges to “smooth” the rending or to reduce some strange patterns.
That said, we have the same problem with Gigapixel. Especially the skin. Sometimes, it adds dark lines to the skin or bizarre and ugly patterns.
I can tell by the picture quality of your original video that you’re probably using one of those EasyCap USB, or maybe an AV-to-HDMI converter with a HDMI to USB capture card. If that’s the case let tell you it’s destroying the picture quality. I’ve both of these options, and finally an old Pixelview PlayTV Pro PCI capture card I bought in 2001, and the quality is miles away. Either way, if your capture card is deinterlancing your video, you need to change it to something that gets you interlaced, so then you can process it properly with yadiff or QTGMC. I hope this helps.
I’m not sure what original video quality you are speaking about?
With my captures, I use professional capture equipment (with SDI cabling), high end VCRs and I de-interlace the captured video prior to upscaling with VEIA. I stay with ProRes files.
I don’t see VEIA providing the best de-interlacing in terms of quality in additional to the time factor.
If you are capturing your video with equipment that sells for less than $300 and you aren’t using a professional TBC – you’ve already ruined your source file.
The problem I see is that Artemis LQ – needs serious help. There is other software that can make a face look like something like a face instead of an alien. I would rather have slightly intelligent fuzzy video in place of straight lines.
I’m barely using Artemis LQ anymore. It is too geometrical. If it can’t interpret a face, then it should keep the face fuzzy.
I wish there was a way to create a slight intelligent grain around geometric lines. You need kind of like a Artemis Medium to LQ mode version.
One thing that I really see as a problem is when blurry landscapes are “enhanced” – the enhancement looks like a pattern. There needs to be some randomness involved.
My big request would be face detection and to deal with it properly instead of alien faces. Keep it fuzzy if it can’t make it less geometrical.
I am also seeing little sections of video that flickers with Artemis medium. I have had to edit out those sections are replace them with the original video, which is a pain.
Finally, adding grain is great – but it is not intelligent. It needs dimension based on where it is applied.
AI technology is rapidly getting better – but I am concerned that this product is not keeping up.
for my part I digitized my videos in firewire in avi (22 - 23 mbp/s) so I think it’s the best possible quality. for the grain, it is applied in the hight, medium and low, it will be cool if it is less accentuated in the low tones. For deinterlacing, qtgcm is the best, but in “slow or placeboo” even if it removes a lot of noise and gives straight edges, it removes a bit of detail, set with the “medium” preset, it is a good compromise
