That's so true. It depresses me to see the amount of vintage clips online which just happened to be digitized when we didn't care or could not have better resolutions. A lot of old cartoons or fun tv ads are only available on overly compressed 320x240 versions or even worse 😭
The best way to preserve analog video tapes is to record them at video head service output and record that to file for further processing or future when better processing will be available. VHS-Decode is the future.
So will a massive amount of digital content to be fair. I expect large amounts have already been lost even from a decade ago, let alone beyond that. There are many tapes that are 30+ years old, with minimal if any degradation(though no doubt many tapes have also been thrown away, regardless of condition) I can imagine a lot of current digital videos will be permanently lost well before 30 years have passed. I suspect a huge amount of youtube videos, for example, aren't stored anywhere else.
I used QTGMC in the past. Tested it against Topaz. While Topaz is quite OK, one area is completely bad: color interpolating. QTGMC is the only script to date I found that does deinterlace color properly and put it in right spot. While most deinterlacers just get rid of one field colors (like Topaz, as well) and there is visible stripes in red fast moving colors, some does even worse and color jumps ahead/behind luma. QTGMC does proper interpolation and colors are fixed where they suppose to be. I had a lot of run and gun HDV footage in the past and it was notorious for blocking and deinterlacing that meant reducing quality and "jumpy" colors with any tool, except QTGMQ. That resulted in 20% boost of resolution if proper (slow) setting was done. I tested one footage in best light, closeup of woman lower legs in red sandals, walking perpendicular to the camera. While Topaz can mask the issue quite well in most shots, there it was revealed that QTGMC is unbeatable. And images were very like each other, as you already mentioned. I concluded that for deinterlacing QTGMC would still be the best option, with some noise /grain shaping and than Topaz for upscaling fo further sharpening and deblocking. Also noise reduction in Topaz is quite bad. It reduces noise, for sure, but it works too much on time axis and blur details if moving scene. Neat Video is still better. Of course for just one click solution for quick improvement, Topaz is good (best).
Offcourse. It is lossless digital video. Videoquality on old DV cameras is dependendant on the Video Sensor Chip. For example 3CCD for better color or high quality CMOS chips. And offcourse optics. Zombie Movie 28 Days later was shot on an inexpensive Semi Pro DV camera (Canon XL1). Though they also used special 35mm camera lens adapter, otherwise they would not get the same kind of Depth of Field as traditional 35mm Film Stock cameras have. Video quality is not comparable today, as we have come a long way with digital video since then. But this were an international feature film. And it still looks good. Lighting and optics and post-production can make a huge difference even with a shitty videocamera.
@@zerazara it's not lossless technically. DV video is 25mbps. It's just such a low rate that there is no perceived loss. I'm not sure how he was capturing his VHS tapes here, but if he's capturing to MPEG to start with then he's already suffering quality loss. I'm really old school. I use a DV camera with digital pass though to capture straight to DV-AVI, then compress for PC or onto DVD for people. If my final output is going to be MPEG I compress to 8mbps DV resolution and leave it interlaced. I've found this works really well. If the person wants DVD I try not to put more than 2 hours on a disc and use a bitrate calculator to get the best results. I've found over the years that 2 hours or less there is no perceived loss to any typical eye viewing the end results.
@@stantheman1976 I don't know the technical specs and how every codec works. Offcourse the best would be no codec at all and have everything RAW. But that takes up enormous amounts of dataspace. But my point was not about the codec. My point was that digital video cameras has way better quality than analog(VHS, Betamax etc) becaus the is minimal with artifacts. But you are right about digital video encoding degrading the potential of video quality of the sensor chip in the camcorder. It is not always just the video codec used. It is also the electronics encoding signals from the video sensor. One example of that is the HW Andromeda Hack for the Panasonic AG- DVX100. It took a commercial semi professional camera and turned it in to a beast for it's time. Quality achieved at the time were a milestone and not far from where RED cameras was at the time. The AG-DVX100 is a very old camera so it shows how great it really were.
I’ve done this but I used separate applications to de interlacing and upscale. I captured the vhs with a canopus advc 110 into Fcp 7 which does a really good job de interlacing. I outputted a 720p ProRes 422 file that I plugged into topaz and upscaled to 4K.
I think with VHS there's a lot of times where they have so much over sharpening/halos where it actually looks better to soften it than to add more sharpness. TV master tapes naturally do not have those super sharp edges. You just have to use a plugin to make sure you're only softening the super sharp points, not everything, and the result looks much more natural.
Well, AI might be able to address some of these things very soon, that's the whole point. Eventually, say in Stable Diffusion, you could include in the prompt, deinterlace or remove dropouts or even correct haloing. Someone will have to feed an AI examples of these issues so the AI will best know.
Oh also video enhance ai isn't really meant for cartoons or anime. For those the best I've found is the DigitalFrames2.1 Aggressive ESRGAN model mixed with the DigitalFrames2.5 ESRGAN model. 2.1 has better colors and line reproduction, 2.5 has better scene detail reproduction. So mixing them something like 70% 2.1 and 30% 2.5 works out quite well. It just takes forever. Also you can use these models easily by using cupscale.
I have some VHS tapes mailed to me by Nintendo Power around 96/97 still! And… about 30 family videos I need to digitize too. Had no idea tales could expire till now
Great stuff! I unfortunately lost all my home movies from childhood in a move back in 2005, but it's awesome that you've been able to hold onto yours and utilize AI in such a unique way. It'll only get better from here as the technology improves.
AI can help with extracting more "quality" from VHS tapes. But it will have much easier job if these tapes are scanned and recorded with direct RF sampling from video heads and then decoded in software to bypass ageing electronics. Cheap hardware like Conexant CX2388X based capture card is enough to get custom RAW mode driver running. Domesday86 VHS-Decode is the future of archiving analog video tapes.
im testing this myself right now and im thinking of keeping the originals as is. The upscalers seem to just add a sharpness filter and maybe a smoothness filter. they end up looking over processed instead of something that would seem smarter for an ai.
A light to moderate(depending on the source tape quality) denoise + sharpen filter seems to work reasonably well, while being free and requiring much less processing power/time.
I've been using my retrotink5x to get my 8mm tapes outputting at svideo to upscale them to 1080p. They look great, but using the AI might be able to make it even better! Thanks for the great info!
You could also try playing it in a Digital8 camera model that supports Video8(& Hi8) playback - these do the analog to digital conversion themselves(presumably in high quality) and send the digital signal out via the firewire connection. These cameras are becoming quite valuable/expensive though.
Can you recommend a laptop and/or the hardware/specs needed to allow AI software to upscale home videos to HD within a reasonable length of time. The Topaz Enhance AI software is telling me it will take >2 days to upscale a 1 hr. long analog video! I'd appreciate whatever you can offer as I have hundreds of home movies in need of being digitized and upgraded. Thank you for your expertise!
I've tried *Topaz* , I see you're also getting weird interlaced artifacting on reds on the video enhance AI, I noticed @ 6:53 . I'm getting those same artifacts on reds
I managed to capture family videos from several miniDV tapes using mini USB port on my Panasonic camera. Just plugged it via mini USB -> USB type A cable. I used WinDV program which is very handy. But still I'm not sure if FireWire (IEEE 1394) capture would be better or not. Visually the quality is OK. Will you cover the process of capturing VHS/miniDV tapes and handling them in the video editors, please? :)
USB on MiniDV cameras is typically used for streaming and such, by compressing the image considerably. A firewire connection allow you to copy the video exactly as it is on the MiniDV tape(it's a digital format), so it's definitely the best option.
I usually use QTGMC to deinterlace and then depending on if the source needs to be 60fps or not I convert it to the appropriate fps and then run the deinterlaced footage through Topaz Video Enhance AI. I tend to like their Artemis models. I never really liked their models that do the deinterlacing themselves because as they deinterlace fine they don't upscale all that well.
20+ years ago I had a VHS DVDR combo machine to digitize all my VHS tapes. I then ripped the discs to HDD and encoded with the 3ivx codec. I still have many of those videos and the quality isn't too bad. I suppose I could now use Topaz or similar to upscale.
Any idea when your VHS capture guide will be out? I’ve been using an Sony digital 8 camcorder as a capture device from a VCR and capturing with Adobe premiere pro. The audio sync has been challenging, but I learned so much from your last video. Thanks for your hard work. You’re helping me preserve all the home movies my mother made who passed away a few years ago. Cheers!
Honestly, the best results I've even able to get has been by using a Digital8 camcorder as a destination for my VHS deck (my Sony can be set to use the composite port as an input... not just for output). In other words, I star by recording the VHS tape to a Digital8 tape. Once done, I use firmware to transfer the Digital8 tape to my PC (just like DV, I use the same format too). At present, I am experimenting with my Elgato HDMI capture card. This requires me to use either an composite to HDMI upscaler or an old USB composite capture device (by Ion). The latter option ditched in lieu of the aforementioned VHS to Digital8 to firmware/DV file. I'm 90% sure there is no composite to DV pass-through feature and so I am forced to use a D8 tape as the middleman, which still yields ideal results. I can't honestly remember the interlacing status of the resulting file but the data was so clean that it was a breeze to run through an encoder and decombing filter. I'm still working on it. Feel free to drop a comment on my YT channel if you want to check notes or have questions.
@@DJRonnieG Some MiniDV cameras also have analog inputs, which can passthrough to firewire (I have a couple myself) They can be found considerably cheaper than Digital8 cameras(particularly the models which can playback Video8 & Hi8, as they're valuable for conversion) The DV codec does limit the quality a little, but they're still good. Plain old DVD/HDD recorders are typically good too, though it can be a nuisance to get the video off on many, and also typically record to mpeg2. Some DVRs have an analog input, and at least some models allow you to copy the files off over the network(such as my Panasonic HD/DVB-T DVR) They typically also seem to record to mpeg2, in similar quality though.
Love your work! I have a question... I have a Lenovo Legion 5i laptop, with Intel i7 processor, 16gb RAM, and a GeForce RTX 3070 graphics card, which I thought was a powerful machine. Why does Topaz software take 17+ hours to upscale only 1 hour of captured home video? Are there specific settings to hasten the process you can share? Do I need more RAM? Please help!! Keep up the great work!
AI can inherently be extremely taxing on hardware, it's very different to standard upscaling which is typically straightforward repeated calculations. AI can be more like the software deciding what objects/shapes are in an image, then recreating a higher resolution version of those objects. ie: a bit like a human brain interpreting what we're looking at vs a calculator doing multiplications. Topaz's instructions should give details on recommended specifications, and which options take the longest.
Hey dude, could you suggest some software or way of working here? I have very scratchy S-VHS footage that has white little needlescratches flickering on it all the time. Is there some kind of software that can clean it up digitally?
First deinterlace with QTGMC (lossless) then use Video Enhance with the Artemis models. You can hack the config file to download older model versions. I Use Artemis LQ v8 at 1.2x. Turns out there are different models for different scales, at 1.21x render time increases and the results differs. There's no point in increasing VHS resolution beyond 720x480, the goal is just to reduce noise and sharpen the image. More observations: -The current Artemis LQ v12 is way too aggresive with noise reduction and sharpening. -LQ v11 produces a weird glowy effect, it's unusable. -LQ v10 it's really good for VHS, it has it's uses, but the again LQ v8 turned out to be the best, specially in low light homemade grainy footage.
@@schnullertroll4173 I was refering to the upscale percentage. Above 120% (1.2x) VEAI changes the models used (and the end result) and almost doubles the rendering time. I personally don't like how clear the edges become in relation to the low level of detail of the image itself as you increase the resolution.
Really wish Topaz would separate their deinterlacing algorithm from the denoising/sharpening garbage. It's overall much cleaner than even QTGMC. Also, have you tried ld/vhs-decode?
I haven’t tried vhs-decode because no one will give/sell me a complete package And it seems like from what I’ve been told the software side still has quite a ways to go, so there’s no point in rushing to it
@@EposVox Yeah, having to cobble together the hardware necessary is a problem, and I've also been holding off on trying it in the hopes audio decode will be added.
@@EposVox Indeed. I think ld-decode was more straight forward (in a general sense) since everything comes off the disc as one signal. Since VHS has separate output lines for video and audio, video decoding was started on first.
Getting the hardware is not that hard, as for VHS capture cheap CX2388X cards are enough for the format. It's possible to use them in multichannel configuration and they are still available but now as PCIex1 expansion with extra PCI brigde. Once capture is done the FLAC-compressed video/audio stream can be stored and archived once software side of things is mature.
You provide very interesting content.. I was struck at times during your delivery. (I mean absolutely no disrespect. Your voice has a great basis for Kermit the Frog. With just a little training, you could find another side hustle.) Thanks a great video.
@@EposVox Sir, As I mentioned, I mean NO disrespect. I am a TV producer and have mentored many news readers and weather presenters. If you took the time to read my message, I suggest there is a great vocal ability to conjure up a new talent. When you calm down, and listen again, I think you might see your talent. I am NOT putting you down. I am praising you for what may be a hidden talent. I am open and honest about this. Honesty.
S'funny how analogue tape / CCD sensors left everything with a cyan tinted mid colour band. There must be some correction for that. I havve an old Video8 camera so I might set it up on a colour card and calibrate it next to my Sony A7Rii and get the footage looking the same.
Is it better if simply upscaled to a "near hd equivalent" of about 960i? Or is fully going 4k better in terms of looking like it inferes fairky basic and efficiently the 240p ish res of the vhs feed?
I used several AI 4k enhancers on home movies from the late 80s, using every advanced option to make it look as best as possible- taking days to do each short video- and the end result has been the same every time: no changes in video quality. This is unacceptable. I paid a lot of money for these softwares that don't work.
You want the output res to be about 2x or else you just get painting-like artifacting, and also, the source footage needs to be on the verge of aliasing or else it might not restore anything. The second part was covered by 2kliksphilip and I believe LTT actually realised that too high of an upscale does nothing to the footage.
I bought topaz ai less than a week ago. I bought it to upscale dvds. I tried gaia, also arthemis then finally dione interlace. The interlaced looked terrible some of video seemed like the picture like sliced. I don't know how to describe it. Tbh I don't even know if dvds are interlace, I couldn't get the information on Google which was surprising. What model and settings would you recommend for dvds? P. S same here btw I wish I could change the fps output. I was sure I could do it but disappointed to find that you obviously can't.
From my understanding, almost all DVDs are interlaced. It seems DVD does support progressive(non-interlaced) video, but it was rarely used. A number of DVD players also converted the signal to progressive, which confused matters. Interlacing is part of the reason they're well suited to VHS+DVD recorder combo units - as they can record in the same interlaced format as VHS. You can often view video file information such as resolution, interlacing etc from the menu in players like VLC, or even sometimes just in the normal file properties in Windows.
Wow - thanks for this! I am trying Topaz but my ancient i7-4th gen and 2GB vid card, are not playing along. Any thoughts on RUclips Studio for fixing up old VHS material. Yes, I am a noob.
RUclips Studio isn't much use for anything, from what I've seen, it's extremely limited(they even removed previous features) There's a lot of free software around that can deinterlace, sharpen, scale etc though, if wanted.
From what I've seen, the difference between the best/most expensive capture device and quite affordable capture devices(like DVD recorders, passing through a MiniDV camcorder, etc) is quite small. The most important aspect seems to be the VCR itself, with an S-VHS deck(even if only playing VHS tapes) often being best, though they've become expensive.
I really like your videos. I have a bunch of VHS tapes. I am capturing them using a Canopus ADVC 110 on fire wire in 720 x 480 NTSC 29.976 FPS. I purchased Topaz Video Enhance AI. How do you setup your Topaz? Do I select Interlaced or Progressive? How do I deinterlace and what is the best present to use? Thanks
This may be a bit late for you, wow, still using firewire?? I had one of those with the ADVC110. I would assume you'd set up Topaz in interlace since the Canopus output the footage in SD interlace. I've since moved on from Canopus after several crapped out on me. I tried a bunch of cheap to relatively expensive upscalers that can do a good job deinterlacing but blows out the highlights. Currently trying out a Retrotinx 2x, made really for video game output. It does an okay job at deinterlacing, highlights aren't blown out, but it doesn't upscale. I'm wondering if AI can get rid of those few extra jaggies that the Retrotinx might have missed. But as they say "garbage in, garbage out". There is nothing that's going to make your SLP VHS home movies look like they're shot in HD.
@@inkermoy I don't really get the appeal of retrotinx or similar for recording, other than being quick - they're just doing what can be done in software. They're obviously useful for watching something with a potentially improved image, without needing to record first, though.
@@DoubleMonoLR Just like hardware, software can be hit and miss with deinterlacing. I've always had a hard time with software deinterlacing analog captures, not getting satisfactory results. I'm not a script-kitty enough to use AviSynth or the like, so with my workflow I'd rather have something to deinterlace and send a progressive signal to capture, rather than doing it in post. I've been messing around with Topaz and some analog sources and found that there are limitations in that I don't think anyone has really taught an AI how to really clean up noisy analog frames. I've seen AI do wonders with shoddy DVD transfer/compression of TV shows, bringing it up to the level of a 1080p stream. But there's just too much variance and warble in VHS sources for an AI to truly fix and stabilize the content.
I just stumbled across this so I'm brand new to your channel. I'm curious about your workflow when doing the initial capture from VHS to digital. Are you using a capture card and starting with MPEG files? I'm old school on the way I do things but I think it gives the best results possible. I use older DV cameras that have digital pass through. I take the footage straight from VHS to DV-AVI over firewire. Then if my final format is playback on PC or smart TV I'll encode MP4 at 720x480i, lower field first, 8mbps CBR. VLC Media Player does a really good job of deinterlacing during playback and TVs use hardware decoding which works best. If the person wants a DVD for playback I'll use a bitrate calculator to figure out how much I want to put on a disc. I've found with good starting footage that I can get anything 2 hours or less on a disc and there is no perceived loss to any normal eye. I recently transferred 90+ hours of VHS and VHS-C footage for a family friend this way. It takes a while to do it, but it's a workflow I learned MANY years ago and I know it gives good results.
It varies widely, the most important thing is probably just to get them done. It's tempting to look for the absolute best option, but it can end up being a waste of time, the improvements can also end up being pretty marginal. Even using an older combination VCR + DVD recorder unit typically gives good results, and is very easy.
Thanks for the review. I'd love to see more of VHS and 8mm restoration. I have one tape with game promos. Wouldn't like to send it away but maybe I could digitize it and send the mp4 to you?
9:15 This child in the video is probably the author of this video? That's where his craving for green came from - first he eats a green cake, and then dyes his hair green! There was definitely something in the food coloring!
You mentioned in this video that you do something with old video samples of games and demos included with old magazines. Do you just archive the data? If yes, do you have a list of what you already have? I have a handful from the 90's and I could see if what I have is of any use.
Perhaps in the future, like some of the photo enhancing AI we have now, we will have an AI competent at recognizing specific objects in noisy low resolution video and can perform similar enhancements. Still wouldn't be perfect and would take 'forever,' but since it's becoming possible for photos, it's not too unrealistic to expect it to become possible for videos on a frame by frame basis with enough data and training for an AI. Would be a pretty huge project, though, and I still wouldn't expect anything near perfect results. I think enhancing faces/people would go a long way to preserving important aspects of these old videos, but without similar enhancements for surrounding scenery and objects, the resulting videos would look very odd...
I'd imagine faces are probably the most difficult thing though unfortunately - AI could potentially make it look like *A* perfectly sharp face, but for lower quality/smaller images(ie: not much detail to base the AI generation on) not *THE* face of the person in the video (short of having a reference image attached for each person in the video)
This is a great idea, would really help a lot, I have many of them as well, I am only a little familiar with DaVinci I know the cake was not the focus but it sure looks good lol, yes you can definitely see a difference on some, it sure is fascinating what can be done today....gosh I remember the days...not only "Be kind Rewind" but they charged if you didn't lol
In certain global regions it’s unfortunately “improper” to specifically ask how to solve a “signal turns to a black screen” when some HDMI devices are connected in a certain way they aren’t supposed to. (I know, it’s sad)
I recently transferred some vhs tapes and used QTGMC Placebo and VEAI Artemis Dehalo with some sharpening afterwards. I don't like the VEAI Dione Filters. The Deinterlacing is ineffective and the upscale looks too aggressive. To retain the aspect ratio in VEAI I rendered QTGMC into a lossless intermediate and edited the metadata with mkvmerge. I think for Content in SD 720p is sufficient. Just use a player with madVR and SSIM or NGU. Upscaling to 4K takes too much time and storage. It's more important to export in ProRes in VEAI for editing and encode to 10 bit H265 afterwards. The H264 option in VEAI is only 8 bit.
I am still looking for a solution that does something similar to what MyHeritage does - i.e. adds extra details based on it's understanding of objects, but for video.
Frame rate is problem. Audio not synced. Either export frames and than combine it with audio in VirtualDub or reframe video to fix problem. Cumbersome additions step.
Those are all things AI is doing great work on, so yes? Though clearly not all are created equal as we saw here. I was really rooting for Resolve’s native options
@@EposVox i appreciate your videos and thoroughness I was just more speaking on the trendiness of the concept itself, peoples seemingly new ability to misunderstand what AI actually is and throwing it into random terms these days... even though everything is clearly all still closed loop. Now it's nothing more than a tag for their product that says "look im bleeding edge too" its gotten to me but i realize this isnt the place to gripe. my bad
@@smatchimo55 "Computer science" is a weird field full of misnomers. Even that name is wrong, it's actually a field of engineering for data processing solutions, such as data structures and algorithms, not about computers (which are subject of electrical engineering), and not a science (as it doesn't explore the laws of nature). Then "artificial intelligence" is a field of "computer science" that loosely collects various families of algorithms deemed useful to create autonomous agents, which are... i would say systems that are designed to exhibit useful behaviours when presented with faulty and incomplete information, but this is my personal definition, but don't you think if this is definition is acceptable, that image reconstruction would be a part of it? Convolutional neural networks and especially generative adversarial networks have been a major recent research focus in AI field, driven by machine learning, have demonstrated massive breakthroughs in reconstructing missing or damaged information from images and presumably they form the foundation of Topaz algorithms. I don't feel the "AI" moniker is anyhow abused by them, if that is the case. Closed loop how? GANs are generally open-loop at the application stage, closed-loop at training stage. But all you get in the software are pretrained models.
Need help restoring and digitising old VHS and camcorder tapes, tried to do it on my own and turned out not so good. I there a lab or someone I can send this to get it done professionally?
Hey man! Been watching your videos a lot. Been very helpful! Wanted to know your thoughts on this. I'm choosing between the Avermedia pw513, Logitech Brio, and maybe splurge a little bit more on the Sony a5100 and work my way up to buying a Sigma 16mm 1.4 lens. After around 7 months since your Avermedia Pw513 video, do you still think it's the best streamer Webcam in the market? Thanks, and hope you're well! :)
Looking for suggestions on what capture card, or component I can pick up to capture my 8mm and VHS tapes to convert to 1080 or 4k. I do have a Atomos Ninja 5.
@@EposVox I see :) I am just curious as I sent in a question a while ago, just something about command line options for stream encoding. Sorry for the late reply.
for now, my only options are using a panasonic s-vhs with a high-quality s-video cable that is short along with an older Avermedia Hybrid Volar Max USB device. I've been putting off the conversions but think I should shortly begin b4 it's too late. I may order a Retrotink 5x but then I'd need to buy an HDMI capture device. I will probably just capture as .avi to get best quality, then compress via handbrake, and hopefully sometime in near future, I can implement some of these new cutting edge programs to clean up the files if only a little. there is also super 8 footage which I may use a high res scanner to convert as I saw on a couple of videos, but that's...another story...
I have found old clips of one of my favorite drummers…Peter Criss of KISS…I wanna enhance the footage so much because the audio is funny as hell…love the guy…wish I can meet him…unfortunately he’s retired but if I ever meet him imma pour my heart out in as little time as possible.
@@EposVox I mean, it is. For example, "native" at 2:40 looks ok, but deinterlaced and scaled at 2:55 is stretched. Seems that you did not set PAR correctly. Same with 7:45 and 7:55. At 5:44 Dione AI is stretched compared to Resolve Super Scale. If a tool does not let you specify either PAR or DAR, it is garbage. Same with frame rate and scanning type - no override options, no cigar.
What i have done is My jvc camcorder on s video plug to a DVD recorder which is set on component Blue Green Red output to a aver game recorder on SD card frankly it looks Pretty Nice and the bitrate is 16 and get this 50 fps😊
This would be very cool. My mom had tapes made in 2006 (yes I know) with a long lost relative she found after our family got split 100 years ago. I am looking for ways to improve the quality a bit.
Topaz video AI 4.08 = hardly any difference upscaling/enhancing VHS video. In stark contrast to their still image upscaling which works amazingly well on all images.
I’m struggling with this now. Trying to do home videos and not as an expert. Then watching RUclips videos and the experts tell you how it has to be done the right way. I want an easy way we’re I don’t have to be an engineer and take this on as a career change! Haha! I guess all I’m saying is that I want a good result (maybe not the best) with the least amount of learning curve to get this project done and move on with my life.
AI upscaling isn’t something that has a generic “good easy” option, it’s very much a “tinker and cater to the specifics you want” kind of things as a brand new emerging technology heh
I'd recommend just buying an old VHS+DVD recorder(just be aware that VHS+DVD *Players* are more common, but obviously no use for copying VHS tapes) unit (or a separate VCR and DVD recorder, though a bit less convenient), and you'll be able to capture tapes in good quality with very minimal effort. The machines were, after all, largely designed and mass produced for exactly this purpose. For example, the video below was recorded on a similar device(a DVR, but is extremely similar to a normal DVD recorder) Note that youtube made it darker (and compresses the video quite a lot for low resolution video, though that's less of an issue): ruclips.net/video/6Z6Aazu27no/видео.html I did no processing on it, not even deinterlacing.
Yes: Topaz lacks aspect ratio control. If I feed original DVD (VOB) it will do great job for aspect - it scale horizontal properly more than vertical. If I deinterlace 1st, than I loose aspect and Topaz does worse job; image is not in proper aspect and is those missing details from horizontal is not "reproduced". If I scale image when deinterlacing, Topaz does even worse. Image image is visible odd.
Hear ya Playing around with idea of buying $40 camcorder standard definition See capture card to get on SIM card Also see card that scales say to 1080p You did not seem high on Ai solution As good as cards
I'm not seeing anything impressive with this except a heavy hand on the denoise switch & a sharpening filter on a source that already has major ring artifacts. Also doesn't use a dropout fixer, has choppy motion & the footage is unstabilised despite the addition of resolution. I've seen better example 15 years ago without any AI marketing crap. I know real AI restoration is around the corner in 5 years but these Topaz & Davinci examples are very mediocre.
QTGMC is just a deinterlacer, it will do nothing on upscale quality. While bicubic upscaler is like the lowest on any upscaling method, second to bilinear. A good quality one is nnedi3, but don't expect magic on home VHS. And the Topaz AI stuff are just scam
To upscale VHS videos makes absolutely 000 sense. Also bicubic is the most primitive upscaling method. It is made live by your media player or the tv set. During digitalizing VHS importand is to remove interlacing. The best devices offer field extension method. It is the best method which could be made live, during capturing the video. 5$ digitalizing devices and its attached software offer only taking odd or even frame, which means descreasing original VHS quality by 50%.
![Noob To Max With DRAGON REWORK In Blox Fruits [FULL MOVIE]](http://i.ytimg.com/vi/LnBMOinoOvA/mqdefault.jpg)
There's a lot of obscure content stuck on VHS, βeta and other tape formats that will disappear forever if not preserved.
That's so true. It depresses me to see the amount of vintage clips online which just happened to be digitized when we didn't care or could not have better resolutions. A lot of old cartoons or fun tv ads are only available on overly compressed 320x240 versions or even worse 😭
@@stereo123 we are lucky we can even access this stuff so fast.
The best way to preserve analog video tapes is to record them at video head service output and record that to file for further processing or future when better processing will be available. VHS-Decode is the future.
@@Zcooger the VHS Decode project is neat. I'm going to see if I can get the Library of Congress to fund that guy.
So will a massive amount of digital content to be fair. I expect large amounts have already been lost even from a decade ago, let alone beyond that.
There are many tapes that are 30+ years old, with minimal if any degradation(though no doubt many tapes have also been thrown away, regardless of condition)
I can imagine a lot of current digital videos will be permanently lost well before 30 years have passed. I suspect a huge amount of youtube videos, for example, aren't stored anywhere else.
I used QTGMC in the past. Tested it against Topaz. While Topaz is quite OK, one area is completely bad: color interpolating. QTGMC is the only script to date I found that does deinterlace color properly and put it in right spot. While most deinterlacers just get rid of one field colors (like Topaz, as well) and there is visible stripes in red fast moving colors, some does even worse and color jumps ahead/behind luma. QTGMC does proper interpolation and colors are fixed where they suppose to be. I had a lot of run and gun HDV footage in the past and it was notorious for blocking and deinterlacing that meant reducing quality and "jumpy" colors with any tool, except QTGMQ. That resulted in 20% boost of resolution if proper (slow) setting was done. I tested one footage in best light, closeup of woman lower legs in red sandals, walking perpendicular to the camera. While Topaz can mask the issue quite well in most shots, there it was revealed that QTGMC is unbeatable. And images were very like each other, as you already mentioned. I concluded that for deinterlacing QTGMC would still be the best option, with some noise /grain shaping and than Topaz for upscaling fo further sharpening and deblocking. Also noise reduction in Topaz is quite bad. It reduces noise, for sure, but it works too much on time axis and blur details if moving scene. Neat Video is still better. Of course for just one click solution for quick improvement, Topaz is good (best).
You should do this again with the updates that have been made to the entire spectrum of AI and there should be noticeable differences now.
Im always amazed at the quality of old DV tapes, ripped a pile and they still look good today just 4:3.
Offcourse. It is lossless digital video. Videoquality on old DV cameras is dependendant on the Video Sensor Chip. For example 3CCD for better color or high quality CMOS chips. And offcourse optics. Zombie Movie 28 Days later was shot on an inexpensive Semi Pro DV camera (Canon XL1). Though they also used special 35mm camera lens adapter, otherwise they would not get the same kind of Depth of Field as traditional 35mm Film Stock cameras have. Video quality is not comparable today, as we have come a long way with digital video since then. But this were an international feature film. And it still looks good. Lighting and optics and post-production can make a huge difference even with a shitty videocamera.
@@zerazara it's not lossless technically. DV video is 25mbps. It's just such a low rate that there is no perceived loss. I'm not sure how he was capturing his VHS tapes here, but if he's capturing to MPEG to start with then he's already suffering quality loss. I'm really old school. I use a DV camera with digital pass though to capture straight to DV-AVI, then compress for PC or onto DVD for people. If my final output is going to be MPEG I compress to 8mbps DV resolution and leave it interlaced. I've found this works really well. If the person wants DVD I try not to put more than 2 hours on a disc and use a bitrate calculator to get the best results. I've found over the years that 2 hours or less there is no perceived loss to any typical eye viewing the end results.
@@stantheman1976 I don't know the technical specs and how every codec works. Offcourse the best would be no codec at all and have everything RAW. But that takes up enormous amounts of dataspace.
But my point was not about the codec. My point was that digital video cameras has way better quality than analog(VHS, Betamax etc) becaus the is minimal with artifacts. But you are right about digital video encoding degrading the potential of video quality of the sensor chip in the camcorder. It is not always just the video codec used. It is also the electronics encoding signals from the video sensor. One example of that is the HW Andromeda Hack for the Panasonic AG- DVX100. It took a commercial semi professional camera and turned it in to a beast for it's time. Quality achieved at the time were a milestone and not far from where RED cameras was at the time. The AG-DVX100 is a very old camera so it shows how great it really were.
MinDV was a massive leap in quality!
@@stantheman1976 You capture Dv tapes in the DV codec.
I’ve done this but I used separate applications to de interlacing and upscale. I captured the vhs with a canopus advc 110 into Fcp 7 which does a really good job de interlacing. I outputted a 720p ProRes 422 file that I plugged into topaz and upscaled to 4K.
What kind of computer are you using and speed and memory?
7:45 what camera was used for the birthday party ? Its beautiful!
I think with VHS there's a lot of times where they have so much over sharpening/halos where it actually looks better to soften it than to add more sharpness. TV master tapes naturally do not have those super sharp edges. You just have to use a plugin to make sure you're only softening the super sharp points, not everything, and the result looks much more natural.
Good thoughts
Well, AI might be able to address some of these things very soon, that's the whole point. Eventually, say in Stable Diffusion, you could include in the prompt, deinterlace or remove dropouts or even correct haloing. Someone will have to feed an AI examples of these issues so the AI will best know.
Oh also video enhance ai isn't really meant for cartoons or anime. For those the best I've found is the DigitalFrames2.1 Aggressive ESRGAN model mixed with the DigitalFrames2.5 ESRGAN model. 2.1 has better colors and line reproduction, 2.5 has better scene detail reproduction. So mixing them something like 70% 2.1 and 30% 2.5 works out quite well. It just takes forever. Also you can use these models easily by using cupscale.
I have some VHS tapes mailed to me by Nintendo Power around 96/97 still!
And… about 30 family videos I need to digitize too.
Had no idea tales could expire till now
The Nintendo Power tapes!!
Try VHS-Decode them!
@@Zcooger VHS-decode isn't a realistic option for most people though.
Great stuff! I unfortunately lost all my home movies from childhood in a move back in 2005, but it's awesome that you've been able to hold onto yours and utilize AI in such a unique way. It'll only get better from here as the technology improves.
AI can help with extracting more "quality" from VHS tapes. But it will have much easier job if these tapes are scanned and recorded with direct RF sampling from video heads and then decoded in software to bypass ageing electronics. Cheap hardware like Conexant CX2388X based capture card is enough to get custom RAW mode driver running. Domesday86 VHS-Decode is the future of archiving analog video tapes.
im testing this myself right now and im thinking of keeping the originals as is. The upscalers seem to just add a sharpness filter and maybe a smoothness filter. they end up looking over processed instead of something that would seem smarter for an ai.
A light to moderate(depending on the source tape quality) denoise + sharpen filter seems to work reasonably well, while being free and requiring much less processing power/time.
I've been using my retrotink5x to get my 8mm tapes outputting at svideo to upscale them to 1080p. They look great, but using the AI might be able to make it even better! Thanks for the great info!
You could also try playing it in a Digital8 camera model that supports Video8(& Hi8) playback - these do the analog to digital conversion themselves(presumably in high quality) and send the digital signal out via the firewire connection. These cameras are becoming quite valuable/expensive though.
Can you recommend a laptop and/or the hardware/specs needed to allow AI software to upscale home videos to HD within a reasonable length of time. The Topaz Enhance AI software is telling me it will take >2 days to upscale a 1 hr. long analog video! I'd appreciate whatever you can offer as I have hundreds of home movies in need of being digitized and upgraded. Thank you for your expertise!
I've tried *Topaz* , I see you're also getting weird interlaced artifacting on reds on the video enhance AI, I noticed @ 6:53 . I'm getting those same artifacts on reds
I managed to capture family videos from several miniDV tapes using mini USB port on my Panasonic camera. Just plugged it via mini USB -> USB type A cable. I used WinDV program which is very handy. But still I'm not sure if FireWire (IEEE 1394) capture would be better or not. Visually the quality is OK. Will you cover the process of capturing VHS/miniDV tapes and handling them in the video editors, please? :)
USB on MiniDV cameras is typically used for streaming and such, by compressing the image considerably.
A firewire connection allow you to copy the video exactly as it is on the MiniDV tape(it's a digital format), so it's definitely the best option.
I usually use QTGMC to deinterlace and then depending on if the source needs to be 60fps or not I convert it to the appropriate fps and then run the deinterlaced footage through Topaz Video Enhance AI. I tend to like their Artemis models. I never really liked their models that do the deinterlacing themselves because as they deinterlace fine they don't upscale all that well.
20+ years ago I had a VHS DVDR combo machine to digitize all my VHS tapes. I then ripped the discs to HDD and encoded with the 3ivx codec. I still have many of those videos and the quality isn't too bad. I suppose I could now use Topaz or similar to upscale.
Any idea when your VHS capture guide will be out? I’ve been using an Sony digital 8 camcorder as a capture device from a VCR and capturing with Adobe premiere pro. The audio sync has been challenging, but I learned so much from your last video. Thanks for your hard work. You’re helping me preserve all the home movies my mother made who passed away a few years ago. Cheers!
Honestly, the best results I've even able to get has been by using a Digital8 camcorder as a destination for my VHS deck (my Sony can be set to use the composite port as an input... not just for output).
In other words, I star by recording the VHS tape to a Digital8 tape. Once done, I use firmware to transfer the Digital8 tape to my PC (just like DV, I use the same format too).
At present, I am experimenting with my Elgato HDMI capture card. This requires me to use either an composite to HDMI upscaler or an old USB composite capture device (by Ion). The latter option ditched in lieu of the aforementioned VHS to Digital8 to firmware/DV file.
I'm 90% sure there is no composite to DV pass-through feature and so I am forced to use a D8 tape as the middleman, which still yields ideal results. I can't honestly remember the interlacing status of the resulting file but the data was so clean that it was a breeze to run through an encoder and decombing filter.
I'm still working on it. Feel free to drop a comment on my YT channel if you want to check notes or have questions.
@@DJRonnieG Some MiniDV cameras also have analog inputs, which can passthrough to firewire (I have a couple myself) They can be found considerably cheaper than Digital8 cameras(particularly the models which can playback Video8 & Hi8, as they're valuable for conversion)
The DV codec does limit the quality a little, but they're still good. Plain old DVD/HDD recorders are typically good too, though it can be a nuisance to get the video off on many, and also typically record to mpeg2. Some DVRs have an analog input, and at least some models allow you to copy the files off over the network(such as my Panasonic HD/DVB-T DVR) They typically also seem to record to mpeg2, in similar quality though.
it;s already done. google vhs-decode project
At 5:45 there is an obvious change in horizontally stretching.
Love your work! I have a question...
I have a Lenovo Legion 5i laptop, with Intel i7 processor, 16gb RAM, and a GeForce RTX 3070 graphics card, which I thought was a powerful machine. Why does Topaz software take 17+ hours to upscale only 1 hour of captured home video? Are there specific settings to hasten the process you can share? Do I need more RAM? Please help!! Keep up the great work!
Good question, I'd like to know too
AI can inherently be extremely taxing on hardware, it's very different to standard upscaling which is typically straightforward repeated calculations. AI can be more like the software deciding what objects/shapes are in an image, then recreating a higher resolution version of those objects. ie: a bit like a human brain interpreting what we're looking at vs a calculator doing multiplications.
Topaz's instructions should give details on recommended specifications, and which options take the longest.
Hey dude, could you suggest some software or way of working here? I have very scratchy S-VHS footage that has white little needlescratches flickering on it all the time. Is there some kind of software that can clean it up digitally?
Not yet. All actual options are disappointing when dealing with normal VHS content. I am sure there will be some time, but we need to wait more.
First deinterlace with QTGMC (lossless) then use Video Enhance with the Artemis models. You can hack the config file to download older model versions. I Use Artemis LQ v8 at 1.2x. Turns out there are different models for different scales, at 1.21x render time increases and the results differs. There's no point in increasing VHS resolution beyond 720x480, the goal is just to reduce noise and sharpen the image.
More observations:
-The current Artemis LQ v12 is way too aggresive with noise reduction and sharpening.
-LQ v11 produces a weird glowy effect, it's unusable.
-LQ v10 it's really good for VHS, it has it's uses, but the again LQ v8 turned out to be the best, specially in low light homemade grainy footage.
Which one is the config file?
brilliant !
What do you mean with at 1.2? Grain Amount and size, both the same?
@@schnullertroll4173 I was refering to the upscale percentage. Above 120% (1.2x) VEAI changes the models used (and the end result) and almost doubles the rendering time. I personally don't like how clear the edges become in relation to the low level of detail of the image itself as you increase the resolution.
I may have missed it, but what are you using fir the initial VHS-to-PC rip, and what format are you using for the capture (avi, mp4, etc)? Thanks!
Thanks for this.. can we have your TUTORIAL for STAXRIP bicubic upscaling and QTGMC deinterlacing encoding???
Really wish Topaz would separate their deinterlacing algorithm from the denoising/sharpening garbage. It's overall much cleaner than even QTGMC. Also, have you tried ld/vhs-decode?
I haven’t tried vhs-decode because no one will give/sell me a complete package
And it seems like from what I’ve been told the software side still has quite a ways to go, so there’s no point in rushing to it
@@EposVox Yeah, having to cobble together the hardware necessary is a problem, and I've also been holding off on trying it in the hopes audio decode will be added.
I didn’t even know it didn’t support audio yet. Yikes
@@EposVox Indeed. I think ld-decode was more straight forward (in a general sense) since everything comes off the disc as one signal. Since VHS has separate output lines for video and audio, video decoding was started on first.
Getting the hardware is not that hard, as for VHS capture cheap CX2388X cards are enough for the format. It's possible to use them in multichannel configuration and they are still available but now as PCIex1 expansion with extra PCI brigde.
Once capture is done the FLAC-compressed video/audio stream can be stored and archived once software side of things is mature.
I'll be recapturing our family trip to Newfoundland for the rest of my life. Good video
why bicubic upscaling? aren't lanczos4 and spline16-144 better?
Ehhhh
spline scaling is ideal here
We should wait until generative AIs will be able to analyze all details on the videos and reconstruct them from scratch
You provide very interesting content.. I was struck at times during your delivery. (I mean absolutely no disrespect. Your voice has a great basis for Kermit the Frog. With just a little training, you could find another side hustle.) Thanks a great video.
My voice is nothing like Kermit, that’s incredibly insulting and just not even remotely accurate
@@EposVox Sir, As I mentioned, I mean NO disrespect. I am a TV producer and have mentored many news readers and weather presenters. If you took the time to read my message, I suggest there is a great vocal ability to conjure up a new talent. When you calm down, and listen again, I think you might see your talent. I am NOT putting you down. I am praising you for what may be a hidden talent. I am open and honest about this. Honesty.
S'funny how analogue tape / CCD sensors left everything with a cyan tinted mid colour band. There must be some correction for that. I havve an old Video8 camera so I might set it up on a colour card and calibrate it next to my Sony A7Rii and get the footage looking the same.
Is it better if simply upscaled to a "near hd equivalent" of about 960i? Or is fully going 4k better in terms of looking like it inferes fairky basic and efficiently the 240p ish res of the vhs feed?
You’ve really inspired me to keep going with my gaming channel.,!
I used several AI 4k enhancers on home movies from the late 80s, using every advanced option to make it look as best as possible- taking days to do each short video- and the end result has been the same every time: no changes in video quality. This is unacceptable. I paid a lot of money for these softwares that don't work.
You want the output res to be about 2x or else you just get painting-like artifacting, and also, the source footage needs to be on the verge of aliasing or else it might not restore anything. The second part was covered by 2kliksphilip and I believe LTT actually realised that too high of an upscale does nothing to the footage.
I bought topaz ai less than a week ago. I bought it to upscale dvds. I tried gaia, also arthemis then finally dione interlace. The interlaced looked terrible some of video seemed like the picture like sliced. I don't know how to describe it. Tbh I don't even know if dvds are interlace, I couldn't get the information on Google which was surprising. What model and settings would you recommend for dvds?
P. S same here btw I wish I could change the fps output. I was sure I could do it but disappointed to find that you obviously can't.
From my understanding, almost all DVDs are interlaced. It seems DVD does support progressive(non-interlaced) video, but it was rarely used.
A number of DVD players also converted the signal to progressive, which confused matters.
Interlacing is part of the reason they're well suited to VHS+DVD recorder combo units - as they can record in the same interlaced format as VHS.
You can often view video file information such as resolution, interlacing etc from the menu in players like VLC, or even sometimes just in the normal file properties in Windows.
Wow - thanks for this! I am trying Topaz but my ancient i7-4th gen and 2GB vid card, are not playing along. Any thoughts on RUclips Studio for fixing up old VHS material. Yes, I am a noob.
RUclips Studio isn't much use for anything, from what I've seen, it's extremely limited(they even removed previous features) There's a lot of free software around that can deinterlace, sharpen, scale etc though, if wanted.
This is great. Did you do a video on the next hardware to capture vhs yet?
From what I've seen, the difference between the best/most expensive capture device and quite affordable capture devices(like DVD recorders, passing through a MiniDV camcorder, etc) is quite small. The most important aspect seems to be the VCR itself, with an S-VHS deck(even if only playing VHS tapes) often being best, though they've become expensive.
Brah, how much time does it take to use your method?
I really like your videos.
I have a bunch of VHS tapes. I am capturing them using a Canopus ADVC 110 on fire wire in 720 x 480 NTSC 29.976 FPS. I purchased Topaz Video Enhance AI. How do you setup your Topaz? Do I select Interlaced or Progressive? How do I deinterlace and what is the best present to use? Thanks
This may be a bit late for you, wow, still using firewire?? I had one of those with the ADVC110. I would assume you'd set up Topaz in interlace since the Canopus output the footage in SD interlace.
I've since moved on from Canopus after several crapped out on me. I tried a bunch of cheap to relatively expensive upscalers that can do a good job deinterlacing but blows out the highlights. Currently trying out a Retrotinx 2x, made really for video game output. It does an okay job at deinterlacing, highlights aren't blown out, but it doesn't upscale. I'm wondering if AI can get rid of those few extra jaggies that the Retrotinx might have missed.
But as they say "garbage in, garbage out". There is nothing that's going to make your SLP VHS home movies look like they're shot in HD.
@@inkermoy I don't really get the appeal of retrotinx or similar for recording, other than being quick - they're just doing what can be done in software. They're obviously useful for watching something with a potentially improved image, without needing to record first, though.
@@DoubleMonoLR Just like hardware, software can be hit and miss with deinterlacing. I've always had a hard time with software deinterlacing analog captures, not getting satisfactory results. I'm not a script-kitty enough to use AviSynth or the like, so with my workflow I'd rather have something to deinterlace and send a progressive signal to capture, rather than doing it in post.
I've been messing around with Topaz and some analog sources and found that there are limitations in that I don't think anyone has really taught an AI how to really clean up noisy analog frames. I've seen AI do wonders with shoddy DVD transfer/compression of TV shows, bringing it up to the level of a 1080p stream. But there's just too much variance and warble in VHS sources for an AI to truly fix and stabilize the content.
I just stumbled across this so I'm brand new to your channel. I'm curious about your workflow when doing the initial capture from VHS to digital. Are you using a capture card and starting with MPEG files? I'm old school on the way I do things but I think it gives the best results possible. I use older DV cameras that have digital pass through. I take the footage straight from VHS to DV-AVI over firewire. Then if my final format is playback on PC or smart TV I'll encode MP4 at 720x480i, lower field first, 8mbps CBR. VLC Media Player does a really good job of deinterlacing during playback and TVs use hardware decoding which works best. If the person wants a DVD for playback I'll use a bitrate calculator to figure out how much I want to put on a disc. I've found with good starting footage that I can get anything 2 hours or less on a disc and there is no perceived loss to any normal eye. I recently transferred 90+ hours of VHS and VHS-C footage for a family friend this way. It takes a while to do it, but it's a workflow I learned MANY years ago and I know it gives good results.
What would currently be my best option for enhancing old vhs family recordings?
It varies widely, the most important thing is probably just to get them done. It's tempting to look for the absolute best option, but it can end up being a waste of time, the improvements can also end up being pretty marginal.
Even using an older combination VCR + DVD recorder unit typically gives good results, and is very easy.
Thanks for the review. I'd love to see more of VHS and 8mm restoration. I have one tape with game promos. Wouldn't like to send it away but maybe I could digitize it and send the mp4 to you?
9:15 This child in the video is probably the author of this video? That's where his craving for green came from - first he eats a green cake, and then dyes his hair green! There was definitely something in the food coloring!
You mentioned in this video that you do something with old video samples of games and demos included with old magazines. Do you just archive the data? If yes, do you have a list of what you already have? I have a handful from the 90's and I could see if what I have is of any use.
Perhaps in the future, like some of the photo enhancing AI we have now, we will have an AI competent at recognizing specific objects in noisy low resolution video and can perform similar enhancements. Still wouldn't be perfect and would take 'forever,' but since it's becoming possible for photos, it's not too unrealistic to expect it to become possible for videos on a frame by frame basis with enough data and training for an AI. Would be a pretty huge project, though, and I still wouldn't expect anything near perfect results.
I think enhancing faces/people would go a long way to preserving important aspects of these old videos, but without similar enhancements for surrounding scenery and objects, the resulting videos would look very odd...
I'd imagine faces are probably the most difficult thing though unfortunately - AI could potentially make it look like *A* perfectly sharp face, but for lower quality/smaller images(ie: not much detail to base the AI generation on) not *THE* face of the person in the video (short of having a reference image attached for each person in the video)
This is a great idea, would really help a lot, I have many of them as well, I am only a little familiar with DaVinci I know the cake was not the focus but it sure looks good lol, yes you can definitely see a difference on some, it sure is fascinating what can be done today....gosh I remember the days...not only "Be kind Rewind" but they charged if you didn't lol
Is there a good specific product recommendation to solve “connection issues” (*cough*) for sources that use 18 Gbit/s HDMI 2.0b (HDR & all)?
I don’t know what “connection issues cough” is supposed to mean
In certain global regions it’s unfortunately “improper” to specifically ask how to solve a “signal turns to a black screen” when some HDMI devices are connected in a certain way they aren’t supposed to.
(I know, it’s sad)
@@EposVox HDCP (copy protection) presumably.
I recently transferred some vhs tapes and used QTGMC Placebo and VEAI Artemis Dehalo with some sharpening afterwards. I don't like the VEAI Dione Filters. The Deinterlacing is ineffective and the upscale looks too aggressive. To retain the aspect ratio in VEAI I rendered QTGMC into a lossless intermediate and edited the metadata with mkvmerge. I think for Content in SD 720p is sufficient. Just use a player with madVR and SSIM or NGU. Upscaling to 4K takes too much time and storage. It's more important to export in ProRes in VEAI for editing and encode to 10 bit H265 afterwards. The H264 option in VEAI is only 8 bit.
I am still looking for a solution that does something similar to what MyHeritage does - i.e. adds extra details based on it's understanding of objects, but for video.
Frame rate is problem. Audio not synced. Either export frames and than combine it with audio in VirtualDub or reframe video to fix problem. Cumbersome additions step.
I like how you can just peg AI on to any new encoding/image recognition/sampling program and just have people eat that shit up.
Those are all things AI is doing great work on, so yes?
Though clearly not all are created equal as we saw here. I was really rooting for Resolve’s native options
@@EposVox i appreciate your videos and thoroughness I was just more speaking on the trendiness of the concept itself, peoples seemingly new ability to misunderstand what AI actually is and throwing it into random terms these days... even though everything is clearly all still closed loop. Now it's nothing more than a tag for their product that says "look im bleeding edge too" its gotten to me but i realize this isnt the place to gripe. my bad
@@smatchimo55 "Computer science" is a weird field full of misnomers. Even that name is wrong, it's actually a field of engineering for data processing solutions, such as data structures and algorithms, not about computers (which are subject of electrical engineering), and not a science (as it doesn't explore the laws of nature). Then "artificial intelligence" is a field of "computer science" that loosely collects various families of algorithms deemed useful to create autonomous agents, which are... i would say systems that are designed to exhibit useful behaviours when presented with faulty and incomplete information, but this is my personal definition, but don't you think if this is definition is acceptable, that image reconstruction would be a part of it? Convolutional neural networks and especially generative adversarial networks have been a major recent research focus in AI field, driven by machine learning, have demonstrated massive breakthroughs in reconstructing missing or damaged information from images and presumably they form the foundation of Topaz algorithms. I don't feel the "AI" moniker is anyhow abused by them, if that is the case.
Closed loop how? GANs are generally open-loop at the application stage, closed-loop at training stage. But all you get in the software are pretrained models.
@@SianaGearz Well said I am going to digest your comment for a bit. Thanks for your response.
Need help restoring and digitising old VHS and camcorder tapes, tried to do it on my own and turned out not so good. I there a lab or someone I can send this to get it done professionally?
Cool video, thanks for testing!
Did you try ESRGAN upscaling?
Hey man! Been watching your videos a lot. Been very helpful! Wanted to know your thoughts on this.
I'm choosing between the Avermedia pw513, Logitech Brio, and maybe splurge a little bit more on the Sony a5100 and work my way up to buying a Sigma 16mm 1.4 lens.
After around 7 months since your Avermedia Pw513 video, do you still think it's the best streamer Webcam in the market?
Thanks, and hope you're well! :)
Looking for suggestions on what capture card, or component I can pick up to capture my 8mm and VHS tapes to convert to 1080 or 4k. I do have a Atomos Ninja 5.
How often do you check messages sent with donations? Just curious about something :)
Usually every time! My email is always flooded but I generally try to always get to them
@@EposVox I see :) I am just curious as I sent in a question a while ago, just something about command line options for stream encoding. Sorry for the late reply.
Do you have any updates to this after 1year? :)
I can't find the video of the hardware used to capture the VHS footage, can someone link it?
Video Enhance AI is great...when it works and doesn't shut down the PC entirely.
for now, my only options are using a panasonic s-vhs with a high-quality s-video cable that is short along with an older Avermedia Hybrid Volar Max USB device. I've been putting off the conversions but think I should shortly begin b4 it's too late. I may order a Retrotink 5x but then I'd need to buy an HDMI capture device. I will probably just capture as .avi to get best quality, then compress via handbrake, and hopefully sometime in near future, I can implement some of these new cutting edge programs to clean up the files if only a little.
there is also super 8 footage which I may use a high res scanner to convert as I saw on a couple of videos, but that's...another story...
I have found old clips of one of my favorite drummers…Peter Criss of KISS…I wanna enhance the footage so much because the audio is funny as hell…love the guy…wish I can meet him…unfortunately he’s retired but if I ever meet him imma pour my heart out in as little time as possible.
Wonder if it is possible to surpass dvd quality using ai remastering + upscaled vhs..
Good that you convert to 60p. But the aspect ratio is wrong.
I mean, it's not
@@EposVox I mean, it is. For example, "native" at 2:40 looks ok, but deinterlaced and scaled at 2:55 is stretched. Seems that you did not set PAR correctly. Same with 7:45 and 7:55. At 5:44 Dione AI is stretched compared to Resolve Super Scale. If a tool does not let you specify either PAR or DAR, it is garbage. Same with frame rate and scanning type - no override options, no cigar.
What i have done is My jvc camcorder on s video plug to a DVD recorder which is set on component Blue Green Red output to a aver game recorder on SD card frankly it looks Pretty Nice and the bitrate is 16 and get this 50 fps😊
Can we get a Betamax guide? Asking for my parents 🙈
1:03 Funny, that's exactly why I'm here :) (yay, Jet Force Gemini)
Do you upload them somewhere?
This would be very cool. My mom had tapes made in 2006 (yes I know) with a long lost relative she found after our family got split 100 years ago. I am looking for ways to improve the quality a bit.
8:41 Game?
Jet Force Gemini
@@EposVox thank you
Topaz video AI 4.08 = hardly any difference upscaling/enhancing VHS video. In stark contrast to their still image upscaling which works amazingly well on all images.
I have a VHS tape that's nearly 40 years old and as good now as then.
is there a possible way to completely restore a distored vhs and get it cleared up with AI?
Not an easy and good one yet. All options are disappointing.
I’m struggling with this now. Trying to do home videos and not as an expert. Then watching RUclips videos and the experts tell you how it has to be done the right way. I want an easy way we’re I don’t have to be an engineer and take this on as a career change! Haha! I guess all I’m saying is that I want a good result (maybe not the best) with the least amount of learning curve to get this project done and move on with my life.
AI upscaling isn’t something that has a generic “good easy” option, it’s very much a “tinker and cater to the specifics you want” kind of things as a brand new emerging technology heh
I'd recommend just buying an old VHS+DVD recorder(just be aware that VHS+DVD *Players* are more common, but obviously no use for copying VHS tapes) unit (or a separate VCR and DVD recorder, though a bit less convenient), and you'll be able to capture tapes in good quality with very minimal effort. The machines were, after all, largely designed and mass produced for exactly this purpose.
For example, the video below was recorded on a similar device(a DVR, but is extremely similar to a normal DVD recorder) Note that youtube made it darker (and compresses the video quite a lot for low resolution video, though that's less of an issue):
ruclips.net/video/6Z6Aazu27no/видео.html
I did no processing on it, not even deinterlacing.
We need a good dust and scratch ai filter or app
Yes: Topaz lacks aspect ratio control. If I feed original DVD (VOB) it will do great job for aspect - it scale horizontal properly more than vertical. If I deinterlace 1st, than I loose aspect and Topaz does worse job; image is not in proper aspect and is those missing details from horizontal is not "reproduced". If I scale image when deinterlacing, Topaz does even worse. Image image is visible odd.
Someone gave that baby a giant piece of cake. Good on them.
Is it me, but I feel like most tapes look good on 4:3, I also like the aesthetic of it
How powerful does your computer have to be to upscale videos like this?
If I was good with up-scaling or tech savvy, I would upscale every seasons of "Tiny Toon Adventures" I wish...
The true choke on Ai is it only has as much info as we do. It just knows how to utilize it like a pro.
Such a strapping lad ☺️
Ranma and Transformers, my bro
should have tried gaia
VHS tapes should last for years if you store them in a cool, dark place.
"For years" Sure. We're now approaching 3-4 *decades* for many of them. Big difference.
Hear ya
Playing around with idea of buying $40 camcorder standard definition
See capture card to get on SIM card
Also see card that scales say to 1080p
You did not seem high on Ai solution
As good as cards
God I was so confused clicking on this video because I only looked at the thumbnail and thought this was a video by JackSepticEye...
He just caught the youtube-hair disease.
If i were to send you my tapes, would you also provide me a copy of the digital file, as well as my tapes back?
Ill pay for the shipping, both ways.
I'm not seeing anything impressive with this except a heavy hand on the denoise switch & a sharpening filter on a source that already has major ring artifacts. Also doesn't use a dropout fixer, has choppy motion & the footage is unstabilised despite the addition of resolution. I've seen better example 15 years ago without any AI marketing crap. I know real AI restoration is around the corner in 5 years but these Topaz & Davinci examples are very mediocre.
QTGMC is just a deinterlacer, it will do nothing on upscale quality. While bicubic upscaler is like the lowest on any upscaling method, second to bilinear. A good quality one is nnedi3, but don't expect magic on home VHS.
And the Topaz AI stuff are just scam
Savage black outline artifacts of VHS were not handled at all. They had one job.
Too bad Video Enhance crashes contasntly on the Nvidia RTX 3000 series.
try underclocking like 20 or 30%, then work your way up until it crashes again. I have an RX 480 and had the same issue
Why not use stabilization? To me it is more impressive than much of AI.
To upscale VHS videos makes absolutely 000 sense.
Also bicubic is the most primitive upscaling method.
It is made live by your media player or the tv set.
During digitalizing VHS importand is to remove interlacing.
The best devices offer field extension method.
It is the best method which could be made live, during capturing the video.
5$ digitalizing devices and its attached software offer only taking odd or even frame, which means descreasing original VHS quality by 50%.
Bizarre post dubbing of a presenter's voice.
Definitely not
Ay lot!
If you can do this with a rare 1994 VHS R-Rated movie The Raffle with Mark Hamill and Jennifer Clements
upscale AND preserve ? lol
Ai upscales of VHS tapes always look terrible.
Too many artifacts
Vhs movies more free whiching in your house now days everything proven