No it's not. We are watching a man using an, at this time 3 yr old whatever mac without CUDA and Tensorflow acceleration, reviewing his 10 hour upscaling attempts of 10 seconds of video. By the way this software matured massively since this video's upload. Compression settings are included, encoders, multiple new and "old" improved ai models were added. Which of course is not his fault reviewing wise. It's actually best when u have great 1080p footage downscale it to 720p via davinci studio (2x neural might do a trick here, too) and then upscale with topaz labs video enhance ai. For old footage I recommend doing multiple upscales on a video. So instead of 720p to 2160p u go 720p 2x > 1440p then downscale to 1080p 1080p 2x > 2160p U might want to do denoising first in editing software or use the function in topaz. 480p upscaling can be tricky, since that footage might as well be badly encoded. 480p crisp material on the other hand is no problem at all. A rule of thumb to mention: You want an ai to work on the smallest resolution possible. The smallest res without losing detail. You can do it on your Mac. 30 minute footage might cost u a year then. Or use this software after u read the system requirements... Cheers 😑
I think this kind of software will be huge in a couple of years. I get the feeling it's on an exponential scale of improvement in technology. Imagine a world where you could take films from the 1940s and upscale them to 4k in todays picture quality.
@@theoptimist4771 That's true of more recent films, however films over say 50 years old have a reducing limit to the quality that can be obtained from them due to film degradation over the years. Take a film like Double Indemnity from 1944, it's original prints will be 80 years old soon, and at that age it's probably not usable at all for any modern resolution as the physical film will be too aged to get anything from. On top of that many old films like that don't have much demand to be remastered or rereleased as most people today have never heard of them, so studios wouldn't want to bother with it. If the technology reaches a stage where smaller companies can digitally remaster old films like that using AI for fairly cheaply, then there's no limit to the lifespan and longevity of any film that's ever been made, and I think that's brilliant. Every film will basically be watchable and upscalable forever. I've got a pretty high education in film and media production so I've got a good understanding of how exiting this technology is for historic films.
I am SO looking forward to this! 🤩 Especially as someone who was into a very technologically-limited hobby around the turn of the millennium lol & the potential of not only being able to ~salvage fanvids in general, but not having to re-edit my own from "scratch" if ever there ends up a version of the tech where it can refer to actual 1080/4K source & use that info to upconvert / AI-upscale them.. what a dream! (& it's been over a decade since I lost the ability / tech / living situation to make vids, but I still do hope to get back into it ~someday..) 💻😅
I really enjoyed this video - I happen to have some old videos and photos I'd like to enhance and I don't see much difference with the trial version. I like your presentation - it's short, but very informative and delivered in a nice, calm manner.
This is an awesome approach, and your feedback is very professional to hear !!! Being in Video industry for 20+ years, I love the details that you shared, kudos to make much more such videos
This video really needs an update. I just downloaded the trial, and the interface has completely changed. You can now tell it what kind of input footage it is, and what you think is wrong with it (noise, compression artifacts, etc) and it will recommend one or more algorithms to apply. You can also tell it that your input footage is interlaced, and it will deinterlace it for you.
Thanks for doing this. I recently saw the ads and gave this software a try as I had a bunch of old SD footage that I was archiving. Luckily, the trial version has few limitations except for the watermark. I thought the software did make a difference, but not enough to justify the cost in time or money for what I was doing. If I had a paying job, where I needed to up res some footage, I’d give it another look. The nice thing about the trial version is you can try it on your footage, before you buy it. I think that the effectiveness of this software is very dependant on the footage you are working with.
Wow, this was a very professionally done video! I recently tried this program for upscaling rather low bitrate 720p gameplay to 4K. The results were very impressive considering I used youtube-compressed footage rather than the source file which I no longer have.
Thanks for the review. Hmmm, I am currently in the process of digitising the family videos, put them all on hard drives to give to my parents and brothers so we've all got a copy of the footage through the years. I am trying to find ways to improve the image quality. This looks like it might do it. I might hold off for a while just now until I've done them all. I'm hoping I'll see a deal on this or I'll see if a cheaper alternative for an amateur such as myself emerges.
What a quality video you made. The tests and questions I would ask. thanks. I intend to subscribe but... I was already subscribed, I wonder why CZcams doesn't push a quality youtuber to our feed. Thanks again.
This is a great video! really breaks down things and sadly, not much content on the web showing things in action other than product demo videos, which you can't always trust.
As someone who's sick of the BBC not even releasing a bluray of their classic dinosaur shows, Topaz VEAI has done such a better job than individually upscaling frame by frame with Topaz Gigapixel. The SD to HD upscale is clean and passable if you get the input settings right. If your result is too plastic, try selecting a higher input dataset. Even with the dreaded watermark, the difference is amazing. I will be purchasing this. On a side note, I think PAL will be better to upscale than NTSC because PAL SD has a higher resolution by 20%. This extra data will help a lot.
I've been doing some tests with this software recently, and I've noticed that I get much better results if I first process a video with preset 100% Scale Denoise/Deblock, as this cleans up videos a LOT, it's almost even more impressive than the upscale itself sometimes, and then process this result video a second time, this time with a 200% or more upscale. Upscaling the cleaned up video instead of the original one gives me a cleaner image, as things like noise or possible blocking artifacts aren't upscaled since they were reduced in the first pass.
@Dragan Petronijevic I had poor results with Artemis so I mostly don't use it at all. I've noticed that CG seems to be better for cleaning up very poor quality image, it reduces noise / blocking better, especially with the denoise/deblock preset. However, when using it to upscale, it smooth things a bit too much in my opinion, so Gaia HQ seems to be better for upscaling in the tests I did, it exagerates things a bit less, especially for faces and such, it's more natural. I either go with Gaia HQ for both denoise/deblock and upscaling, or with Gaia CG for denoise/deblock if the source quality is really bad, and then gaia HQ to upscale. Here is a test I did on a 1997 tv show in a not so great dvdrip quality, but it's the best source available. Source quality: i.imgur.com/yStEE6t.jpg After a first pass using Gaia CG with the denoise/deblock preset: i.imgur.com/a8Eq4BR.jpg And then after upscaling it to 300%: i.imgur.com/opVkryl.jpg I'm still experiencing with it, I wish it was a bit faster :p
Now this is exactly how a review should be. These days reviews are usually just advertisements. This video was perfectly balanced to my opinion. Thank you very much for your great work again. Maybe I'll buy this software one day to enhance my old VHS and Super VHS digitized footage.
Fantastic but I'm not sure what intrigued me more - Video Enhance AI or your review, script and narration! And if I'm not mistaken, this was before you attended Casey's class - well done.😎
We recently had many priceless 8mm reels converted to mp4 by a service. These films were from the 30's to the 50's. The conversions in many cases had significant noise in many scenes. The service did minimal post scan image processing. While the family was thrilled just be able to see and share them, I got intrigued with Video Enhance AI to see if it can improve them. It turns out VE AI does a great job in two areas. It removed most the noise with little degradation and exported the source frame by frame into a jpg, png or tiff for editing out defects. I processed the source mpg using A.I. Model Artimus-LQ, Preset 100% (denoise/deblock) and Model Output of both MP4 and JPG's (8bit compressed). I used the MP4 output file to review for defective frames (scratches, blotches, etc..). If unacceptable defects occur, I found the errant JPG frames in the subdirectory of JPGs created by VE AI. I then edited them with Affinity Photo frame by frame, (depending on the number of affected frames, this step could be labor intensive). Once finished I combined the JPGs back into a MP4 with ffmpeg: >> ffmpeg -r 16 -f image2 -i /%06d.jpg -s 640x480 /.mp4 -r = frame rate (16 for 8mm, chk yours with >> ffmpeg -i source.mp4 or ffprobe source.mp4 ) -f image2 = tells ffmpeg that the input file format is a sequence of images (ie. 000000.jpg, 000001.jpg, 000002.jpg) NOTE: The numbers must be sequential with no missing numbers or ffmpeg aborts. VE AI will export jpgs, tiffs and pngs in this sequence. -s HxW = height and width of output -i %06d.jpg = describes the sequence format (%06d is a six digit number with leading zeros followed by the file extension such as .jpg) There were no audio tracts with my films, but an extracted audio file can be integrated by adding another input to the command line. Run ffprobe to find your audio codec and frame rate of the source video to set the ffmpeg options. >> ffprobe source.mp4 >> ffmpeg -r -f image2 -i /%06d.jpg -i audio.aac -s HeightxWidth /.mp4 -i audio.aac = tells ffmpeg to use this audio file as input You can extract the audio with: >> ffmpeg -i input-video.avi -vn -acodec copy output-audio.aac where the file extension (aac) is determined by ffprobe of the audio codec. I'm not an ffmpeg expert, but the combination of Video Enhance AI with ffmpeg greatly enhanced these old films.
Wow... that's a comprehensive guide to get the most out of this. Very useful info here and thank you very much for sharing. it's appreciated! I'm glad it worked well for you Don.
Great video, I have been using an online platform to enhance my SD videos and have been super impressed with the results. Thinking about purchasing this, and I appreciate your even-handed result.
Thank you for the comment, much appreciated. If you're looking to upscale SD video's then the latest version of Enhance AI is significantly better than when I tested. Not just faster for people like me, with rubbish graphic cards, but nicer looking results too. They've given us more options. Look for the holiday pricing at the moment... it makes it a little easier to justify!
I find the Video Enhance AI very interesting, and I'm looking forward to seeing how it improves in the future! Also, thanks for showing me how to upscale 1080p to 4K in Final Cut Pro, I just tried it and it looks really nice!
Thanks for the great video; such a pleasure to watch! I've just bought a "proper" laptop to replace my 10 year old one, and looking forward to trying this out on very poor quality 1950s Sci-Fi TV shows. As others have said here and elsewhere, the software has improved a lot in the last year or two, so fingers crossed for some great results. I smiled at the suggestion that Topaz is ideal for someone with "too much money" . Rather, I think it's for people like me with too few skills. ;-)
The software has improved a lot of the last year... so much I made a new video using an M1 Mac.... wow , what a difference! czcams.com/video/IlKd3_N1fEs/video.html
Great video and I agree with all your points. Tried it out on my 50s and 60s rock and roll vhs films and though a bit sharper the resultant upscale looked more like a painting and somewhat unnatural. I think there is potential for old SD video but not quite there yet considering the time needed for rendering. The topaz promo videos looked immensely better than my own results.
I know what you mean with the painted look sometimes... did you try short clips with all the different settings? I found the Gaia-HQ was slightly less aggressive with the smoothing! I hope they are still developing this software because it definitely has potential. Thanks for the kind comment.
@@extrashot Thanks for the suggestion. I had tried it earlier and again yesterday and though I agree that it has a less painted look, the improvement is very minor. In fact it almost looked more blurry because of the smoothing out of the original film. In general it's almost like comparing a 65 and 70 inch tv. Side by side I can see a difference but on its own, not really. As much as I'm ready to buy the software, their one year of free updates included in the purchase makes me think that it might be better to wait and see what 2021 has to hold. I look forward to more of your videos. Regards from Canada.
There definitely seems to be noticeable improvement you managed to show. I guess the hope is it really is AI and with more "training" through maintenance and upgrades by Topaz Labs the results can improve. Thankfully I'm on Windows and my Nvidia card is relatively high end so rendering will hopefully be quicker than what you experienced. I think I'll give the trial a shot and purchase if it works well to scale up my old videos. Thanks for the review.
Thank you. You'll definitely get a faster result with a Nvidia card and it seems to work best on older videos. I need apple to make up with Nvidia! It's worth testing the trial first... I would have done that, but didn't think it would be fair on Topaz to then post a review here.
Great analysis, I got their advertisement on instagram and kinda curious about how it works. I googled and found your video, your video is so helpful for me to understand what this program all about, thanks!
This was a worthwhile test. Interesting to see the results you were able to obtain. As usual a useful examination of a particular tool! And yes, I go as far back as the umbilical cord. LOL!
I think you have clearly the best videotech channel i've seen! You have a great sense of humour. Your content is very deep. And your videos are full of small details (irony). I think (hope) it takes a lot of time to produce this videos! Otherwise I would feel myself incapable, because I needed a lot of time for this quality level of video making. :) I am curious, do you work with teleprompter?
Thank you... you're very kind. It does take too long... I need to perfect the art of producing videos in a more efficient way. It's an area I'm sadly not very good at but trying to improve! Not using a prompt at the moment, mainly because I'd then have to sit down and write a script !!! Thanks for your comment - it's appreciated.
many thanks for saving me some coin!! and the tip on how to do in editor. BTW though, for the stills shooters out there, have just tried the noise reduction software that TopazLabs has, and it is very very good!! used on Mavic stills, and results are fantastic
Loved this video. It's over a year since this video was posted, have you had more success with the Topaz software later? It would be interesting if you could make an updated video about this. Especially if you've upgraded your Mac (which doesn't seem to be a speed monster ... :D ) I just downloaded a trial and wanted to see others' experience. I found the interface PC looks somewhat different and has several interlaced modes now. So far I've been quite amazed by how good it works on some 1080p dashcam footage.
One thing this software does brilliantly is upscaling rendered 3d footage. I render at 960x540 and use the cg mode to upscale it to 1920x1080. The results are very good (not as good as rendering at 1080p, but very good nonetheless), and it saves me a lot of time of rendering. FYI, I have a i7 8700K with a 1050ti - each frame takes about 5 seconds to process - so with a faster graphics card it will definitely speed up your workflow even more.
You're the second person on here who's mentioned using it for upscaling CGI for time saving... I've just upgraded to the latest version and that seems a lot quicker even on an old Mac!
@@extrashot I bought the program while it was discounted for Black Friday. I haven't spent much time digging into it yet but thanks to you revealing how long the upscale time is I was prepared in knowing that it is not a fast process and I would have probably been much more impatient feeling like I should have forewarned had I not seen your review.
Great review and demonstration! I’m currently testing out the Gigapixel AI & Sharpen AI for photos and the results were amazing! Quick, easy, and automatic, but you can always override the manual tab to get even greater results. I forgot to mention Gigapixel Video AI the one you are using in this video is also amazing. I have a gaming laptop...a pretty decent beefy computer, but upscaling videos can be very time consuming. I believe if a video really needs it and you happen to really love that video or clip then by all means enhance them.
Thanks for the review, good info. I am a fan of a lot of the Topaz Labs photography products and was also looking into the Video Enhance AI product for some of my video projects, specifically for three use case scenarios: 1) I record and use a fair amount of 360 camera footage (Samsung Gear 360 before, and now the GoPro Max 360) and want to be able to enhance the 16:9 or 1:1 framed and exported clips from the 360 footage into existing 2K or 4K footage without it having a highly noticeable difference in resolution. 2) There are times when I record drone footage in 2K or 4K, but the actual framing I want to use is a cropped in view within the footage. This is usually due to instances where it would not be safe to fly and track a subject while having the drone that close to the either subject or the ground; or needing to make sure I keep the drone within signal range and still get a distant object in the shot. Being able to enhance these cropped in clips back to 2K or 4K will be a great boon. 3) High frame rate footage. Some of my gear has the option to shoot in higher frame rates of 60fps, 120fps or even 240fps. But NONE of it does so at the higher 2K/4K resolutions, you always have to drop down one or two categories of resolution in order to shoot at those high speed frame rates. When wanting to be able to slow down action by shooting at high frame rates, I want to see if using Enhance AI gives you the ability to get usable high resolution enhanced clips that can be mixed back in with the other 2K/4K clips.
That's a very interesting use... and I could see there might be useful improvements to be had. I have an Insta360 which shoots 5.7K but gives you a poor 1080 as a crop. It certainly would be interesting to see what improvements I can get from that footage... I'll run a few tests and try to post them here. Thank you so much for the suggestion. 👍
Very instructive video. Thank you, as someone interested in the subject without the professional knowledge, I really do appreciate your clear explanations. Thanks for the hard work !!!
The thing to remember about AI, vs a specific process driven improvement (resolve, final cut), is that AI is trained, and will get better as future updates are released. We've already seen this with their other AI tools. Their upscaling, noise reduction, JPEG->RAW conversion tools today produce better results than when they first launched. For example, the shot @ 4:35 should get better as the AI learns to detect context better. The trees aren't detected as trees particularly well, and therefore its processing, and introduction of AI detail isn't great. Expect that to get better. An alternative option would be to run your content through their standalone, picture-based AI upscaler. This can produce significantly better shots on content that doesn't have a huge amount of motion. E.g. your drone panning shots will definitely come out better - the downside is you need to convert your video to frames, process them in the tool, and re-assemble in to a video (or use a frame sequence if your NLE supports it) Secondly, your interlaced content, run it through a proper deinterlacer. I can recommend I can highly recommend QTGMC, since it's excellent at taking, for example, 25i content, and reproducing 50p. Since interlaced content is essentially double framerate encoded in to a single interlaced frame. Again, or low motion content this produces exceptional results, and it's free.
Good point about the interlaced... merging fields isn't good enough! Are you saying that the AI in the picture based scaler is more advanced than the Video Enhance program... is that why it does a better job or is it just the additional computational time in can take when looking at individual frames? The thought of the AI being constantly improved is exciting as it's definitely doing something more than just edge correction!
extrashot i watched a presentation some time ago that really summed up the power, or potential power of ai image enhancements. It went something like this. If you can look at an image and imagine how it should look then so can ai. For example we can watch old hand cranked silent movies and imagine fluid motion, clarity, and maybe even colour. Or watch old video games and see smooth polygon edges, or high colour depth.. Technically, the current training that’s going on works with the best quality footage available. 8k, shoot on the best cameras. As more and more of that is processed, and as higher quality (>8k) content becomes available the better ai processing will also become. Somewhat there’s a reliance on processing power to train ai on too; nvidia train its video game upscaling on super computers, that perhaps topaz can’t afford just yet? As for QTGMC, i actually haven’t tried using it for footage that is going to end up being ai scaled, but I’ve found QTGMC to personally be the best deinterlacer out there, and is generally very good at temporally interpolating to remove interlacing artifacts. Potentially in a high morion scene it might produce artifacts for ai upscaling, but i suspect it will still be better than not using it.
I agree with everyone complaining and concerned about the render times. However, it does some crazy voodoo and if you put in the time like this gentleman has, you'll eventually get amazing results. I've been using it for an attorney that I've been enlarging security cam footage for use in mediation and courtroom trials. Before I found this AP, I was just cleaning it up as best I could and enlarging it in AfterEffects. With Topaz Video Enhance I'm able to bring out a lot of detail that I couldn't before. My client has even credited my Topaz output files with winning a big money case. The best part is that I charge by the hour so slow rendering, for the first time in my career, is actually generating more money. Luckily, Lady Justice moves slowly so I haven't really felt pressured to miss a deadline because of the intense render times. One enhanced video that I set up to render overnight makes a nice little piece of change for my bank account. If I could land a few more attorneys for the same service, I could end up in a comfortably secure financial situation. Still, slow sux.
@@extrashot Most of the attorneys I know (and I know a lot.For some reason, my city and state are overflowing with attorneys) bill their clients for every tiny little thing they do for a client including asking their assistants to put the case file back in the filing cabinet. I'm sure some of them bill their clients if they happen to have a dream about the case in the middle of the night. I'm not working with defense lawyers or prosecutors. These guys are ambulance-chasing, Slip& Fall attorneys, and they're loaded. I've done this service for about five of them now and not one of them has so much as blinked when I give them a quote or an invoice. It's not their money anyway. They're going to add to their fee when they win $$$ money for a client. AND, they don't take cases they don't think they can win. I learned to stop selling myself short and began factoring in the skills that I spent 25 years developing.and that they're hiring me for. Speaking of selling yourself short, those of us who have actually put in the time and the work to get where we need to unite and warn the thousands of young designers and editors who are accepting ridiculously low prices to edit ridiculously complex projects that if they don't cut the crap, we're going to take them to an abandoned building and beat them senseless with a rubber hose. They're driving the value of our craft into the ground by editing people's podcasts or youtube shows for as low as $25/hour. The first production company I ever worked for charged $175/hour with a two-hour minimum. And those were 1991 dollars!
@@SCmattburrows How true... totally matches my experiences of how lawyers charge as well. They wouldn't last five minutes in an open or unregulated market... but then they are never going to let that happen! The point about how new people to our industry charge is much more complicated. I do understand why it's tempting to start out cheap and get a foot in the door... but all that happens is you undervalue your own skills for the whole of you career. It's almost worse if you do a good job because you'll undervalue everyone else who's trying to earn a living as well. Once people get used to getting your services for cheap... it's almost impossible to raise the rates. Eventually you'll never be able to support your family !!!
On this subject, the new model of NVidia Shield proposes a real time upscaling option that uses AI. It upscales 480p, 720p or 1080p videos to 4k. The result is interesting although it can look a bit plasticky sometime.
Oh my god, I've heard the word interlaced about 10 times in the past month, but didn't really understand what it was. I recently saw the first interlaced footage that I recognized as terrible looking, too. It had this weird blurring in the motion areas and it pissed me off. almost like TV bars in the motion of a clip. And you just taught me that that weird motion was called interlaced footage. I VERY strongly dislike it. UGh. Anyway, thanks for showing me your video. I really enjoyed watching it!
I'm old enough to remember when everything was interlaced.... unless you went to the movies. Dark days indeed! Thank you for the kind comment, it's appreciated.
@Planet Xtreme, haha, this comment made me feel old in a second. Welcome to the past world of strange NLE products like "DPS PAR" or "Matrox Digisuite", where interlace or not was vital for smooth movements, but also flickering typography on screen. Thats of course on phosphorized scanline monitors only, with heavy tubes and double screen refresh. With half the y-resolution, by skipping every second line, and double the frame rate (50/60Hz, PAL/NTCS) it was made sure that either all odd lines or all even lines of a full frame were visible at the same time. So unlike those very visible bars nowadays in real interlace times those bars were not visible onscreen while the clips were running but provided a smoother playback.
Great presentation. I have both FCPX and DaVinci Resolve Studio and I was wondering what process did you use with Resolve to enhance 1080P footage to 4K. If you have an existing tutorial on upscaling SD or 1080P footage with DaVinci Resolve Studio, I would be very interested to see it.
@@lucasleal4035 an i7 5500u only has two cores with hyperthreading. It is a fairly weak processor, especially for something like video editing which tend to utilise more cores. Mobile processors also generally have lower frequencies and thus lower performance over desktop computers.
Thanks - I've seen those damn adds constantly and assumed it was rubbish. Appreciate how you took the time to use FCPX to replicate it - made the overall review that much better. What is the bg music you've used at the end of video?
Thanks Chris, good news is it's definitely not rubbish, it's doing something! I've just downloaded the latest update and it's also improved quite a bit with more picture quality options... also a fair bit faster on a Mac as well. The music was my own mix from the Filmstro service, which I used to use before switching to Artlist. Unfortunately, because I no longer subscribe, I don't have access to it any more (which is one of the reasons I switched).
I have been using this for a while and for anyone interested in the time to do the project element I have come across this. I have an older HP pc which is in any terms fast when I first used the program with the pc as is a 5 min clip took 4.5 days to convert original footage 720 x 576 taking it up 200% I changed the processor to an AMD A10 5800K with only a slight improvement (but still happy) I then changed the graphics card to a RTX750ti (second hand 60 euros) and what a difference 4.5 days down to around 4 hours. the latest version of Topaz does have interlacing De interlacing function which is superb. I ran some DV cam footage and again upscaled 200% the end result was very good no jagged edges at all and much better quality. I am currently saving to get a gaming computer for flight sim software but cant wait to see what the AI video does on that machine as the graphics ram goes from 2gb to 10 gb. This software needs a good graphics card ... there is an option within the software to use cpu or gpu if you have a gpu run it through that.
Did you mean GTX 750 TI instead of RTX 750 TI? Because trust me, if an RTX had 2 GB of video memory, it would be a terrible choice. Most RTX's have well above 4 GB of video memory... Just sayin!
lmao that zoom and enhance cliches from the films.. We'll get there eventually, its hard to fill in detail you was unable to capture because tech wasn't good enough or you just recorded it with a potato. Its a good start though plenty of people will have really old crappy clips left around from like old mobile phones or something so this might be good to improve that. Nice video dude been looking at this software for ages but its a bit pricey for me at the moment
They often run deals on the price, but it's worth testing on a trial version first because you need a nice fast graphics card to speed things up a bit. Oh... and I normally use a potato mark iii ....
You should check the program again, the improvements they have made are quite interesting. There are almost 20 different AI rendering models now as opposed to the three you saw.
Watching your example enhance ai seems to do something a little more than sharpening alone. I think this technology just needs a few more years before it’s in every editor and we can all call it “enhance”
It's definitely doing something more than I can do with FCPX or Resolve alone... works well with poor quality footage but is quite subtle once the rushes are good. Cheers for the comment.... and I love your videos by the way... keep them coming!
Really hoping that eventually, you'll be able to give this (or a similar program) a few still shots from the old locations (and people) and it can incorporate that into its learning. There are plenty of old TV shows and movies out there in SD, but with production photography at much higher res available.
Super-resolution AI models are already trained with many millions or even billions of samples. Retraining it can take days, and selecting the right learning parameters is not a simple process. Only thing that will make it better are new architectures, which thankfully pop up every year or so, with increasing frequency.
Hello. If I render a video from adobe premiere with high quality with best render settings and filmed with professional camera,lenses and lighting. it the Video Enhance AI can make a better difference for the video? it can boost quality of video or not?
you don't have the render the videos fully to try each upscale AI setting, you can actually preview the upscale setting with the preview button on the upper right hand corner of the app which will generate a short preview.
Another interesting, well put together review! And ‘this time of some magic software... enhance it. Personally I thought it might be fun and “quick” to try and enhance some old progressive 3CCD footage from the 90’s (Sony trv900, i think?). Fun stuff and thanks for the thoughtful, in depth review. Now back to my FX9 👍
Thank you for your testing and insights! I'd downloaded the trial last night. This was quite a help. One minor quibble: You say the VE setting to upscale good HD to 4K is Gaia-HQ, but your textbox over the footage still says Gaia-CG. So which did you use? Also, did you try any of the image sequence settings or just the mp4? Thanks again!
Ahh, I see what you mean.... yes the graphic is correct to what you're seeing. My words were recorded later and I'd re-run some of the tests. Sorry for the confusion. I didn't try outputting stills and then compiling into video later... so all mp4. The stills route probably would be better for anything you want to edit later as you can then choose the compression settings... especially with .tiff's
This was a interesting video. I too seen this software advertised and wondered if it was worth anything. By watching your findings it seems that it is worth getting. Especially if you are in a tight spot and need to deliver a project that is in HD and want 4K. Also if you are editing and ready to deliver a 4K project and you notice some clips you have are in HD and want them upscaled to 4K for the clarity to show the clips in your 4K project, then i can see the worth. Anyway I probably won't buy it but I can see where someone may be interested in having this software.
When I started making this video I thought I'd have the chance to take the mickey out of their exaggerated claims, so seeing how well it worked, was a genuine surprise!
main problem is with people's faces. They just don't handle faces properly. It is a bit like turning up the sharpness on regular video of buildings or trees, you can get away with a lot of sharpening, but on a face you can't. It is a bit strange as if it's really AI it should recognise faces and handle them in a different way.
Enhance 4k to 8k
Processor: This little maneuver is going to cost us 51 years
Ha!!!... you'd need something fast... with a few fast Nvidia cards too !!!
Excellent review. Seems the primary benefit will be for modernizing old footage.
True... definitely bigger improvements with older footage! 👍
We are already having benefits in modern games with DLSS
No it's not. We are watching a man using an, at this time 3 yr old whatever mac without CUDA and Tensorflow acceleration, reviewing his 10 hour upscaling attempts of 10 seconds of video.
By the way this software matured massively since this video's upload. Compression settings are included, encoders, multiple new and "old" improved ai models were added. Which of course is not his fault reviewing wise.
It's actually best when u have great 1080p footage downscale it to 720p via davinci studio (2x neural might do a trick here, too) and then upscale with topaz labs video enhance ai.
For old footage I recommend doing multiple upscales on a video.
So instead of 720p to 2160p u go
720p 2x > 1440p then downscale to 1080p
1080p 2x > 2160p
U might want to do denoising first in editing software or use the function in topaz.
480p upscaling can be tricky, since that footage might as well be badly encoded. 480p crisp material on the other hand is no problem at all.
A rule of thumb to mention: You want an ai to work on the smallest resolution possible. The smallest res without losing detail.
You can do it on your Mac. 30 minute footage might cost u a year then. Or use this software after u read the system requirements... Cheers 😑
new versions can work with interlaced shots now, and also way more options
There you go!... czcams.com/video/IlKd3_N1fEs/video.html
Me watching this video in 360p: 🤡🤡🤡
😂😂😂
Enhance it lol
I did the same with 480p. Realized after reading your comment. Thanks bro. I've officially joined the Circus.
@@yoeymeme lol
I think this kind of software will be huge in a couple of years. I get the feeling it's on an exponential scale of improvement in technology. Imagine a world where you could take films from the 1940s and upscale them to 4k in todays picture quality.
They can already be remastered up to 8K since film stock is an analog medium and thus has no defined resolution like digital
@@theoptimist4771 That's true of more recent films, however films over say 50 years old have a reducing limit to the quality that can be obtained from them due to film degradation over the years. Take a film like Double Indemnity from 1944, it's original prints will be 80 years old soon, and at that age it's probably not usable at all for any modern resolution as the physical film will be too aged to get anything from. On top of that many old films like that don't have much demand to be remastered or rereleased as most people today have never heard of them, so studios wouldn't want to bother with it. If the technology reaches a stage where smaller companies can digitally remaster old films like that using AI for fairly cheaply, then there's no limit to the lifespan and longevity of any film that's ever been made, and I think that's brilliant. Every film will basically be watchable and upscalable forever. I've got a pretty high education in film and media production so I've got a good understanding of how exiting this technology is for historic films.
It will be awesome
I am SO looking forward to this! 🤩 Especially as someone who was into a very technologically-limited hobby around the turn of the millennium lol & the potential of not only being able to ~salvage fanvids in general, but not having to re-edit my own from "scratch" if ever there ends up a version of the tech where it can refer to actual 1080/4K source & use that info to upconvert / AI-upscale them.. what a dream! (& it's been over a decade since I lost the ability / tech / living situation to make vids, but I still do hope to get back into it ~someday..) 💻😅
@@theoptimist4771 czcams.com/video/9R3VxPz1aE0/video.html
Great demonstration from a professional with no nonsense presentation, glad i found this channel, subscribed and liked.
Thank you. I'm glad you're here now and the sub is appreciated!
I really enjoyed this video - I happen to have some old videos and photos I'd like to enhance and I don't see much difference with the trial version. I like your presentation - it's short, but very informative and delivered in a nice, calm manner.
This is an awesome approach, and your feedback is very professional to hear !!! Being in Video industry for 20+ years, I love the details that you shared, kudos to make much more such videos
Great voice. This isn't the sort of thing I'd watch normally. You made it engaging!
Thank you!
Your review was very helpful. Thanks! You also brought back memories of being tethered between a Nagra and an Eclair NPR!
Fantastic as usual!.. Born to do it!.. Thanks for yet another great video!
precisely the sort of testing and commentary I wanted from this program. Thank you
Thank you for the comment... it's appreciated.
This video really needs an update. I just downloaded the trial, and the interface has completely changed. You can now tell it what kind of input footage it is, and what you think is wrong with it (noise, compression artifacts, etc) and it will recommend one or more algorithms to apply. You can also tell it that your input footage is interlaced, and it will deinterlace it for you.
It certainly deserves another look with the latest version and a new Mac.... hopefully soon.
Thanks for doing this. I recently saw the ads and gave this software a try as I had a bunch of old SD footage that I was archiving. Luckily, the trial version has few limitations except for the watermark. I thought the software did make a difference, but not enough to justify the cost in time or money for what I was doing. If I had a paying job, where I needed to up res some footage, I’d give it another look. The nice thing about the trial version is you can try it on your footage, before you buy it. I think that the effectiveness of this software is very dependant on the footage you are working with.
Wow, this was a very professionally done video! I recently tried this program for upscaling rather low bitrate 720p gameplay to 4K. The results were very impressive considering I used youtube-compressed footage rather than the source file which I no longer have.
Thank you for the comment... and you're right to start with a low bitrate, that's where the big improvements are!
Thanks for the review. Hmmm, I am currently in the process of digitising the family videos, put them all on hard drives to give to my parents and brothers so we've all got a copy of the footage through the years. I am trying to find ways to improve the image quality. This looks like it might do it. I might hold off for a while just now until I've done them all. I'm hoping I'll see a deal on this or I'll see if a cheaper alternative for an amateur such as myself emerges.
What a quality video you made. The tests and questions I would ask. thanks.
I intend to subscribe but... I was already subscribed, I wonder why CZcams doesn't push a quality youtuber to our feed. Thanks again.
Need to look at that... but thank you for the sub, very kind!
This is a great video! really breaks down things and sadly, not much content on the web showing things in action other than product demo videos, which you can't always trust.
Glad it was helpful!
As someone who's sick of the BBC not even releasing a bluray of their classic dinosaur shows, Topaz VEAI has done such a better job than individually upscaling frame by frame with Topaz Gigapixel. The SD to HD upscale is clean and passable if you get the input settings right. If your result is too plastic, try selecting a higher input dataset. Even with the dreaded watermark, the difference is amazing. I will be purchasing this.
On a side note, I think PAL will be better to upscale than NTSC because PAL SD has a higher resolution by 20%. This extra data will help a lot.
What a brilliant job you did making this piece. Broadcast quality material on all fronts, well done.
Thanks James... kind of you!
I've been doing some tests with this software recently, and I've noticed that I get much better results if I first process a video with preset 100% Scale Denoise/Deblock, as this cleans up videos a LOT, it's almost even more impressive than the upscale itself sometimes, and then process this result video a second time, this time with a 200% or more upscale.
Upscaling the cleaned up video instead of the original one gives me a cleaner image, as things like noise or possible blocking artifacts aren't upscaled since they were reduced in the first pass.
That's a good idea... makes sense. I'll try it, and thank you for sharing
@Dragan Petronijevic I had poor results with Artemis so I mostly don't use it at all.
I've noticed that CG seems to be better for cleaning up very poor quality image, it reduces noise / blocking better, especially with the denoise/deblock preset.
However, when using it to upscale, it smooth things a bit too much in my opinion, so Gaia HQ seems to be better for upscaling in the tests I did, it exagerates things a bit less, especially for faces and such, it's more natural.
I either go with Gaia HQ for both denoise/deblock and upscaling, or with Gaia CG for denoise/deblock if the source quality is really bad, and then gaia HQ to upscale.
Here is a test I did on a 1997 tv show in a not so great dvdrip quality, but it's the best source available.
Source quality:
i.imgur.com/yStEE6t.jpg
After a first pass using Gaia CG with the denoise/deblock preset:
i.imgur.com/a8Eq4BR.jpg
And then after upscaling it to 300%:
i.imgur.com/opVkryl.jpg
I'm still experiencing with it, I wish it was a bit faster :p
@@Larxia The Upscaling settings please.
Bookmark
@@Larxia I want to try some TV shows like Stargate SG1
Love your content and your playful/calming disposition.
Cheers Perry... calm on the outside!
@@extrashot 😂🤯
Now this is exactly how a review should be. These days reviews are usually just advertisements. This video was perfectly balanced to my opinion. Thank you very much for your great work again. Maybe I'll buy this software one day to enhance my old VHS and Super VHS digitized footage.
Thank you computerjantje... very kind and appreciated!
This is just the video i needed; answering all my questions and taking not much time. Thanks mate!
No worries, and thank you for the comment. Appreciated!
Fantastic but I'm not sure what intrigued me more - Video Enhance AI or your review, script and narration! And if I'm not mistaken, this was before you attended Casey's class - well done.😎
Clear explanation and I really like your soothing voice, please keep making content.
Wow you really had a lot of work to create this video, thank you so much it made many things clear to me! Amazing quality video!
Thank you. Appreciated!
We recently had many priceless 8mm reels converted to mp4 by a service. These films were from the 30's to the 50's.
The conversions in many cases had significant noise in many scenes. The service did minimal post scan image processing. While the family was thrilled just be able to see and share them, I got intrigued with Video Enhance AI to see if it can improve them.
It turns out VE AI does a great job in two areas. It removed most the noise with little degradation and exported the source frame by frame into a jpg, png or tiff for editing out defects.
I processed the source mpg using A.I. Model Artimus-LQ, Preset 100% (denoise/deblock) and Model Output of both MP4 and JPG's (8bit compressed). I used the MP4 output file to review for defective frames (scratches, blotches, etc..). If unacceptable defects occur, I found the errant JPG frames in the subdirectory of JPGs created by VE AI. I then edited them with Affinity Photo frame by frame, (depending on the number of affected frames, this step could be labor intensive).
Once finished I combined the JPGs back into a MP4 with ffmpeg:
>> ffmpeg -r 16 -f image2 -i /%06d.jpg -s 640x480 /.mp4
-r = frame rate (16 for 8mm, chk yours with >> ffmpeg -i source.mp4 or ffprobe source.mp4 )
-f image2 = tells ffmpeg that the input file format is a sequence of images (ie. 000000.jpg, 000001.jpg, 000002.jpg)
NOTE: The numbers must be sequential with no missing numbers or ffmpeg aborts.
VE AI will export jpgs, tiffs and pngs in this sequence.
-s HxW = height and width of output
-i %06d.jpg = describes the sequence format (%06d is a six digit number with leading zeros followed by the file extension such as .jpg)
There were no audio tracts with my films, but an extracted audio file can be integrated by adding another input to the command line.
Run ffprobe to find your audio codec and frame rate of the source video to set the ffmpeg options.
>> ffprobe source.mp4
>> ffmpeg -r -f image2 -i /%06d.jpg -i audio.aac -s HeightxWidth /.mp4
-i audio.aac = tells ffmpeg to use this audio file as input
You can extract the audio with:
>> ffmpeg -i input-video.avi -vn -acodec copy output-audio.aac
where the file extension (aac) is determined by ffprobe of the audio codec.
I'm not an ffmpeg expert, but the combination of Video Enhance AI with ffmpeg greatly enhanced these old films.
Wow... that's a comprehensive guide to get the most out of this.
Very useful info here and thank you very much for sharing. it's appreciated!
I'm glad it worked well for you Don.
Pocket 6K footage exported at 8K via Topaz yields a drastically better image than the overheating Canon R5 :)
I'm slightly biased... but I think the BMPCC 6K does anyway !!!
You can get same results with open source avisynth
Great video, I have been using an online platform to enhance my SD videos and have been super impressed with the results. Thinking about purchasing this, and I appreciate your even-handed result.
Thank you for the comment, much appreciated. If you're looking to upscale SD video's then the latest version of Enhance AI is significantly better than when I tested. Not just faster for people like me, with rubbish graphic cards, but nicer looking results too. They've given us more options. Look for the holiday pricing at the moment... it makes it a little easier to justify!
What a great review/analysis! Subbed!
Thank you for the kind comment... and the sub is also appreciated!
I find the Video Enhance AI very interesting, and I'm looking forward to seeing how it improves in the future!
Also, thanks for showing me how to upscale 1080p to 4K in Final Cut Pro, I just tried it and it looks really nice!
Thanks for the kind comment... happy I saved you money while we wait for the future improvements!
Thanks for the great video; such a pleasure to watch! I've just bought a "proper" laptop to replace my 10 year old one, and looking forward to trying this out on very poor quality 1950s Sci-Fi TV shows. As others have said here and elsewhere, the software has improved a lot in the last year or two, so fingers crossed for some great results. I smiled at the suggestion that Topaz is ideal for someone with "too much money" . Rather, I think it's for people like me with too few skills. ;-)
The software has improved a lot of the last year... so much I made a new video using an M1 Mac.... wow , what a difference! czcams.com/video/IlKd3_N1fEs/video.html
Great video and I agree with all your points. Tried it out on my 50s and 60s rock and roll vhs films and though a bit sharper the resultant upscale looked more like a painting and somewhat unnatural. I think there is potential for old SD video but not quite there yet considering the time needed for rendering. The topaz promo videos looked immensely better than my own results.
I know what you mean with the painted look sometimes... did you try short clips with all the different settings? I found the Gaia-HQ was slightly less aggressive with the smoothing!
I hope they are still developing this software because it definitely has potential. Thanks for the kind comment.
@@extrashot Thanks for the suggestion. I had tried it earlier and again yesterday and though I agree that it has a less painted look, the improvement is very minor. In fact it almost looked more blurry because of the smoothing out of the original film. In general it's almost like comparing a 65 and 70 inch tv. Side by side I can see a difference but on its own, not really. As much as I'm ready to buy the software, their one year of free updates included in the purchase makes me think that it might be better to wait and see what 2021 has to hold. I look forward to more of your videos. Regards from Canada.
There definitely seems to be noticeable improvement you managed to show. I guess the hope is it really is AI and with more "training" through maintenance and upgrades by Topaz Labs the results can improve. Thankfully I'm on Windows and my Nvidia card is relatively high end so rendering will hopefully be quicker than what you experienced. I think I'll give the trial a shot and purchase if it works well to scale up my old videos.
Thanks for the review.
Thank you. You'll definitely get a faster result with a Nvidia card and it seems to work best on older videos. I need apple to make up with Nvidia! It's worth testing the trial first... I would have done that, but didn't think it would be fair on Topaz to then post a review here.
I love your videos, always informative 👍🏻
Thank you... very kind.
Great analysis, I got their advertisement on instagram and kinda curious about how it works. I googled and found your video, your video is so helpful for me to understand what this program all about, thanks!
Glad it helped... and thank you for the comment.
What a nice review, good job.
Awesome video. I had wanted to run similiar tests. Grateful for your comprehensive review. Thankyou!
Thanks Tyler, very kind!
One of the very best produced reviews I have ever seen. Excellent work. Many thanks.
You'll make me big headed.... but thank you, it's appreciated!
I need some work on this page plz open this link crackother.com/wp-admin/edit.php
This was a worthwhile test. Interesting to see the results you were able to obtain. As usual a useful examination of a particular tool! And yes, I go as far back as the umbilical cord. LOL!
Thank you for the kind words David... appreciated, (but I bet my cord's longer!!!)
I think you have clearly the best videotech channel i've seen! You have a great sense of humour. Your content is very deep. And your videos are full of small details (irony). I think (hope) it takes a lot of time to produce this videos! Otherwise I would feel myself incapable, because I needed a lot of time for this quality level of video making. :) I am curious, do you work with teleprompter?
Thank you... you're very kind. It does take too long... I need to perfect the art of producing videos in a more efficient way. It's an area I'm sadly not very good at but trying to improve! Not using a prompt at the moment, mainly because I'd then have to sit down and write a script !!!
Thanks for your comment - it's appreciated.
many thanks for saving me some coin!! and the tip on how to do in editor. BTW though, for the stills shooters out there, have just tried the noise reduction software that TopazLabs has, and it is very very good!! used on Mavic stills, and results are fantastic
Thank you... and yes, I've heard others saying the noise reduction routines are great!
Very informative. How I do hope all reviews on software be like this.
Loved this video. It's over a year since this video was posted, have you had more success with the Topaz software later? It would be interesting if you could make an updated video about this. Especially if you've upgraded your Mac (which doesn't seem to be a speed monster ... :D )
I just downloaded a trial and wanted to see others' experience. I found the interface PC looks somewhat different and has several interlaced modes now.
So far I've been quite amazed by how good it works on some 1080p dashcam footage.
I've been thinking about updating this to cover the latest versions... seems a lot better!
One thing this software does brilliantly is upscaling rendered 3d footage. I render at 960x540 and use the cg mode to upscale it to 1920x1080. The results are very good (not as good as rendering at 1080p, but very good nonetheless), and it saves me a lot of time of rendering. FYI, I have a i7 8700K with a 1050ti - each frame takes about 5 seconds to process - so with a faster graphics card it will definitely speed up your workflow even more.
You're the second person on here who's mentioned using it for upscaling CGI for time saving... I've just upgraded to the latest version and that seems a lot quicker even on an old Mac!
Thanks for the review, now if know exactly what to expect if I get this :)
Very good review! Davinci Resolve upscaler hasn't give so good results, but this seems promising!
omg, very detail and impressive review. i am impressed
Thank you... very kind!
8:50 Such a satisfying looking set you've got for your video background
Subscribed just because you really know what you are doing. :-)
Thank you... too kind!
Very well conducted review. Appreciate it.
Thanks Leo... very kind!
I ran a less than standard def version of Roger Rabbit through it and again... WOW...
thank u i think ur the first that explains about upscaling
Thanks Alex.
Thanks for such an informative review! It's nice to get a sincere honest insightful look into a product before spending a good chunk of money.
... and thank you for the kind comment.
I do need to take another look at this program. The updates since I made this video are huge.
@@extrashot I bought the program while it was discounted for Black Friday. I haven't spent much time digging into it yet but thanks to you revealing how long the upscale time is I was prepared in knowing that it is not a fast process and I would have probably been much more impatient feeling like I should have forewarned had I not seen your review.
@@extrashot how can use free without paying is it possible in adope primer proo
Great review and demonstration! I’m currently testing out the Gigapixel AI & Sharpen AI for photos and the results were amazing! Quick, easy, and automatic, but you can always override the manual tab to get even greater results. I forgot to mention Gigapixel Video AI the one you are using in this video is also amazing. I have a gaming laptop...a pretty decent beefy computer, but upscaling videos can be very time consuming. I believe if a video really needs it and you happen to really love that video or clip then by all means enhance them.
Thank you. The results from SD to HD video are amazing.... and much quicker if you have an Nvidia GPU
I came to see review of Video Enhance software, and I was quite surprised to see a lot of nice shots of Dubrovnik :-)
Spotted... taken some time ago. One of my favourite places!
Great professional presentation. Subscribed!
Thank you... it's appreciated. Have a great New Year!
@@extrashot A great New Year to you too!
Thanks for the review, good info. I am a fan of a lot of the Topaz Labs photography products and was also looking into the Video Enhance AI product for some of my video projects, specifically for three use case scenarios:
1) I record and use a fair amount of 360 camera footage (Samsung Gear 360 before, and now the GoPro Max 360) and want to be able to enhance the 16:9 or 1:1 framed and exported clips from the 360 footage into existing 2K or 4K footage without it having a highly noticeable difference in resolution.
2) There are times when I record drone footage in 2K or 4K, but the actual framing I want to use is a cropped in view within the footage. This is usually due to instances where it would not be safe to fly and track a subject while having the drone that close to the either subject or the ground; or needing to make sure I keep the drone within signal range and still get a distant object in the shot. Being able to enhance these cropped in clips back to 2K or 4K will be a great boon.
3) High frame rate footage. Some of my gear has the option to shoot in higher frame rates of 60fps, 120fps or even 240fps. But NONE of it does so at the higher 2K/4K resolutions, you always have to drop down one or two categories of resolution in order to shoot at those high speed frame rates. When wanting to be able to slow down action by shooting at high frame rates, I want to see if using Enhance AI gives you the ability to get usable high resolution enhanced clips that can be mixed back in with the other 2K/4K clips.
That's a very interesting use... and I could see there might be useful improvements to be had. I have an Insta360 which shoots 5.7K but gives you a poor 1080 as a crop. It certainly would be interesting to see what improvements I can get from that footage... I'll run a few tests and try to post them here. Thank you so much for the suggestion. 👍
Thanks for the super useful review!
This is a really great video. Awesome!
Thanks Stephen, very kind!
Very instructive video. Thank you, as someone interested in the subject without the professional knowledge, I really do appreciate your clear explanations. Thanks for the hard work !!!
Thank you Teva, very kind and appreciated!
It so lovely watching you! Such a kind face :)
Too kind! 🥴
I need some work on this page plz open this link crackother.com/wp-admin/edit.php
Thank you for this review, much appreciated.
Thanks Daniel... very kind!
Great review, thanks!
Thanks for watching!
You are excellent at what you do, thank you for the video.
Cheers, very kind of you to say so....
great video. thank you so much for all the footage and doing all the hard work for us so we don't have to. Thank you
You're welcome!
The thing to remember about AI, vs a specific process driven improvement (resolve, final cut), is that AI is trained, and will get better as future updates are released. We've already seen this with their other AI tools. Their upscaling, noise reduction, JPEG->RAW conversion tools today produce better results than when they first launched. For example, the shot @ 4:35 should get better as the AI learns to detect context better. The trees aren't detected as trees particularly well, and therefore its processing, and introduction of AI detail isn't great. Expect that to get better. An alternative option would be to run your content through their standalone, picture-based AI upscaler. This can produce significantly better shots on content that doesn't have a huge amount of motion. E.g. your drone panning shots will definitely come out better - the downside is you need to convert your video to frames, process them in the tool, and re-assemble in to a video (or use a frame sequence if your NLE supports it)
Secondly, your interlaced content, run it through a proper deinterlacer. I can recommend I can highly recommend QTGMC, since it's excellent at taking, for example, 25i content, and reproducing 50p. Since interlaced content is essentially double framerate encoded in to a single interlaced frame. Again, or low motion content this produces exceptional results, and it's free.
Good point about the interlaced... merging fields isn't good enough! Are you saying that the AI in the picture based scaler is more advanced than the Video Enhance program... is that why it does a better job or is it just the additional computational time in can take when looking at individual frames? The thought of the AI being constantly improved is exciting as it's definitely doing something more than just edge correction!
extrashot i watched a presentation some time ago that really summed up the power, or potential power of ai image enhancements. It went something like this.
If you can look at an image and imagine how it should look then so can ai.
For example we can watch old hand cranked silent movies and imagine fluid motion, clarity, and maybe even colour. Or watch old video games and see smooth polygon edges, or high colour depth..
Technically, the current training that’s going on works with the best quality footage available. 8k, shoot on the best cameras. As more and more of that is processed, and as higher quality (>8k) content becomes available the better ai processing will also become. Somewhat there’s a reliance on processing power to train ai on too; nvidia train its video game upscaling on super computers, that perhaps topaz can’t afford just yet?
As for QTGMC, i actually haven’t tried using it for footage that is going to end up being ai scaled, but I’ve found QTGMC to personally be the best deinterlacer out there, and is generally very good at temporally interpolating to remove interlacing artifacts. Potentially in a high morion scene it might produce artifacts for ai upscaling, but i suspect it will still be better than not using it.
@@sheppo I will take a look at QTGMC... thank you, very interesting stuff and excited to see which way AI goes in video creation!
Great video. Very impressive and just what I needed to see.
Glad it helped... and they've improved it a lot in recent versions!
This is actually useful thanks!!
...and thank you!
I agree with everyone complaining and concerned about the render times. However, it does some crazy voodoo and if you put in the time like this gentleman has, you'll eventually get amazing results. I've been using it for an attorney that I've been enlarging security cam footage for use in mediation and courtroom trials. Before I found this AP, I was just cleaning it up as best I could and enlarging it in AfterEffects. With Topaz Video Enhance I'm able to bring out a lot of detail that I couldn't before. My client has even credited my Topaz output files with winning a big money case. The best part is that I charge by the hour so slow rendering, for the first time in my career, is actually generating more money. Luckily, Lady Justice moves slowly so I haven't really felt pressured to miss a deadline because of the intense render times. One enhanced video that I set up to render overnight makes a nice little piece of change for my bank account. If I could land a few more attorneys for the same service, I could end up in a comfortably secure financial situation. Still, slow sux.
Are you sure you should be telling the Attorney's how the charging works? I admire your productive use of the tech though.... nice one! 👍😀
@@extrashot Most of the attorneys I know (and I know a lot.For some reason, my city and state are overflowing with attorneys) bill their clients for every tiny little thing they do for a client including asking their assistants to put the case file back in the filing cabinet. I'm sure some of them bill their clients if they happen to have a dream about the case in the middle of the night. I'm not working with defense lawyers or prosecutors. These guys are ambulance-chasing, Slip& Fall attorneys, and they're loaded. I've done this service for about five of them now and not one of them has so much as blinked when I give them a quote or an invoice. It's not their money anyway. They're going to add to their fee when they win $$$ money for a client. AND, they don't take cases they don't think they can win. I learned to stop selling myself short and began factoring in the skills that I spent 25 years developing.and that they're hiring me for. Speaking of selling yourself short, those of us who have actually put in the time and the work to get where we need to unite and warn the thousands of young designers and editors who are accepting ridiculously low prices to edit ridiculously complex projects that if they don't cut the crap, we're going to take them to an abandoned building and beat them senseless with a rubber hose. They're driving the value of our craft into the ground by editing people's podcasts or youtube shows for as low as $25/hour. The first production company I ever worked for charged $175/hour with a two-hour minimum. And those were 1991 dollars!
@@SCmattburrows How true... totally matches my experiences of how lawyers charge as well. They wouldn't last five minutes in an open or unregulated market... but then they are never going to let that happen! The point about how new people to our industry charge is much more complicated. I do understand why it's tempting to start out cheap and get a foot in the door... but all that happens is you undervalue your own skills for the whole of you career. It's almost worse if you do a good job because you'll undervalue everyone else who's trying to earn a living as well. Once people get used to getting your services for cheap... it's almost impossible to raise the rates. Eventually you'll never be able to support your family !!!
I need some work on this page plz open it crackother.com/wp-admin/edit.php
Underrated Channel
Of course, I agree!
On this subject, the new model of NVidia Shield proposes a real time upscaling option that uses AI. It upscales 480p, 720p or 1080p videos to 4k. The result is interesting although it can look a bit plasticky sometime.
That's a lot to do in real time... like to have a look at that. This tech is moving fast now...
This was so fun to watch!
Thank you Doda... glad you found my channel!
Oh my god, I've heard the word interlaced about 10 times in the past month, but didn't really understand what it was. I recently saw the first interlaced footage that I recognized as terrible looking, too. It had this weird blurring in the motion areas and it pissed me off. almost like TV bars in the motion of a clip.
And you just taught me that that weird motion was called interlaced footage. I VERY strongly dislike it. UGh.
Anyway, thanks for showing me your video. I really enjoyed watching it!
I'm old enough to remember when everything was interlaced.... unless you went to the movies. Dark days indeed! Thank you for the kind comment, it's appreciated.
@Planet Xtreme, haha, this comment made me feel old in a second. Welcome to the past world of strange NLE products like "DPS PAR" or "Matrox Digisuite", where interlace or not was vital for smooth movements, but also flickering typography on screen. Thats of course on phosphorized scanline monitors only, with heavy tubes and double screen refresh. With half the y-resolution, by skipping every second line, and double the frame rate (50/60Hz, PAL/NTCS) it was made sure that either all odd lines or all even lines of a full frame were visible at the same time. So unlike those very visible bars nowadays in real interlace times those bars were not visible onscreen while the clips were running but provided a smoother playback.
@@jiffijoff9780 aw man. Thanks for the legendary replies! You definitely get a sub and like for that!! :)
@@jiffijoff9780 wow... all stuff I'd prefer to forget!... I still get confused by odd or even line first 😂
@@extrashot haha, I'd too. Including mpeg2, what a nightmare codec. Remember tmpeg professional? Always fiddling with the profiles and bandwidths.
Great presentation. I have both FCPX and DaVinci Resolve Studio and I was wondering what process did you use with Resolve to enhance 1080P footage to 4K. If you have an existing tutorial on upscaling SD or 1080P footage with DaVinci Resolve Studio, I would be very interested to see it.
I'm only just getting into learning Resolve... impressed so far! Thanks for the suggestion though.
Great review. Really good, I upscalled 40 seconds of a video clip 480p to 1080p, really a good job, but It takes 4 hours in my notebook i7 5500u.
Thank you... it does take a while... especially without an NVidia GPU... but 480 to 1080 is usually worth the wait!
extrashot Yes, is amazing the upscaling .. really worth.. i’m upscaling a concert.
@@lucasleal4035 an i7 5500u only has two cores with hyperthreading. It is a fairly weak processor, especially for something like video editing which tend to utilise more cores. Mobile processors also generally have lower frequencies and thus lower performance over desktop computers.
Thanks for this great review
Thank you for the comment!
Excellent presentation!
Thank you! Cheers!
Thanks - I've seen those damn adds constantly and assumed it was rubbish. Appreciate how you took the time to use FCPX to replicate it - made the overall review that much better. What is the bg music you've used at the end of video?
Thanks Chris, good news is it's definitely not rubbish, it's doing something! I've just downloaded the latest update and it's also improved quite a bit with more picture quality options... also a fair bit faster on a Mac as well. The music was my own mix from the Filmstro service, which I used to use before switching to Artlist. Unfortunately, because I no longer subscribe, I don't have access to it any more (which is one of the reasons I switched).
You're very believable...and that's obviously very important. Thank you
I have been using this for a while and for anyone interested in the time to do the project element I have come across this. I have an older HP pc which is in any terms fast when I first used the program with the pc as is a 5 min clip took 4.5 days to convert original footage 720 x 576 taking it up 200% I changed the processor to an AMD A10 5800K with only a slight improvement (but still happy) I then changed the graphics card to a RTX750ti (second hand 60 euros) and what a difference 4.5 days down to around 4 hours. the latest version of Topaz does have interlacing De interlacing function which is superb. I ran some DV cam footage and again upscaled 200% the end result was very good no jagged edges at all and much better quality.
I am currently saving to get a gaming computer for flight sim software but cant wait to see what the AI video does on that machine as the graphics ram goes from 2gb to 10 gb.
This software needs a good graphics card ... there is an option within the software to use cpu or gpu if you have a gpu run it through that.
Did you mean GTX 750 TI instead of RTX 750 TI? Because trust me, if an RTX had 2 GB of video memory, it would be a terrible choice. Most RTX's have well above 4 GB of video memory... Just sayin!
@@QueenJeremy yes you are right its the gtx I know it has 2gb. I have been looking at rtx cards as i want to build a high spec gaming pc. got mixed up
lmao that zoom and enhance cliches from the films.. We'll get there eventually, its hard to fill in detail you was unable to capture because tech wasn't good enough or you just recorded it with a potato. Its a good start though plenty of people will have really old crappy clips left around from like old mobile phones or something so this might be good to improve that. Nice video dude been looking at this software for ages but its a bit pricey for me at the moment
They often run deals on the price, but it's worth testing on a trial version first because you need a nice fast graphics card to speed things up a bit. Oh... and I normally use a potato mark iii ....
You should check the program again, the improvements they have made are quite interesting. There are almost 20 different AI rendering models now as opposed to the three you saw.
True... I've been playing with the latest version and it's improved a lot. I should do an update as soon as I have a moment.
Watching your example enhance ai seems to do something a little more than sharpening alone. I think this technology just needs a few more years before it’s in every editor and we can all call it “enhance”
It's definitely doing something more than I can do with FCPX or Resolve alone... works well with poor quality footage but is quite subtle once the rushes are good. Cheers for the comment.... and I love your videos by the way... keep them coming!
Really hoping that eventually, you'll be able to give this (or a similar program) a few still shots from the old locations (and people) and it can incorporate that into its learning. There are plenty of old TV shows and movies out there in SD, but with production photography at much higher res available.
That's a really good idea... give it your own frames to help it learn... I wonder how long before we see this sort of stuff?
Super-resolution AI models are already trained with many millions or even billions of samples. Retraining it can take days, and selecting the right learning parameters is not a simple process. Only thing that will make it better are new architectures, which thankfully pop up every year or so, with increasing frequency.
@@pawala7 True... it's easy to underestimate the processing involved here!
It'd be great to get your thoughts on it now that theres ai models that can handle fixing interlaced footage
On first try, it seems to do well... as soon as I get some time I'll have another look.
Hello. If I render a video from adobe premiere with high quality with best render settings and filmed with professional camera,lenses and lighting. it the Video Enhance AI can make a better difference for the video? it can boost quality of video or not?
If you're starting with pro quality, I'm not sure what you'd be trying to achieve unless you're increasing resolution or sorting a noise issue?
you don't have the render the videos fully to try each upscale AI setting, you can actually preview the upscale setting with the preview button on the upper right hand corner of the app which will generate a short preview.
That's useful... I didn't realise you could save the preview. 👍
Thank you so much for this video
i love this professional reviews, they know their subject ! everyone get very hyped with Ai, but sometimes thats lots of calculations for not much
Apparently the latest version v1.9.0 can work with interlaced video now. I'm thinking about whether or not I can justify the expense.
Yes, just seen they've added that... it could be useful. Try the trial first !!!
Another interesting, well put together review! And ‘this time of some magic software... enhance it.
Personally I thought it might be fun and “quick” to try and enhance some old progressive 3CCD footage from the 90’s (Sony trv900, i think?).
Fun stuff and thanks for the thoughtful, in depth review. Now back to my FX9 👍
Ahhh... just the phrase 3CCD brings back memories!
You just saved me £200 bucks as I dont have any SD or dodgy footage, just clean 1080p upwards. Thanks.
If you're still curious... a lot of people seem to have success with the trial versions!
Also try mode stacking. I use CG without any up scale then upscale on a second pass.
Thank you for your testing and insights! I'd downloaded the trial last night. This was quite a help. One minor quibble: You say the VE setting to upscale good HD to 4K is Gaia-HQ, but your textbox over the footage still says Gaia-CG. So which did you use?
Also, did you try any of the image sequence settings or just the mp4?
Thanks again!
Ahh, I see what you mean.... yes the graphic is correct to what you're seeing. My words were recorded later and I'd re-run some of the tests. Sorry for the confusion. I didn't try outputting stills and then compiling into video later... so all mp4. The stills route probably would be better for anything you want to edit later as you can then choose the compression settings... especially with .tiff's
Amazing review, hands down and in general on AI
Thank you for the kind comment, appreciated!
This was a interesting video. I too seen this software advertised and wondered if it was worth anything. By watching your findings it seems that it is worth getting. Especially if you are in a tight spot and need to deliver a project that is in HD and want 4K. Also if you are editing and ready to deliver a 4K project and you notice some clips you have are in HD and want them upscaled to 4K for the clarity to show the clips in your 4K project, then i can see the worth. Anyway I probably won't buy it but I can see where someone may be interested in having this software.
When I started making this video I thought I'd have the chance to take the mickey out of their exaggerated claims, so seeing how well it worked, was a genuine surprise!
main problem is with people's faces. They just don't handle faces properly. It is a bit like turning up the sharpness on regular video of buildings or trees, you can get away with a lot of sharpening, but on a face you can't. It is a bit strange as if it's really AI it should recognise faces and handle them in a different way.
@@IDKOKIDK I do know what you mean... faces are hard (as the CGI experts will confirm).
Thank you. You were right, Topaz did work.
Glad it worked for you... cheers for the comment.