Extremely long time for video upscaling

I just purchased Video Enhance AI a couple of days ago and was trying it on some old video clips. Let me warn people out there that certain combinations of factors could result in REALLY LENGTHY processing times!

I grabbed (in hindsight, a poor choice) an old video clip shot at 1280x720, 60fps. I selected the “Artemis Medium Quality” option and upscaled to 4K UHD just for fun. My computer is a Dell XPS notebook with an nVidia GTX 1050 GPU–not the strongest on the block.

Even though the clip was only 31 seconds long, it took about 3 hours to render! By comparison, upgrading to HD only took about 45 minutes.

I’m not going to bash VEAI, as it worked as expected. At least it didn’t crash. I set my “Max VRAM Usage” to Medium/Low, as per Topaz’s recommendation.

Back of the napkin calculations, this would take nearly 2 days to convert 10 minutes of footage…

My takeaways: Don’t bother with 60fps clips. Don’t upscale to 4K if you don’t need it (I don’t even have a 4K display anyway). I’m stuck with my existing GPU since it’s a notebook. So it is what it is I guess.

Seriously. You bought AI software for advanced operations and you try to work with a toy-GPU card in… a laptop? :slight_smile: Are you joking man? :slight_smile:

I have 4-core i7 at 4.6 GHz all cores, 32GB of RAM and overclocked 3060 TI. I have to wait 1 hour or 2 for pretty short clips.

What were you expecting with this “chess playing” configuration??? :slight_smile:

1 Like

Mike, I’m not that far behind you. I have a quad-core i7 at 2.8GHz with 32GB RAM. Okay, so I just learned that my GPU is kinda old…can’t do much about that.

If you look at Topaz’s “recommended minimum hardware”, you’ll see that it’s far more of a joke. They are touting this as a “consumer level” video enhancer. Not everyone can afford a $700 GPU, and not everyone can even upgrade their GPU.

I’ll treat VEAI as a novelty software, with limited use. I’ll keep it only because it was on sale at half price plus I got another coupon discount. That brings it down to a reasonable level. There’s no way it’s worth USD $299 to me.

Once you get used to the world AI based processing your expectations will adjust.

Many here are running 1080tis with 8 core CPUs some with dual 3090s and 24 core CPUs, and we’re talking hours of rendering. The real bottle neck is the GPUs mind and with a 1050ti you need a lot patience :slight_smile:

1 Like

Everyone here understands the concept of AI. Of course it has to take hours, also expensive hardware. And FYI, this is just the “consumer level”. If you want to work with “commercial level”, you need to spend $30000+ for the hardware only. So $3000 PC is not a big deal at consumer level. Not everyone can afford a $700: yes. But also not everyone can use AI upscaling software. The product targets a small number of users only. If you meet the requirements, you can join. If not, you still can, but reconsider upgrading your PC or just forget it. This is AI. If you want just 4K upscaling, use adobe instead. Or at least software-based upscaling. VEAI is more about hardware than software.

You drive home a good point: this is AI-based upscaling, not just 4K upscaling. I may be on the outside looking in; not “worthy enough” yet to join this exclusive club.

I fear Topaz is going to alienate many users with this product. All their other AI-based apps run fine enough on typical consumer hardware. They are seemingly pushing VEAI to the same target audience, many of whom are in for a big surprise. I will think carefully whether a refund is in the cards for me.

It’s not that exclusive a club. Topaz software runs fine on typical consumer hardware, it’s just that we’ve grow used to many AV processes running fast. It wasn’t so long ago that a simple conversion from mpeg to mp4 would run overnight and into the next day. You’re quite right that there is a difference between an upscale to 4k and what Topaz does. Essentially you’re talking about applying filters dynamically on a scale and with dynamic tuning that makes single filter sharpening or denoising look feeble. Yes, that takes some processing power. But it’s definitely the future of AV enhancement.

1 Like

My “exclusive club” reference was to the world of AI video upscaling. :wink: The other Topaz AI products are perfectly fine on most consumer-level hardware.

I’ve forgotten about the old days where converting video formats took forever. Maybe it’s because the makers of those converters didn’t push the products as hard as Topaz with VEAI.

Most modern video clips are already shot at HD or higher. I saw VEAI as the ideal solution for upscaling my hours of legacy footage shot on Hi8. That is still possible of course, with a lot of patience (and fingers crossed hoping my computer won’t burn itself out being maxed out for hours on end :grin:).

I have a similar aim. The material I am interested in enhancing is largely film and television video that does not and never will be available in better than SD. That’s life, and it’s quite acceptable to me to spend 4 days with VEAI to get something which does a decent video some justice. A modern video card, anything from the last few years, will happily run flat out for days on end. My 1660 super sits in the high 60s/low 70s Centigrade for hours.

IMO, a $3K machine is just scratching the surface at the consumer level. If you’re serious about AI conversion, and thinking about spending the money, don’t short change yourself! A good (Intel-based) desktop machine with a modern Nvidia RTX Cuda card is really the way to go.

I may bitch and moan about incompatible models with certain NVidia drivers, but I’ve been checking around and this (One) is the best (Consumer-level) AI up-scaling program out there for the price, it’s much better than other crappy rent-ware and online-only AI solutions…

It’s well worth the $300 although I bought it at a discount, but still, aside from my grumbling, this software is worth it’s weight in gold!

EDIT: Edited content in parentheses…

1 Like

I wouldn’t say VEAI is the best. For a consumer it definitely is.
Though if you have the hardware and the capabilities then there are open source projects where you can train an AI yourself and sometimes get much better results since it’s specially trained.

At the very least these open source projects are very popular within a community who uses them for animated shows since something like VEAI doesn’t have super specific models for certain content. I’ve seen a lot used for cartoons and animes that’s pretty primo where you’re really only looking for edge sharpening without halo’ing and no blurry effects as you would see from algorithm based upscales.

And of course these open source projects are free. So make of that what you will.

1 Like

Yeah, I should have been more specific, as VEAI is the top AI consumer AI Upscaler, but I remember a freeware animation upscaler called Waifu AI upscaler; I think I still have it, I preferred it to the command-line progs as I’m a visual artists and learner, so I’m fated to never be a coder…

I am waiting for improvement in performance myself as I have a 5950x and VAEI is not yet optimized for the use of all cores 16 (32 threads) (8, I believe is the most supported). I managed to get a hold of a new GPU and would like to see how effective a 16(32)core CPU pairs in conjunctiuon with a modern GPU Simultaneously .

VAEI has come along way in the last 4 months. It will be interesting to see what the devs have been cooking up as they are always seem to be doing something. Your fellow beta testers here are always giving input and TOPAZ Labs seem to really consider recommendations and feedback. Speed of processing has been mentioned before dont worry. This topic (speed) has been covered.

1 Like

Unless you run 2-3 models at the same time, otherwise, more than 8 cores are not recommended. Also using more than 8 threads in a encoding is also a bad idea because of scaling quality concerns. You can try to google about it. More threads will make it worse. I know more threads will speed up any encoding process, but there’s a catch and most people recommend not to use more than 8 threads while encoding.