Topaze Video AI 4

Hi Topaze Video AI team,

In the MP4 export viewport:

Dynamic/Constant should be Variable/Constant!

In video codecs, we use ABR (Average), CBR (Constant) and VBR (Variable).

Kind regards.


Because it’s hardware encoding, it’s CBR, but can dynamically change some levels within it.

Software or hardware encoding does not change the situation!

What you describe is called CBR Auto or Auto CBR as you wish.

The codec is a CBR with bitrate budget estimated automatically by sampling the content.


So I have to encode in VBR the output!

My goodness :sob:

By the way I lost the last audio settings.

I mean if I set it as AAC 192 kb/s for instance, next new video is again with factory settings.

I think a preference settings could be nice.


I think what he is trying to say is that with Hardware encoding (h264 / h265) there isn’t the same flexibility for encoding like in Software encoding / CPU (libx264 / libx264). there is a limitation in the Hardware encoding protocol for only certain types of encoding. Yes, you gain more speed then Software encoding, but you are limited with the functionality and encoding flexibility.

That is why the best practice is to use TVAI with lossless compression, such as FFV1 or image (PNG/TIFF), then Handbrake it (Avidemux, VidCoder, Hybrid, or any other of your favorite encoding tool), with your favorite settings, that does have Software/CPU (libx264/ libx265) option.

BTW - CRF is the best encoding to use these days for offline playback/Archival (online streaming is a different story), not ABR, CBR, VBR, VBV, CQP(QP)

1 Like

Hi Akila,

CRF or Constant Rate Factor is used with x264 for instance.

I am using ony H.264 VBR with level-3 hardware accelerated process fully powered by NVIDIA.

Which means I go up to 47 times faster on my old laptop computer with an 980M NVIDIA graphic card.

If you prefer CRF, I have also CBR unconstrained but at standard encoding time.

Video quality is only one parameter in the encoding process.



I go for quality vs. file size as priority, not speed.
So for me libx265 CRF is the best deal.
As I said. export as FFV1 lossless compression in TVAI, then encode the output with your favourite encoding tool you are used to that encodes your MP4 as you like it.

1 Like

Hi Akila,

Okay so which stand-alone application is using libx265 with level-3 hardware accelerated by NVIDIA.

I searched and found no solution!

Exporting lossless and reencoding is also a good solution as you explain.

But I am testing actually the export and found weird things in the lossy export.


To let you understand, I transfered all my videos collection, roughly 18 000 videos from 5 mn up to 3 h 20 mn, most are between 43 mn and 1 h 45 mn, on my 120 To Synology server, Hybrid RAID 6.

I reencoded everything in H264 VBR High Quality 5.1 audio at maximum speed in just 6 months, only by night and the week-end.

Impossible to perform in a reasonable time with any other solution.

Also and most important, my television is crashing if I read a x265 video!

My television is only compatible with a limited video and audio profiles of H264.

So I can forget CRF I think so.



Doesn’t handbrake have Nvidia HW acceleration for h265?

Hi jo.vo,

Handbrake does not support NVIDIA HW acceleration for 980M I think so.

It is supporting HW for NVIDIA 1050 and up.

Best regards.


yeah Handbrake uses NVEnc.
it seems your Chip supports it.

I encoded many times using NVEnc in handbrake and OBS studio in the past on my Older GTX 980
I believe the 980M is the same generation of the GTX 980, both are running on Maxwell (2nd Gen)

I’m totally lost with this interface, normally I use a setting for example HD resolution and click preview and then another one the same but with Iris. i only see a preview running in the screen that is not processed. no idea how to see the preview

Hi Akila,

Super to learn that 980M is supported by Handbrake.

I am using Videoproc which is close to Handbrake anyway.

980 and 980M are close effectively!

980 was the computer version, the 980M the laptop version.

I am using an ASUS ROG 17" laptop.

It runs perfectly in the last 8 years without any troubles.


Hi evertvanderhik,

I’m also totally lost with this new interface.

I’m using Video AI since the very first version and later ones, that was including the interface with four viewports for comparing models and parameters.



1 Like

I’m still not convinced you understood that they named it correctly. CRF is the closest thing in H.264 to VBR, but it is never called that. It is not an option in NVENC. Your laptop cannot do the dynamic part of the only encoding option available, that is CBR. Yes it is really fast, but on my GTX 1060 (So newer than a GTX 980) I could never get it to not lose details. Even at really high bit rates. Maybe you’ll have better luck. Your TV might be able to play H.265 but you need to make sure the level is lower, at main L4.0. This can usually be done by changing the pixel format to yuv420. That’s what I had to do to get any of the movies enhanced with TVAI to play on my TV.
Trying to do anything in TVAI with a laptop is going to take days for one 45 minute video. Not a criticism. Just a fact. Software encoding to H.264 is going to take less time then enhancing.
Since you digitized all your audio in 5.1, you should not convert it to aac. Copy it into the final video.

Somehow, digital video will never stop being more complicated than it already seems to be.

Hi ForSerious,

Nice to read you but do you know where is the official explanation/documentation from Topaze Labs support?

Because I can read only this:

Output Settings Menu

The Output Settings menu allows you to alter the export encoder, container, and audio settings.

At this web address:

Also I digitized all Stereo contents in AAC CBR 192 kb/s Stereo and all DD 4.0, 5.1, AC-3 in AAC CBR 384 kb/s 5.1.

Because at the very beginning of this long process, Windows 10 had problems sometimes with DD AC-3 contents.

And by the way my television set is not compatible with H.265/HEVC.

Sony KD-65X9005A, first 4K 3D passive QLED television available in Europe in 2013/2014.

It was 4K, but only H.264. That’s the reason I used these codecs.



That is the official documentation. Some of us have been asking for greater details in it for years now.

In my opinion, TVAI does not offer any output formats that are good enough to be stored and viewed. That’s why we say to export to a lossless format, then use a program that can convert it into something optimized for storage and viewing.

1 Like

I tried to upscale to 4k using a low qulity video. i must say this was a big mistake, this program is still in the developing stage and should not be priced, also to get a refund is somwhat missing in action. then a one year update, What, whos behind that crazy idea. Ai is always advancing, just in three months ChatGpt took over the world so. this is a no go for me.

I think the mistake is attempting to upscale an SD source to 4K and expecting it to be smooth to the eye.

1 Like

Hi Christian,

I think this is not the right use of AI video upscaling!

In fact, you may upscale from low quality SD, 480i/p, 576i/p to HD 720p or FHD 1080p only.

You may upscale from high quality HD 720p or FHD 1080i/p to 4K only.

Other uses lead to bad quality.

By the way when upscaling from low quality SD to HD or FHD, you may use Artemis Low Quality model.

I noticed better results than Proteus which is selected by default when using the pull-down menu preconfiguration SD to HD.



1 Like