I now enlarge both images and I compare.
I personally like the images made in avc better than in av1.

Thank you for sharing your impressions.

I like objective comparisons better than subjective ones, because no two sets of eyes, brains, or expectations are the same. Ever.

That said, I agree that I prefer the relative accuracy of AVC over the smaller files produced by AV1.

There are quantitative benchmarks freely available to you, if you should choose to go that route. The most commonoly used ones are called SSIM and PSNR.

The PSNR is based on tried-and-true Root Mean Square Regression analysis to measure the Signal against Noise in a common ratio. If you took Algebra II, you may recall this.

The SSIM test goes one step further, introducing human perceptual weighting, which can be measured and factored in large population samples.

I enjoy reading others' subjective choices, and I try to figure out the criteria they are using, perhaps subconciously. It is important to know that the eyes themselves are a constantly moving target, making the observer an inseperable part of the equation. Here's one demonstrations of the fallability of human vision over short periods of time:

https://www.vegascreativesoftware.info/us/forum/color-fatigue-demo--118176/

I read that it makes videos rendered in av1 small files and with better quality than avc and hevc.
I tested this and this is how I proceeded :
First, I import an original avc video
I render it back with the default settings of Vegas Pro to the format av1 and to the format acv.
I reimport theses video-files and re-render both video-files to an image-sequences in png.
I now enlarge both images and I compare.
I personally like the images made in avc better than in av1.
Am I wrong or did I not understand ?

 

There are major artifacts here with the AV1 render. Can you share the settings you used to make it?

What's the source media resolution?
What's the project resolution?
What resolution did you render it to?
Was project render quality set to best?

Do these artifacts disappear with a higher bitrate?

@RogerS Hi, Magix Movie Studio has a warning that AV1 requires 40xx series GPU or Intel ARC, any export I do from that program shows flashing black frames when the media is dragged back onto the timeline.

If I render with Vegas Magix AV1 I get the same flashing black frames, does the same 'requires 40xx series GPU or Intel ARC' apply to Vegas also? I can't find it online but admittedly I've only skimmed through the pages with a 'Find' search.

This is a Vegas AV1 rendered clip dragged back onto the timeline.

@Howard-Vigorita @RogerS

I used the preset settings that were set in Vegas Pro 21 for rendering in av1 and avc.

I will soon do the same test with hevc and proreg. I will include the settings and file length used.

Hi, Magix Movie Studio has a warning that AV1 requires 40xx series GPU or Intel ARC, any export I do from that program shows flashing black frames when the media is dragged back onto the timeline.

If I render with Vegas Magix AV1 I get the same flashing black frames, does the same 'requires 40xx series GPU or Intel ARC' apply to Vegas also? I can't find it online but admittedly I've only skimmed through the pages with a 'Find' search.

This is a Vegas AV1 rendered clip dragged back onto the timeline.

@Former user I don't believe VEGAS implemented hardware encoding yet for any GPU but can't test as I know mine isn't supported. Intel UHD770 can decode AV1 but this doesn't appear to be implemented either.

The * for Magix Movie Studio is for "GPU acceleration" only and shouldn't have any connection to simply functioning or black frames.

Do the black frames show up in a media player as well? I did test renders with AV1 and hadn't noticed any issues with it. I can test more later.

Do the black frames show up in a media player as well? I did test renders with AV1 and hadn't noticed any issues with it. I can test more later.

@RogerS No, the rendered files play fine in media players

Well, that's odd! @Amivideotek Does your file look any better in a media player?

Trying to understand if this is a decoding or encoding issue or both.

Edit: tried a source AV1 file and one I then rerendered in VEGAS and neither had black frames upon playback with my laptop. I realize there aren't any bitrate options- my apologies. I do have project set to "best" quality. I'm rendering the sample "ad" project now to test. It's quite CPU intensive, though that shouldn't be a surprise.

With moving high detailed video something is wrong:

Screenshots from my media player (MPC Black) from the VEGAS AV1 render:

With MagicAVC/ NVENC it looks better than that

I'm out of time to test more but here the AV1 output isn't acceptable.

@RogerS

It is difficult for me to compare the quality of moving images from videos played with a media player. That's why I do it with still images (snapshots or image sequences). That's the way I use.

I honestly think there is a drop in quality when coding in av1! But I will try again later and more accurately with the same number of bitrates, framerates, screen dimensions, etc

With a media player I just paused the video and did screenshots above. With MPC Black, control + right or left arrow lets you advance a frame at a time.

Based on that I don't think this is a VEGAS playback issue, something is wrong with the AV1 render itself. I'm just trying to isolate what the issue is as too many variables makes it hard to understand what is happening.

Yes, from your photos it is very clear that there is a very big problem with rendering in av1. This diamond pattern was not visible on a normal scale in av1! Only when I zooming in. My snapshot was enlarged many times! But like I said, I'm going to try this again later. So if this rendering problem is solved, rendering in av1 might be worth it after all!

I've tested Vegas av1 on Intel graphics with objective ffmpeg quality analysis, visual inspection on a 4k Samsung big-screen tv, and YouTube uploads. And have seen none of the issues cited here. Only issues I see are performance and option related.

Also tested with Voukoder, Vopukoder Pro, QsvEncc64, and ffmpeg, which allow selection of much faster qsv_av1, and that's good too. Unfortunately there's still a Vegas quality hit using any 3rd party render plugin, including FrameServer. So I don't consider that an option till Vegas opens up the pipe it gives render plugins.

The one standout I noticed testing an av1-lossless clip is that Vegas av1 decoding quality is pretty good... right up there with vp21 build 108 legacy-hevc and build 187 qsv. Fwiw, I think Vegas is well positioned if new cameras start delivering av1 clips. I'm still getting higher performance rendering hevc so I'll still be using that for YouTube uploads for now. Probably stay that way till Vegas gets direct support for specific av1 gpu hardware and some bitrate options.

I tested again the different video formats that interest me. See the various photos (zoom 600%) in the attachment. When I compare the different formats, I currently continue to use the hevc format. The prores format is better, but the files are very long

.

@Howard-Vigorita Can you elaborate on what you mean, or be more specific:

Unfortunately there's still a Vegas quality hit using any 3rd party render plugin, including FrameServer. So I don't consider that an option till Vegas opens up the pipe it gives render plugins.

Jøran Toresen

AV1 encoding from a NLE doesn't seem to have much of a point currently. I thought maybe for discord as a playable video format. Discord(free) only allows 25MB uploads, so you want as much efficiency as possible, but currently only accepts AVC and VP9 as video formats.

The Nvenc AV1 hardware encoder unlike H.264 and H.265 NVENC and software x264/x265 doesn't keep improving the higher bitrate you give it, It seems like it's only purpose currently is with streaming if you were limited to lower bandwidths. In the case of Youtube streaming that accepts AV1, HEVC and AVC, it you only had the ability to stream 1080P60 at 6-8Mbit's AV1 should have a good win, but for an NLE doesn't seem useful given the lower quality of the Nvenc AV1 hardware encoder, or the slowness of the software AV1 encoder. Doesn't have enough going for it.

The meme that hardware encoders are lower quality than software encoders at higher bitrates is actually true with Nvenc AV1

@J-Toresen I've tried a variety of 3rd party render plugins but FrameServer is the most telling. It gets it's render feed from Vegas just like any other render plugin. Transcodes to Vegas internal presets like Magix MainConcept, Nvenc, or Vce yield significantly higher quality than any render via the plugin pathway. I get similar lower quality readings from every 3rd party plugin I've tried, with ffmpeg via FrameServer the best of the bunch. There is a quality chart linked in my signature cataloging results I've obtained over the years. A link to the lossless clip I use is there too. I check periodically to see if anything's changed for 3rd party plugins, but nothing yet. I hope it's being worked on but am not sure if further development is contemplated.

@Howard-Vigorita Can you see the difference in quality with your eye or are you just going by benchmarking software?

I've noticed before you getting VMAF's in the 70's while I was getting values in 90's. I suspected you were doing something wrong.

@Former user After doing this for so long, I don't second guess the numbers as much as I used to. On a big screen 4k tv, significant difference in the numbers are clearly visible from a couple feet away. I tend to render with cbr 28 mbps 10-bit 4k hevc for YouTube and the differences are very obvious between high-quality decoding with vmaf in the upper 80's compared to those 30 or so points lower.

Looking at your graph, it appears it's not apples to apples. You need to match output bitrates which are not documented. Your reference clip and decoding are also undocumented. I've used a lossless hevc reference with legacy-hevc decoding in the past... equivalent to build 187 qsv decoding quality. If you are using anything but a lossless reference, your results would not be meaningful. The subject matter of your reference is also relevant.... a low complexity reference, like a still photo or worse, a solid color bar, would not generate meaningful results. My preference is the high-complexity reference of the ocean in constant motion which I use for most. When I do all these things, I get the results I've described. I try to carefully document the reference and settings in my charts to insure they are repeatable by anyone.

@Howard-Vigorita The comparison here was about defaults, argument I was making was default Vegas bitrates were too low, they perform much better at higher bitrates. I compared Voukoder X264 , and Voukoder H.264 Nvenc using their default 'recommended' setting.

I use a Prores422 video of a complex scene, a boat on the water with waves and ripples, something compression doesn't like very much. I don't see the degradation using voukoder that you do, the voukoder scores are higher, but I am not verifying results on a large TV as you're doing. Do you feel it's impossible to get VMAF's in the 90's from Voukoder, and if so, where could I be going wrong?

@Former user Since this thread is about av1, if you want to go farther afield with a ProRes sample and Voukoder, it might be better if we carry on the discussion it a thread about that. I promise to chime in.

@Howard-Vigorita Here's AV1 encoding from Voukoder and Vegas's internal version.

Also tested with Voukoder, Vopukoder Pro, QsvEncc64, and ffmpeg, which allow selection of much faster qsv_av1, and that's good too. Unfortunately there's still a Vegas quality hit using any 3rd party render plugin, including FrameServer. So I don't consider that an option till Vegas opens up the pipe it gives render plugins.

Very controversial statement that needs to be peer reviewed.

Do others see artifacts in sample projects like the to ones in my signature when rendered to AV1? I'll do more testing later this week with both my systems.

@Former user if everyone wants to continue with this here, I think I might have figured it out. It's probably the ProRes reference clip being used. It's lossy by nature. In the parlance of ffmpeg quality analysis, if it's already distorted to start with, using that as a reference for further distortion may yield invalid results. To verify this, I took a lossless hevc clip and transcoded it with Vegas to ProRes 422 using nvidia exp-hevc decoding, which is not the best. Got these metrics on comparison to lossless:

psnr: 35.3705; ssim: .9156; vmaf: 51.0876

Not particularly great. Then I took that not so great ProRes clip and transcoded it again with the Voukoder default x264 template. And compared the ProRes clip to the x264 render and got this:

psnr: 44.0108; ssim: .9752; vmaf: 86.1885

Looks great but is it really? It seems the lower the reference quality, the better subsequent renders appear compared to it. To complete the analysis, I also took the lossless clip and compared it to a direct Voukoder transcode with nvidia exp decoding:

psnr: 37.1752; ssim: .9321; vmaf: 61.5343

and only slightly better results using Voukoder direct with exp-hevc unchecked:

psnr: 37.9227; ssim: .9395; vmaf: 64.9605

compared to the same render with Magix Avc Nvenc default:

psnr: 43.0873; ssim: 0.9729; vmaf: 84.7526

But so as to not completely hijack this thread, here's Magix av1's only 4k preset:

psnr: 42.3639; ssim: 0.9681; vmaf: 81.3015

compared to Magix MainConcept hevc 28mbps 10-bit that I use for YouTube:

psnr: 44.8650; ssim: 0.9826; vmaf: 90.0010

Looks like Magix av1 isn't bad but has a little ways to go yet to come up to hevc in quality.