Topaz Video AI Beta v3.4.0.0.b

Hello everyone,

We have another beta available for your testing.

3.4.0.0.b

Changelog from 3.3.9.1.b:

  • Added Second Pass Enhancement
  • Added Dynamic Compression Options (H264 and H265)
  • Added ability to create presets from anywhere
  • Added new user experience tips
  • Quality Improvement for first few frames while using enhancement models (Except Gaia).
  • Added Iris to benchmark
  • Speed up for Iris on macOS Ventura (1x and 2x upscale)
  • Apollo can now run on M1/M2 when used with HD or above input videos
  • Fixes first few frames repeating in frame interpolation
  • Login now uses proxy settings
  • Minimum number of frames needed for processing is now 4
  • Processing on Intel CPU machines (12th, 13th gen) no longer slows down when app is minimized or out of focus
  • Fixes image sequence issues when non ASCII characters are in the path
  • Fixes ANE process using up all Memory on Mac
  • Tesla GPUs now show NVEnc H264 and H265 encoders
  • Log file can now be opened when app is open (Ctrl+K, Cmd+K)
  • Removed encoders.json use video-encoders.json

Potential Fixes

  • Processing failures when exporting large number of frames
  • Intel Arc/iGPU processing issues

Known Issues:

  • Videos with mismatched metadata and streams will display incorrect duration
  • Frame number preview length may shorten on app restart
  • Crop to Fill / Letterbox / Pillarbox preset values may not migrate correctly
  • The “Copy” button for ffmpeg commands is now showing the “Get Logs for Support” button from the error pop-up window
  • Presets do not work with second pass enhancement.
8 Likes

Just confirming since the wording isn’t clear. The ANE now uses up all the memory on Mac? Or the issue has been fixed?

I assume it’s that the issue has been fixed, but just double checking.

The issue has been fixed (already in the previous alpha).

And, finally: Iris is included in the benchmark :+1:

BUT:
As with the previous alpha versions in this Beta the Iris model goes havoc on the actual Beta of MacOS 14.0 Sonoma, creating crazy moiree like artifacts:

This is just with Iris; Proteus, Artemis, Gaia and Theia do work well.

P.S.: The actual release version 3.3.10 doesn’t have this bug.

1 Like

On macOS (I’m using Apple Silicon), the tool tip for “Dynamic Compression Level” shows up when it shouldn’t when hovering your mouse over the gap under the “Target bitrate” section of the export settings.

This shows up when selecting either the H264, H265, or VP9 encoder options.

This same “issue” impacts Windows (tested with a Nvidia GPU) when the VP9 and AV1 encoders are selected.

Which of the three dynamic modes matches the former export on the versions before?

TVAI always used a constant bitrate from version 3.0 up until this release. So to match older versions you need to set your export settings to:
Bitrate -> Constant
Target bitrate -> The setting you picked in older versions

The new dynamic modes offer new functionality.

I did not set up the bitrate myself, it was set to Auto! :eyes:

In which case, the settings you should pick would be:
Bitrate -> Constant
Target bitrate -> Auto

Sadly it seems this option has been removed from all the devices that now support a dynamic bitrate mode. Maybe an unintentional change?


For reference, in older versions of TVAI, if you select “Auto” TVAI will pick a bitrate that it thought would work for your settings, then just export in “constant bitrate mode” with the selected bitrate.

H264 and H265 on macOS (Using Apple Silicon) doesn’t have this option for me. Is this an oversight or intentional and just missed off the release notes?

hevc_videotoolbox (and probably h264_videotoolbox) do support a “constant quantization” or “constant quality” mode.

ffmpeg -i input.mp4 -c:v hevc_videotoolbox -q:v 92 -c:a copy output.mp4

The -q:v is the “constant quantization” or “constant quality” activator. And it operates in reverse compared to most other modes like this. The range of values you can use are 0-100. 100 is the highest quality, 0 is the lowest quality (or auto? If it’s auto, then 1 is the lowest quality).


av1_nvenc also supports a “constant quantize” or “constant quality” mode, same for VP9. Can we expect the encoders with missing dynamic bitrate options, but with support for that feature, to get them by the final release of this version?

The preference window doesn’t have these options, it just shows the old options from older versions of TVAI.

Here’s an example on a Windows machine with a Nvidia GPU:
Screenshot 2023-08-10 130927

The Copy button is missing from the UI when you select Show Export/Preview Command.

Steps to reproduce:

  1. Setup a video for processing.
  2. Select from the top bar Process -> Show Export Command
  3. Notice the missing Copy button that was in previous versions of TVAI.

Screenshot 2023-08-10 131741

The issue has been fixed (and so has the wording in the post, good catch!)

@20rushtonj The new dynamic bitrate settings are using the CQP quality control options available on NVIDIA and AMD GPUs. The dynamic medium setting should create the closest result to the previous auto mode, which uses ‘-b:v 0’ to set a variable bitrate automatically.

The three dynamic modes were tested using VMAF to create different options for balancing file size and video quality.

Here is the average VMAF score for each setting in the current Beta:

High - 98.4
Medium - 88.3
Low - 79.3

2 Likes

Iris is totally broken with MacOS Sonoma Beta.
It gives a totally blurred picture with moire; about the only detail left comes from the Recover original details.

Iris with Recover original details set to 0:


and Recover original details to 100%:

This is just personal preference, but I would of also liked an option that targeted a VMAF of 95.

1 Like

Also, it constantly crashes when quitting the app:
Topaz Crash.zip (15.7 KB)

1 Like

We’re rolling this out to NVIDIA and AMD GPUs on Windows for now, but we plan to add Intel and Apple Silicon options as soon as possible after next week’s initial release.

AV1 and VP9 will definitely also be updated to use CQ rate control.

For this first beta release we’re hoping to get some feedback that will help fine tune the three levels of quality offered in this mode.

3 Likes

Right now the High setting is producing files that are significantly larger than the Medium setting, so we could look into targeting 95 for the High setting. Thanks for trying it out!

This Beta version is noticeably slower than the previous alpha version with Proteus 2x model on AMD. I’m using this setting quite often and this performance regression is quite significant.

Topaz Video AI Alpha  v3.4.0.1.a
System Information
OS: Windows v11.22
CPU: AMD Ryzen 9 5950X 16-Core Processor              63.923 GB
GPU: AMD Radeon RX 7900 XTX  23.941 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	22.51 fps 	2X: 	10.78 fps 	4X: 	02.67 fps 	
Proteus		1X: 	20.12 fps 	2X: 	11.91 fps 	4X: 	02.63 fps 	
Gaia		1X: 	10.95 fps 	2X: 	07.00 fps 	4X: 	02.53 fps 	
4X Slowmo		Apollo: 	18.81 fps 	APFast: 	46.03 fps 	Chronos: 	14.52 fps 	CHFast: 	22.59 fps 	

Topaz Video AI Beta  v3.4.0.0.b
System Information
OS: Windows v11.22
CPU: AMD Ryzen 9 5950X 16-Core Processor              63.923 GB
GPU: AMD Radeon RX 7900 XTX  23.941 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	21.64 fps 	2X: 	10.64 fps 	4X: 	02.74 fps 	
Iris		1X: 	21.98 fps 	2X: 	10.15 fps 	4X: 	02.65 fps 	
Proteus		1X: 	20.96 fps 	2X: 	09.47 fps 	4X: 	02.59 fps 	
Gaia		1X: 	10.93 fps 	2X: 	07.03 fps 	4X: 	02.57 fps 	
4X Slowmo		Apollo: 	18.94 fps 	APFast: 	45.34 fps 	Chronos: 	14.54 fps 	CHFast: 	23.34 fps 	

Thanks for reporting this. It’s an unintended change related to the new button for automatically collecting log files from the error pop-up dialog.

1 Like