H265 Support

NVenc is not very good imho. I’d rather see support for using x264 or x265 (and I mean specifically the X 265 / X 264 encoders, not just “any” ffmpeg implementation). Currently I have VEAI output in ProRes and then reencode that.

2 Likes

Ignoring that it probably is using the x264 library compiled in ffmpeg which is the same library used in x264’s encoder I can’t see Topaz enabling any of the x264 parameters considering that only use CRF for the ffmpeg implementation.

Topaz would have to handle the YUV conversion too which they will be getting for free from ffmpeg.

1 Like

What about other video encoders? Such as AV1, VP9… AV1 would be great, it has the best compression, and it’s lossless.

1 Like

AV1 is not mature enough, still in development, it’s not ready yet. Not all PC can decode AV1. VP9 is the same or worse than H265.

2 Likes

Well, it is still necessary to use video compression encoders to be able to share the results over the internet, because many websites have file size limits. A clear example is Discord, which with Nitro has a 100MB file size limit. When I have interpolated to 60fps and scaled to 4k a 30 second video (or even less) the video has taken up 200MB or more. With AV1 or other codecs it wouldn’t take up that much. Many times I have to use external programs to change the codec and compress the video without losing quality.

VP9 iirc is good because it is embeddable on the internet and the final file size is acceptable, HEVC/H.265 is not and the file size is quite large.

Also, more video extensions would be nice, like M4V, WEBM etc.

Idk, VEAI is an amazing program, but it should have more video formats and codecs since it’s a video post-processing software. And it makes no sense that it has so few export options.

1 Like

Those websites don’t support AV1 or anything like that. That would only cause incompatibility issues. For now, AV1 is not a codec that should be used.

On Discord it only shows as a download link, but with VP9 is embeddable at least. Anyway, it would be great to add support for more codecs and formats.

+1 for x265 support.
Even better would be NVENC Turing support.

2 Likes

I know this has already been said in this thread, but not by me:
It is very easy to see the quality degradation from using NVENC. This is a video enhancing application. It makes no sense to add NVENC. VEAI rarely processes frames faster than software encoding can handle. NVENC is all about real time encoding for streaming, or if you want to encode a video at 600+ fps and don’t mind the quality hit.
Not sure how they’ve implemented H264, but even at CRF 17, dark scenes get gray blocks. If they implement H265 in the same way, it will be useless too.

They could add support for NVIDIA CUDA Upscaling or NCNN Vulkan Upscaling…

…but has nothing to do with support for encoding to H265. Not sure why you’re bringing it up here. Look for a thread that already talks about it. If there is none, make one.

I have already spoken on many occasions about it and it is useless, they will not listen and they won’t add it. So as a topic of conversation in any suggestion thread it is valid…

In my opinion, there is no need for the H265 codec - because then you need to pay a lot of attention to the encoding parameters. It is much more efficient to get ProRes422\444\Tiff\DNxHR on the output and then encode H264\265 with high quality in suitable programs - for example 2pass VBR and so on.

While this has nothing to do with the topic, I´ll shortly answer:

CUDA and VULKAN are just means to call up the GPU in order to calculate “something”… There is no “Vulkan upscaling”. Many scaling algos have been ported to vulkan, others to cuda, filters in video editing sometimes come in an OPENCL variant… Its just teh “Programmng language” one uses, not the actual function (filter, scaler…)

In a product like VEAI, we have a so called inference engine, thats the “tool” that is used. A model is stuffed into it and then it does the transformation we tell it to do - in our case video related filtering/scaling/noise tretament/etc…

this inference engine nowadays calls upon DIRECTML in the windows version, or OPENVINO in case of windows on an Intel CPU/iGPU or the Metal equivalent in the MAC World. Cuda is not used anyore (which is a good thing, it opened up VEAI to the rest of the world, not only NVIDIA).

In many opne source video tools we often find cuda versions or vulkan ones, waifu2x for example exists in many variants, guys like nihuy port a lot of stuff to vulkan…

but again - vulkan and cuda are not the actual algos/inference engines/filters, they are just the backend used…

Thats probably why your reqeust is “not heard” :slight_smile:

Yes, I concur! X265 output encoding with the option to use your PC’s dedicated Nvidia GeForce RTX GPU (Turing, Ampere or Lovelace and newer when available) or any capable AMD Radeon RX6000 or RX7000+ series or future Intel ARC high-end GPU’s would be extremely helpful and preferred.

Hardware GPU X265 encoding output straight from VEAI on my Win10 PC would save me a huge amount of time and free up my system’s resources much earlier and the final H.265 encoded output video files would take up significantly less of my expensive and very limited SSD storage space. I could get so much more work done in so much less time!

this feature too is available since almost day one on the actual Developpement of VEAI 3.0 (actually in beta 3.0.0.8b and early access 3.0.0.6)

1 Like

x265 is the name of one specific HEVC/h265 software encoder. There is no “hardware x265” - Nvidia, Intel and AMD have hardware h264/HEVC encoders :slight_smile:

2 Likes