Photo AI3 details
price: $299 / £240
company: topaz lab
OS: Windows 10 or 11, macOS Big Sur 11 or later
Processor: Intel or AMD, with AVX instructions, released after 2013
memory: 16GB RAM
Graphics card: NVIDIA GTX 900 series or higher, AMD Radeon 500 series or higher, Intel ARC A750
I first encountered Topaz Labs in the early days of Gigapixel, an AI-powered image upscaler that can scale images to crazy sizes. I remember thinking seriously, how on earth was that possible? Dark arts seemed to be at work, but creators continue their habit of creating incredible software that seems to achieve the impossible.
In addition to Gigapixel, there is also Photo AI (currently version 3) and Video AI 5. Both of these applications follow in the same footsteps, using AI to provide simple tools to greatly enhance, and in some cases rescue, your media.
Video AI 5 is the latest video AI 5 to impress your viewers. With users such as Netflix, BBC, Disney, and Apple, this tool is clearly trusted in the industry. This tool upscales low-resolution video to high resolution, generates slow-motion footage from any source, stabilizes subjects, and generally improves detail, sharpness, and clarity.
upscaling
With gigapixel technology, it's no surprise that upscaling is central to video AI applications. Upscale low-resolution videos up to 16K with easy-to-use tools. This tool can not only increase the resolution of the video, but also handle unwanted artifacts that are unsatisfactory.
This is a sign of things to come. I've never owned a 4K film camera, but why would a videographer spend so much money on a 4K or even 8K camera when they know they can get that resolution for less? I know it's not “real”, but there are so many technical calculations going on inside the camera anyway, does it really make any difference? Not to sound too purist, but I think at the end of the day, a lot of people vote with their money.
And if you think Video AI is just duplicating pixels to achieve high resolution, think again. Through an AI algorithm, it creates new pixels that appear as if they were there, helping to keep the image sharp.
Video AI does a great job. I shot some 1080p footage and was able to upscale to 4K and 8K with no issues at all. We found presets to be most useful for quick results, but it is also possible to specify individual settings.
noise removal
One of the most common problems with video cameras is visual noise. Sure, cameras have gotten much better at working in low light and at high ISO levels over the years, but the technology is far from perfect. Along with general sensor improvements, the software is also very good at removing noise in post-production. The introduction of Intel® Open Image Denoise had a huge impact on denoising in renderers, and similar technology is also being introduced in video editors such as his Video AI.
Video AI 5, trained on millions of video frames, can distinguish between noise and detail, making it very good at removing the former. We were able to pick out a few clips that were shot in very low light with less than ideal settings and convert them into usable clips. The new Nyx AI model has proven to be particularly good at not only removing noise, but also fixing compression artifacts in footage where visual noise is an issue.
Slow motion
So far, I've found that it makes little difference what resolution you record your footage at, and if noise issues occur, but can be resolved at the time of shooting, it doesn't matter either. Another question filmmakers may have is whether they need slow-motion capabilities when purchasing their next camera.
Video AI can create smooth slow-motion footage, even if it's footage shot at 24 or 30 frames per second (fps). Believe it or not, with the Aion AI model you can achieve up to 16x fps interpolation, which is absolutely amazing.
Chronos AI models are especially good at converting 24 fps to 60 fps and even 120 fps. Similar to upscaling, these models are actually creating new data between frames. It uses training to parse the contents of the frames and does its best to determine what would have happened if the frames in between had been recorded within the camera.
I took some footage recorded at 30 fps with a lot of action and slowed it down as much as I could visually believe. The results at this stage were a bit unpredictable and seemed to depend heavily on what was being photographed and how fast it was moving. It was helpful to be able to try out different AI models. However, sometimes no matter what I tried, I couldn't get any usable results.
This is proof that you can't rely 100% on AI to do your work in post. It's fine for playing around with clips and editing videos together, but if you absolutely need to guarantee that your clip will be in slow motion in the final edit, it's best to capture it that way in-camera. is the best.
One of the most useful uses of this technology is being able to take multiple files recorded at different frame rates and convert them all to the same frame rate. No more dropped frames or out-of-sync footage.
Is there anything new in version 5?
Most of the improvements in the latest version revolve around the user experience rather than new tools or tool improvements. The cleaned-up user interface allows the footage being processed to take up more screen space, making it essential for comparing results and fine-tuning settings.
It is now also possible to pause the export process. If you've ever used an AI tool, you know how memory-intensive it can be. Therefore, it is a great benefit to have control functions that allow you to proceed with other tasks. Considering memory issues, we're also glad to see the addition of crash recovery features that kick in when the application crashes or loses power.
Should I get Video AI 5?
At $299, I think Video AI will make a lot of people think. It's not super cheap enough for every videographer to buy, but it's also not expensive enough to break the bank. This middle ground leads many people to think that it might be worth buying rather than upgrading your camera kit.
Questions will be asked. Why buy a whole new camera with slow-motion capabilities when you can use them in post-production? Why get a bigger and better sensor when you can filter out the noise later?
Considering all this, I think it depends on how important the results are. If you need to guarantee that you can deliver slow-motion footage in ultra-high resolution with zero noise, you need to get the hardware to deliver it flawlessly every time. On the other hand, if you want a little more flexibility in your editing process, this tool is worth your money.
