Blackmagic Forum

From 299.00 to $99.00 ending 29.Nov

topazlabs.com.

Ricardo Marty

How is the Video NR?

I have Studio Resolve, and its NR is usually good enough

I hear great things about Neat, but have never bought it.

Not sure how much I would use the Topaz AI up-res. I have never felt the need to use the one in resolve.

ZRGARDNE wrote:How is the Video NR?

I have Studio Resolve, and its NR is usually good enough

I hear great things about Neat, but have never bought it.

Not sure how much I would use the Topaz AI up-res. I have never felt the need to use the one in resolve.

It's not only up-res it's used on old videos to restore. Check it out. there is a trial. https://www.topazlabs.com/video-enhance-ai

Ricardo Marty

It's a really good offer, if you need to restore SD or other older sources, even vintage. With modern cameras, it won't do much what DR can't do on its own.

Bought this out of curiosity. I have some super low res 340xsomething 12.5 fps family footage from one of the first digital cameras I bought in 2000. I prefer the up res options compared to Resolve.

But with the motion interpolation it's hit and miss. If there is too much motion between frames it looses track of objects and creates super wobbly morphy blurry artefacts. Also it has a tendency to create textures that move in the opposite direction as the object. Like the fur of a walking cat. Super weird!

Resolve Speed Warp is not perfect on these as well but much better. For slow motion it's good, but so is Speed Warp. My take on this buy it for up res and deinterlace. I find Resolve's Speed Warp to be better and faster.

I took advantage of the sale, too. My first attempts yesterday at restoring early 90s consumer DV footage were quite successful. I saw the same artifacts mentioned above when the software encountered rapid motion, but overall the restoration and upscaling was excellent. I restored a finished five minute AVI from a Discreet Edit system and upscaled the original interlaced 720X480 to progressive 1280X960.

De-noise was very effective, removing nearly all the (horrendous) noise from an available light, after-dark 757 cockpit sequence.

This software has been criticized for being slow to render. I didn't find this to be true. It auto-selected my GPU and my system rendered the above DV footage at about 5 fps, content-dependent.

Equally satisfying was the "timeline/auto detect cuts" function in Resolve, which worked flawlessly on this first-time attempt. It allowed me to do clip-level colour grading on the entire five minute show.

It's extraordinary what modern software can do with low-quality legacy footage.

Got it also. To bad it does not know what Braw is. I converted some Braw from my 12K after basic color management into HDxHR444. Then I put that into Topaz and played around. Topaz worked well on my system. The thing I really like about Topaz is that I can look at the original and samples of three different variations to make decisions and tweak before doing a complete render. This will save a bunch of time and wasted guessing. There are a bunch of AI algorithms to try and more will be coming. It will be interesting to see what more AI will bring and what competition from Topaz might stir in DR.

Edit: I forgot to mention that it is nice that Topaz can output to a tiff for round-trip back to resolve. Maybe they will incorporate some more codex.

It used to be very slow, but they have improved speed a lot. Instead of huge TIFF sequences, it can render ProRes too. Or is that MacOS only? I don't know if they based it on ffmpeg.

It's impressive what it can do for natural textures if movement is not too fast. I upscaled some interlaced DV footage shot from a helicopter and the landscape looked impressive.

I'm not convinced that Topaz is any better than very careful Sharpening in Resolve, particularly with OFX tools like Sharpen/Soften, Texture Pop, and Contrast Pop.

Marc Wielage wrote:I'm not convinced that Topaz is any better than very careful Sharpening in Resolve, particularly with OFX tools like Sharpen/Soften, Texture Pop, and Contrast Pop.

It depends on the AI model used, Artemis Medium Quality is a bit like that, giving people a unhealthy heroin addict look, the same way as over sharping and over contrast pop does on Resolve. This model should not be used on people's faces. Other models have quite different results

Is there any way to combine different models in one pass with this tool? Like up res and FPS change in one pass? While playing around I exported up res to ProRes and then brought that back for Chronos / Motion Interpolation.

Marc Wielage wrote:I'm not convinced that Topaz is any better than very careful Sharpening in Resolve, particularly with OFX tools like Sharpen/Soften, Texture Pop, and Contrast Pop.

I think is a good tool for restoration, it had a good remover for compression artefacts and restore some small dectails like skin texture and hair where you not have it in some low def picture. Sometimes do miracle on animals fur from lowers video upscaled.
But is AI, a roulette where sometimes win, most of times you obtain same result of good ofx resolve tools combined to beat video shapes tool.
It’s matter of fortune .


Inviato dal mio iPhone utilizzando Tapatalk

peterjackson wrote:Is there any way to combine different models in one pass with this tool? Like up res and FPS change in one pass? While playing around I exported up res to ProRes and then brought that back for Chronos / Motion Interpolation.

Not today, actually I often need to run more models for same shooting, later combine or composite together.
Obviously only where original picture are lowers or where original high res is missing.


Inviato dal mio iPhone utilizzando Tapatalk

Marc Wielage wrote:I'm not convinced that Topaz is any better than very careful Sharpening in Resolve, particularly with OFX tools like Sharpen/Soften, Texture Pop, and Contrast Pop.

I just used it in a Doc that is finishing in UHD. I though it did real good in taking SD footage and making it UHD. We did a lot of tests and it was better than just Resolve for that. the de-interlacing is good but I have a toolset for that. For me it's the SD or 720p footage to UHD that is good. I got it on sale for $99 and have already billed way past that for using it.

You can see by the google version in development using AI can make a huge difference to upscaling
https://iterative-refinement.github.io/

It's still a bit experimental. VEAI is pretty good at natural textures, like grass, gravel, leaves etc.
Human faces are hit and miss, though. Mainly depending on their relative size in the source.

P.S. Rippletraining made an interesting test:

That last-referenced Youtube video showed a rendering speed of 0.6 fps !

Wow !

That means 30fps video would render 50X slower, i.e. 50 minutes to render 1 minute of video.

I could not deal with that speed. In DRS I give-up if the render speed is less than 5 fps !!

IB

Use faster hardware.

Uli Plank wrote:It's still a bit experimental. VEAI is pretty good at natural textures, like grass, gravel, leaves etc.
Human faces are hit and miss, though. Mainly depending on their relative size in the source.

Using Proteus and manual settings, you can find the combination of values where human faces will retain a natural look.

When you pass that limit, then they look worse.