Great video once again mr Lane. The reason the Rook footage doesn't work so well is because AI motion interpolation struggles with occluded objects or objects leaving the frame, it works by analyzing how pixels move between frames and since it can't see the hidden or out-of-frame parts, it has to guess, leading to distortions and artifacts. The artifacts are also very visible in the Godwits footage, if you pay attention whenever a bird will overlap another, the Ai can't distinguish which bird went which way and just creates a whole new bird out of thin air. It works great on the Heron footage because the subject is isolated.
Really good basic explanation! In the Heron footage from last week you can see the interpolation struggling with the small fish as it flies into the beak of the Heron, after it tosses it.
I got rid of Topaz software a couple of years ago for several reasons, but you are making me change my mind again. Thanks for that analyze. As a reminder, you can check which camera, lens and settings you did use through exif tool. It works with clips too.
Hi Mike, Excellent Video as always. I have compared the latest Topaz Video AI stabilization with the stabilization performed by Warp Stabilizer in the latest Adobe Premiere Pro software. The degree of stabilization of both software packages is excellent with the Adobe version cropped slightly compared to no crop in the Topaz version. However there is a substantial difference in processing speed. I chose a 30 second long, 4k, 60 fps video clip of a sailboat captured a week ago on an OM-1 Mark 2 using the m.Zuiko150-400 mm lens with the 1.25x teleconverter in use as well as a 1.4x m.Zuiko teleconverter in use. (1,400 mm equivalent full frame field of view). The system was mounted on a tripod with a fluid head but the video had a substantial amount of vibration while tracking the subject. Premiere Pro performed the Warp Stabization in 1.25 minutes. Topaz Video AI required 17+ minutes. Topaz required a lot of time to maintain the original crop but perhaps in some cases this would be important. FTI. I ran the software on a very fast Alienware (Dell) Aurora R16 gaming computer with an Intel i9-14900KF CPU and 64 gig of ram and an Nvidia GeForce RTX 4090 GPU with 24 gig of ram. As a side note, another interesting use of the Topaz software is to upscale 4k video to 8k. The Topaz upscaled video is exellent quality. TWhen imported into Adobe Premiere Pro, the upscalling allows one to double the zoom of a final output 4k video relative to the original 4k. Of course, the crop factor is 2. My video discussed above when processed in this manner has a field of view and degree of magnification the same as being captured with a 2,800 mm equivalent lens. The resulting video quality is also excellent.
I used to look forward to my Saturday morning shows as a kid, and nowadays I look forward to your videos, It would be interesting to see a test done with the Lumix at 120fps and then run through the Topaz AI regarding the quality.
Awesome job as usual! I have a spunky machine so it can handle Topaz quite well. Now, I just have to practice a bit more and see what all it can do. 😊😊
I am currently trying out the similar new function in Final Cut Pro. I am impressed so far with the footage I have tried at relatively modest slo-mo effects.
Ggreat video Mike, thanks. Food for thought. Of course, being able to slow mo standard video means that we can make use Z of af. I don't think there is af in slow mo. I shall definately be playing around with this .
The Topaz AI for the heron grabbing the fish is amazing. It appears that it actually generated frames where the bird closes its eyes as it strikes. This was not in the original footage, but if Topaz is trained upon photos and videos it would know this happens. Very impressive....
Nice one Mike mate very informative sir .. May I ask - have you got a link etc please to the fluid head 'plate and long lens support' cheers. Thanks a lot. Cheers.
Flapping wings seems to be very predictable motion on a clear backdrop. @11:58 the twig in front of the hare's hind is also deformed. But the first swaying twig @11:53 is also in predictable pendulum-like movement. Yet the initiation of the movement is a bit distorted. Bottom line to me is that Topaz is a phenomenal tool but a tool only. It has its applications. But it does not replace figh frame recording. BTW. This is the most educating content on high-level video recording I have seen in months!
Very good, Mike. I'm so tempted to get that. Would it suit my style of video? Maybe, but I'll sit on the fence for now and give it to next year when they may have sorted the render time out.
A great video Mike I’ve been playing with Winxvideo which is similar but 1/10th of the price of Topaz AI, I’ve compared it against Topaz AI and am not sure yet whether there is a significant difference. One thing I have noticed is that whilst upscaling is similar to Topaz AI the difference is marginal and not as dramatic as the advertised images and as for video improvement from poor footage from the DJI NEO drone (which is my real reason for testing WINXVIDEO) not very convincing, but yes a slight improvement; however very slow! Let us know what your final conclusions are Mike.
Clips seems to be G9II,OM-1,G9II,OM-1,G9-1 but not based on details, rather on colors. OM-1 looks oversharpened here and its greens... they seem too green. To the point of being off. But in a pleasing way, I must admit.
Wait a second. It took 30 minutes to generate this 30s clip (slowed down)? I must admit that these generated frames looks very convincing. I would happily trust it to slow down any footage by say x4. But 1 minute to generate 1s of clip seems like a challenge to a workflow.
If you try to slow down a footage of something erratic and unpredictable happening really fast by artificially generating frames, sooner or later you will come across the problem of artifacts. If not on an egret then on a dragonfly. Slow motion recording generally is not intended to make something of normal speed seems very slow, but to make something ultrafast, slow enough to view. In such case interpolated frames may just not fit the frames in between. Trust me, there is no replacement for high frame rates. AI slow down works flawless by up to x2 slower than what you have and up to x4 slower in a pinch, but it is risky. Beyond that it is a gamble. And you don't need to swap cameras. You already have one that is vastly superior. And this is G9II. It is better camera in literally every way. th in terms of stills and vide. Both in terms of stills and video. I also find its ergonomics much better, especially for bigger hands. Not to mention customization.
4K 8-bit from OM-1 looks frankly speaking bad. I watch it on a big screen and even on YT at times I can clearly see the lack of details. But true magic with G9II happens when you shoot in 5.7K and downsample to 4K in post. Or better yet to 5.1K (I happen to have 5120x2880 monitor). Such a footage looks amazing!
Great video once again mr Lane. The reason the Rook footage doesn't work so well is because AI motion interpolation struggles with occluded objects or objects leaving the frame, it works by analyzing how pixels move between frames and since it can't see the hidden or out-of-frame parts, it has to guess, leading to distortions and artifacts. The artifacts are also very visible in the Godwits footage, if you pay attention whenever a bird will overlap another, the Ai can't distinguish which bird went which way and just creates a whole new bird out of thin air. It works great on the Heron footage because the subject is isolated.
Really good basic explanation! In the Heron footage from last week you can see the interpolation struggling with the small fish as it flies into the beak of the Heron, after it tosses it.
Yep, occlusion artifacts are just a fact of every current AI-based image processing software, to varying degrees.
I got rid of Topaz software a couple of years ago for several reasons, but you are making me change my mind again. Thanks for that analyze.
As a reminder, you can check which camera, lens and settings you did use through exif tool. It works with clips too.
Fantastic Mike, very helpful indeed along with some fabulous clips. Thanks for taking the time to make this 👍
You are welcome.
Nice footage. Superb! Have a nice day ! New friend
Thanks for visiting
Excellent and thank you for sharing. It's good to see the comparisons from Topaz
Thanks.
very impressive, especially on the slow motion footage. Great to have that option. Thanks.
Glad you liked it!
That looks really impressive, especially the Heron 👍
Thank you.
Hi Mike, Excellent Video as always.
I have compared the latest Topaz Video AI stabilization with the stabilization performed by Warp Stabilizer in the latest Adobe Premiere Pro software. The degree of stabilization of both software packages is excellent with the Adobe version cropped slightly compared to no crop in the Topaz version. However there is a substantial difference in processing speed.
I chose a 30 second long, 4k, 60 fps video clip of a sailboat captured a week ago on an OM-1 Mark 2 using the m.Zuiko150-400 mm lens with the 1.25x teleconverter in use as well as a 1.4x m.Zuiko teleconverter in use. (1,400 mm equivalent full frame field of view). The system was mounted on a tripod with a fluid head but the video had a substantial amount of vibration while tracking the subject.
Premiere Pro performed the Warp Stabization in 1.25 minutes. Topaz Video AI required 17+ minutes. Topaz required a lot of time to maintain the original crop but perhaps in some cases this would be important.
FTI. I ran the software on a very fast Alienware (Dell) Aurora R16 gaming computer with an Intel i9-14900KF CPU and 64 gig of ram and an Nvidia GeForce RTX 4090 GPU with 24 gig of ram.
As a side note, another interesting use of the Topaz software is to upscale 4k video to 8k. The Topaz upscaled video is exellent quality. TWhen imported into Adobe Premiere Pro, the upscalling allows one to double the zoom of a final output 4k video relative to the original 4k. Of course, the crop factor is 2. My video discussed above when processed in this manner has a field of view and degree of magnification the same as being captured with a 2,800 mm equivalent lens. The resulting video quality is also excellent.
Thanks for your input.
Amazing! Thank you for taking time to do this.
My pleasure!
Another interesting video, Mike. Thanks for posting.
Glad you enjoyed it
Amazing footage from Topaz Ai video editing software even though not perfect. Thanks for sharing. 1 LIKE from Singapore.
Thank you very much!
Very interesting to watch. Looking at the heron clip I got curious of what shutter speed and fps you had on that clip?
I used to look forward to my Saturday morning shows as a kid, and nowadays I look forward to your videos, It would be interesting to see a test done with the Lumix at 120fps and then run through the Topaz AI regarding the quality.
Glad you like them!
Awesome job as usual! I have a spunky machine so it can handle Topaz quite well. Now, I just have to practice a bit more and see what all it can do. 😊😊
Thanks
I am currently trying out the similar new function in Final Cut Pro. I am impressed so far with the footage I have tried at relatively modest slo-mo effects.
Thanks for sharing!
Ggreat video Mike, thanks. Food for thought. Of course, being able to slow mo standard video means that we can make use Z of af. I don't think there is af in slow mo. I shall definately be playing around with this .
The Om-1 is limited to 60 fps with AF.
The Topaz AI for the heron grabbing the fish is amazing. It appears that it actually generated frames where the bird closes its eyes as it strikes. This was not in the original footage, but if Topaz is trained upon photos and videos it would know this happens. Very impressive....
Yes it is.
Nice one Mike mate very informative sir .. May I ask - have you got a link etc please to the fluid head 'plate and long lens support' cheers. Thanks a lot. Cheers.
This film shows it all. Links in the desciption of film. ruclips.net/video/AVQIBM0SDwI/видео.htmlsi=BYT2KJZ8NYMCedLD
Cheers Mike ..
Flapping wings seems to be very predictable motion on a clear backdrop. @11:58 the twig in front of the hare's hind is also deformed. But the first swaying twig @11:53 is also in predictable pendulum-like movement. Yet the initiation of the movement is a bit distorted. Bottom line to me is that Topaz is a phenomenal tool but a tool only. It has its applications. But it does not replace figh frame recording. BTW. This is the most educating content on high-level video recording I have seen in months!
Very good, Mike. I'm so tempted to get that. Would it suit my style of video? Maybe, but I'll sit on the fence for now and give it to next year when they may have sorted the render time out.
There are others too that I have not looked at.
Great video as usual Mike. Any time you are up in Scotland give me a shout and I'll buy you a pint
Thanks.
A great video Mike I’ve been playing with Winxvideo which is similar but 1/10th of the price of Topaz AI, I’ve compared it against Topaz AI and am not sure yet whether there is a significant difference. One thing I have noticed is that whilst upscaling is similar to Topaz AI the difference is marginal and not as dramatic as the advertised images and as for video improvement from poor footage from the DJI NEO drone (which is my real reason for testing WINXVIDEO) not very convincing, but yes a slight improvement; however very slow! Let us know what your final conclusions are Mike.
Thanks. I will have to look at Winxvideo.
Ive never shot video but I did find this very interesting.
Thanks
Great video and as you say, why shoot in slow motion when that makes such a good job.
Thanks.
How much of a CPU and GPU are needed to run Topaz AI?
Better to look at the Topaz website. I would not know.
Clips seems to be G9II,OM-1,G9II,OM-1,G9-1 but not based on details, rather on colors. OM-1 looks oversharpened here and its greens... they seem too green. To the point of being off. But in a pleasing way, I must admit.
Wait a second. It took 30 minutes to generate this 30s clip (slowed down)? I must admit that these generated frames looks very convincing. I would happily trust it to slow down any footage by say x4. But 1 minute to generate 1s of clip seems like a challenge to a workflow.
If you try to slow down a footage of something erratic and unpredictable happening really fast by artificially generating frames, sooner or later you will come across the problem of artifacts. If not on an egret then on a dragonfly. Slow motion recording generally is not intended to make something of normal speed seems very slow, but to make something ultrafast, slow enough to view. In such case interpolated frames may just not fit the frames in between. Trust me, there is no replacement for high frame rates. AI slow down works flawless by up to x2 slower than what you have and up to x4 slower in a pinch, but it is risky. Beyond that it is a gamble. And you don't need to swap cameras. You already have one that is vastly superior. And this is G9II. It is better camera in literally every way. th in terms of stills and vide. Both in terms of stills and video. I also find its ergonomics much better, especially for bigger hands. Not to mention customization.
4K 8-bit from OM-1 looks frankly speaking bad. I watch it on a big screen and even on YT at times I can clearly see the lack of details. But true magic with G9II happens when you shoot in 5.7K and downsample to 4K in post. Or better yet to 5.1K (I happen to have 5120x2880 monitor). Such a footage looks amazing!