Topaz Video AI: Finding the right AI model 

I love Topaz software (I now own a few of the Topaz software titles with Topaz Video AI being one of them) and am impressed at how great of a job it can do. Keep up the great work Topaz team!

It would be nice if the stabilization option had choices for which aspects we wanted stabilized. So for example rotation and zoom could be switched OFF, or tilt could be OFF with only panning ON.

Thank you saved me a lot of time trying to browse videos for an actual working one

I love this software! It's miraculous. I bought Topaz Video AI one month ago and have had time to work with some older videos, so my comment will be about that. I have some old music concerts and TV broadcasts I sang, that were recorded on many various video formats from 25-30 years ago (1994-2000), and am currently doing some "25th Anniversary" uploads to honor the occasions. The results are spectacular! - Whatever the original source was, and however good (or bad) it was, the improvement with Topaz Video AI is astounding. I can't recommend this software enough. BUT I've found that it takes a fair degree of fiddling, experimenting, trial and error, and generating scads of "previews" before determining which models work best for each type of video source. I have different footage shot on various video formats from those bygone eras, ranging from interlaced VHS, VHS transferred and burned to DVD (some interlaced, some not), 8mm, Hi8, mini DV, all the way up to professional (ProDV, Betacam SP, and even some concerts shot on U-matic SP recorded and broadcast in India and Trinidad). The improvement is amazing! But you need a FAST computer. I'm using a base level Mac Studio with M2 and 32g memory, which seems to render 15-25 times faster than my M1 Mac mini with 8g memory. For example, upscaling 2 hours of DVD-SD footage to FHD (eg. a DVD at Standard Definition 720x480 DVD upscaled to FHD 1920x1080) takes 4-9 hours on the Mac Studio, versus about 2 days on the Mac Mini. And if you add frame interpolation, de-interlacing, or any other filters to your output, it can take longer. (Note that de-interlacing typically creates twice as many frames, eg if the original footage was shot at 30fps interlaced, then the de-interlaced version created using Topaz AI is 60fps-so this takes some additional time to render. You may prefer to first de-interlace using HandBrake, then upscale the HandBrake's 720p footage to FHD using Topaz AI. That saves time, and the results seem about equal.) The "Frame Interpolation" feature is just amazing. It uses Artificial Intelligence to create additional frames in between existing frames. This is really helpful for people shooting action, or distance footage with panning. For my purposes, sometimes the camera pans left to right shooting from a distance, so the footage appears jerky. For example, in one of my sample clips, when the large audience is panned from a distance quickly at 30fps, frame to frame seems jerky. This happens because the large audience was shot and panned from a far distance, so as the camera panned, each captured frame (which lasts only 1/30th of a second between frames) appeared to jump quite a bit from one frame to the next. But the Frame Interpolation feature on Topaz AI creates additional frames in between each existing frame. The result is, the audience pans (which were quite jerky in the original footage) become much smoother! So too for any fast action on the stage, for example dancers on stage, or fast-moving sports clips as well. Here's an example of a song I did in the Philippines that was shot on Betacam SP (but the original source given to me was pretty poor to begin with, more like SD). I uploaded it 12 years ago onto RU-vid here, without any processing: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-Btbo42jKRzA.html. You can see the quality is very poor, jagged and ugly. Now look at the improvement in these four upscaled samples using Topaz Video AI at 60fps: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-Hn0e5O_i4nc.html . It's remarkable. I'll be working with this program for years to come, to learn what works best with what types of original video sources. Truly a miracle, that's what Topaz Video AI is. Way to Go, Topaz! UPDATE (10 months later): Topaz AI keeps improving. The new "IRIS" feature really improves upscaling of faces shot in low resolution. I'm going back to some projects I upscaled 10 months ago, that didn't turn out very well despite trying every possible AI model that Topaz had. Now, when I upscale and enhance using "IRIS", the appearance of faces is very much improved. So I'm going back to re-do some of the footage, this time using the "IRIS" model.

The new 3.0.0 is ALOT slower than previous versions. A 1 minute from 1080p to 4K is taking 30mins on a 3090 Ti / R9 5900X

If you're upscaling a video why can't you keep the same frame rate while changing the resolution? Why is the Enhancement pane locked? Thanks. BTW noticing now you don't seem to be responding to anyone's questions. So that's not so good.

After alot of research, i purchased video AI and I am excited to use it. Could you please put together a full tutorial on how to use it and best practices based on the new V3.1 version... thanks!

I have a problem, when I convert a 30 or lower fps video to 60 fps and upload this file to RU-vid, RU-vid only shows 30 fps. How can I fix it

Thank you Topaz Labs form this Video Enhance AI update. There are some welcome additions to your update, along with a busier UI. For example, it is necessary to scroll right up to the top of your control panel to see the scale of the incoming & outgoing file. Also, ganging chronos with proteus in one pass almost virtually stalls preview & render.

We used to have the ability to compare the output of the models up to 4 way in the v2's. It's bizarre to me that you guys scrapped that feature in v3.

I’m curious. Does the recent version add 47.952 and 48fps as options on the drop-down menu for video output or does it have to be added later on in future updates when they happen?

There is a huge bug in this version. It saves the same video twice and loading previews takes forever compared to the previous versions.

What Mac model do you have? And how long does renderning (upscale) approximately take?

This software does a great job on really old family videos. I have some terrible looking videos recorded on bad digital camera from like 2008 and it really helps clean them up and make them legible on a big screen

I imported an rather old interlaced video clip which is black and white and I'm using the trial version. In selecting Deinterlaced and HD, it didn't do wonders for me. Then I selected 4K upscaling and the video type as interlaced. The 2 second preview takes for ever to render, never mind how long the whole clip will take. Are my settings wrong and do you have a suggestion for me please?

I have been using Topaz Labs still image products for about a decade. The introduction of Video Enhance - now rebranded as Video AI - was a wonderful innovation, & I use it regularly. That said the slow motion & stabilisation models in this app are unforgivably slow. If film/ video motion artists are impatient as I am for Topaz Labs to catch up with their competitors, they will use other software vendor’s products to do this job more quickly.

How do I stack filters? For example, if I want to apply fine tune and upscale on the same video at the same time.

hi~ I wanna ask this software is more demand for CPU power or GPU power ? If I just want to upgrade one of these options.

So you can't choose the AI upscale module when Deinterlacing?

I have technical question ⁉️ how this ai work like for example i want to process 1 second footage to get best outcome so i have to give input of 3 to 4 second so it take details form frame. Or just one second is enough to achieve same result for slow motion and upscale. My point is which type of file give best output.

One thing for sure, this software will save me alot of money, don’t need upgrade system, still cut HD using my old software which has load of plugins, and no subscription required. Then upscale 4K for client that needs it. I still shoot 4K though

How do I actually unlock enhancement then if I’m upscaling

if there is a way it can automatically detect interlaced video this would be handy.

Need to go into more detail of the AI model differences

I have a criticism and I think more should be talking about it. The problem is that the before and after looks amazing in the software. But only in the software because when playing back the vid after rendered out it is nowhere near as good looking.

Hey Topaz Labs, how about a non-subsampled (4:4:4) output? while 4:2:0 is often fine, there are some cases where it really shows (pixelated look) and not making that a toggle in a software for enhancing video quality seems odd to me. hevc_nvenc supports 4:4:4. x264 as fallback would also be fine. Also, where did the settings panel for managing models go? :( (Download/Delete)

Hi, In GPU mode, when I choose option 8k + 120 FPS, I always get this error notice: Last FFmpeg messages: Unable to parse option value "0" as video rate Unable to parse option value "0" as video rate In CPU mode, I cannot download any models please help ! Thanks in advance.

in my experience with VEAI v2, Artemis medium/high is also a decent bet for just upscaling at decent speed. ymmv

In Proteus is manual estimate the same as auto?

is there a way to change the color saturation to a video?

I try to update from 2.4 to 3. it downloads and asks update now? and after I select Yes nothing happens Mac Monterey.

Why did you remove h264 from the software?

The v2.x had Computer Graphic setting. WHat is the default for that in the 3.x version?

I would love that TOPAZ VIDEO AI would have a COLOR EDITING INTERFACE OR OPTION !!!!!!!!!! PLEASE include it on your next update!!!!!!!!!!!

THe software most time just sits at Running Model.... I went back to using Video Enchance

So guys, any idea when you are going to fix the whole version 3.x.x car crash?

I'm using the free version but every time I use the enhance ai it always gives me an error. Does anyone have a solution?

is there a model for fixing overexposures?

It is great software once you learn how to use it and you are patient enough to tinker with the settings I probably do about 20-30 previews before deciding on an export

Guys please help, which one should I buy, DeNoise Ai Or Sharpen Ai? For making 4K image

Thank you.

Works good, tnx

why does it show a white square like it's only gonna upscaled the middle of the screen of the video?

Which one would be the most recommended for anime type videos?

Further to my last comment I have been using this latest iteration of Video Enhance AI to time remap, or slow down footage. The result isn’t bad, but the length of time to render is enormous. Topaz Labs need to warn customers of the time involved in rendering slow motion; reduce the time this takes to render in the next iteration, or suggest customers using this function employ a “spare” computer for the task.

If i purchase it. Then is it one time purchase with all future upgrades or with yearly subscription???

I bought it and love it, but my nvidia only 2 years old (asus rog of 1800 euro at the time) already cannot handle it, takes ages.

Do you know how to launch the v3.0 because My version which is the v2.6.4 And I click the bottom for the update And download but after that nothing happened do I need to uninstall the old version for that I don’t know I’m confused ??? Please help me

Why does my video go in slow motion when it’s not

Tutorial is very good but how to remove watermark . Please help !!

Hi Thanks for this video, I am trying out the Video Ai presets and getting mixed results so far. The biggest issue is the time it takes to render and some of the results , maybe I need a faster MacBook to really make use of this program. if work flow takes so long to do anything is not going to be useful to me . For example 3 hours to stabilise a 30 second clip which is not too gittery anyway, that's just not useful! I will keep trying but atm its looking like I am heading for the 30 days money back offer. I use a MacBook 16 from early 2020 maybe I need a M2?

Why Topaz Not Working on AI Rotoscoping

This video is titled "Finding the right AI Model", but the presenter didn't give any help in choosing between Proteus, Artemis, Gaia, and Theia. Those are the AI models.

I also want to ask i have pan shot and one with still shot pan has multiple prospective so pan shot is better for process or still shot is enough for upscale and slowmo for ai.

GENIUS

Gaia seems to work best for me at upscaling, but I'm not sure why that is.

Hi, you should have shown the output. That "Before and After"

It says got me model not found with an x why does it say that

amazing..

Благодарю интересно и полезно.

We need a showcase on what every manual parameter does. It takes forever to load a preview wile testing settings

Wedding ki Shaadi Ka video clear nhi hr...theek ho sakta he kya.

i keep geetting a FFmpeg.error what is it

And I agree with other comments in this thread: this version of the app is a lot slower. The question is why?

how to download models bro its shame that your app costs money and still cant download models

anybody want my build g1.assassin/16gb ram/d9dx cooler/this cpu/3060 for 570 or without the gpu for 300

Al dia de hoy GAIA no funciona en upscale, da un error, lo demas de acuerdo con lo que se dice, lento, deja temporales por todos lados, un desastre

As these AI and other programs get more powerful, your computer also has to become more powerful. I have now invested in a rendering only laptop that I use solely for that purpose, nothing else. So if it takes 8 hours to render I can just let it do its thing. Put it on a laptop cooler fan and off to work I go. I have an I9 paired with a RTX 2070ti that can handle anything and does not get too hot even after 24 hours of rendering. Get a dedicated machine guys if you do a lot of rendering photo or vids like me. And get one with the newer DDR5.

it takes a while to export a 2 second video

I bought a topaz video and the load takes forever. A video of 40 minutes 8 hours is normal?

To convert from 720 to 4k, a movie of 80 minutes, it would take approximately 53 hours with Ai model Proteus ????!!!!

A lot of great answers. 🤦‍♂ Also 2160p is not the same as 4K. 🤦‍♂

Unfortunately absolutely useless in practice. I tried to interpolate a video from the "Insta360 EVO" to 8K with the demo version. For 5 minutes of video, the program needs a whopping 24 hours on a Ryzen 9 with RTX 3070. And then the quality wasn't even particularly good, a little better, but nothing that justifies the effort, and especially not the price .

mayby AV1 support :D

This new version is so much better easy to use and clear

I'm getting great 4K upscale results using my old Canon HV20 HDV footage... here is a recent test ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-bf4iwQQUs2k.html

insanely slow, EST is 2 weeks, sometimes,1 year. we need pause and resume options for video enhancer , and on other products too. :(

i cant even process it wtf

sorry but this may work for a few but this is sooooooooooooo slow! 1 day and 18 hours and 29ec to export a 10 min file just adding a little sharpening ??? usually would take 2-3 hours !!!! so not productive !! very upset!

And it's uploaded in HD hahaha

duuh. How about some instructions that tell us the non obvious things. Less info than the popup help.

Way to not even demonstrate the video stabilization

Bought this, what a load of trash! 1. Cant pause and resume 2. Cant save a project 3. Cant work in frames instead of time for trimming 4. Cant even copy paste timecodes so now i have to Manually type the IN and OUT point in every god damn time!!!! Seriously WTF are you devs doing on a daily basis that you dont even have a save function in a $300 software. I STRONGLY suggest that you people buy Davinci resolve (also $300) and see what REAL software looks like that isn't created by lazy developers!

Blur video bos

Your product is phenomenal and a one of a kind but it ain’t worth $300, a more appropriate price would be like $30 to $50

Upscaling this video to watch in 8k

who needs this when you phone can do better with AI apps hat are FREE!

Totally slow... Please come to Cloud computing era ;)

are you kidding me? you didn't even bother to show us the finished samples? wow

Ok but why is your program $300? Thats too much

I stopped paying for this software when version 3.0 came out.

old interface was better. hm I see ffmpeg.exe and alot of cpu work.

Useless 👍