Topaz Video AI Beta 3.1.0.2.b

Topaz Video AI Video AI Beta
1

Hello Everyone,

Thanks for providing your feedback on 3.1.0.1.b. We have a new beta with new models and other improvements. Please test and let us know your comments.

Downloads:
Win RTX 30 series GPUs | Win All
Mac | Mac Update

Changelog:

  • New motion deblur model.
  • New Apollo models for frame interpolation.
  • Fixes stabilizing getting stuck.
  • Fixes close video not killing the ffmpeg process.
  • Fixes grain and noise sliders to produce correct amounts.
  • Proteus/Theia models now use the correct manual parameters
  • The intermediate export files are now generated in the export as location

Known Issues:

  • Apollo Sharpen model has artifacts at the beginning of the video.
  • Apollo models throw an “out of memory” error on Mac.

New models:

Themis (Motion Deblur) - This model is trained to reduce motion blur caused due to sudden and fast movement (such as panning, rotation, or zoom) of the camera.

Apollo Soft (Frame Interpolation) - Similar to the Chronos models, it does frame interpolation (slow-mo and frame rate conversion). But it looks at four adjacent frames (instead of two) to understand the complex and non-linear motion of objects and interpolate accordingly.

Apollo Sharpen (Frame Interpolation) - It does the same thing as Apollo Soft. Additionally, it reduces the motion blur.

10 Likes
6

Processing speed is faster than previous version but still testing for all models and will update later

7

Software keeps crashing during last part of video

8

Holy crap, that was fast!

Just like the new betas! :wink:

Now to continue to ignore the general release again! :laughing: Thanks a lot DEV team for the speedy fixes!

2 Likes
9

Belated comment: Topaz 3.1.0.1.b’s GUI got hung several times today, making it necessary to kill/close it. (and reboot just in case. )

However, v3.2.0.2.b renders at roughly double the rendering speed of the released v3. 0.12 using an RTX 4090. But the current release version’s GUI is very stable.

Also Proteus is much better at rescaling, provided that it is done in incremental passes to reach the final scale. (I can elaborate on this in a side-topic if anyone is interested in discussing the subject…)

NOTE: UPDATED with as few small corrections.

10

Yes please elaborate. :melting_face:

11

OK. I have created a topic. Hope it will be helpful to everyone participating in the discussion. - I will start adding my content starting sometime on 1/13.

Here’s the link:

4 Likes
12

This version doesn’t seem faster. An 8 minute video took over 4 hours using a preset.

For some reason it set my Processors to 2 from 1 and then crashed. Reset it back and it worked again.

The last version didn’t do anything in terms of AI.

13

What CPU & GPU are you using? May be relevant in terms of performance and glitch issues…

14

On macOS 13.1 presets are not selectable and button “Export” doesn’t start the process.
Did work with 3.1.0.1 beta.

15

I tried all the models on the same clip at first where the Proteus model was generating the 0,1 s preview on my 320 * 240 video forever. It never finished. :neutral_face:

16

Trimming a clip while another clip is being processed still leads to system instability or crashing on the 3090 system. No problem with the 6900XT.

Processing more than 2 streams concurrently leads to one or all of the running processes failing.

Speed is fantastic, great work! And thank you for respecting the custom export as file paths too

2023-01-13-18-27-4-Main.tzlog (582.1 KB)

AMD GPU also still fails with any more than 2 streams:
2023-01-13-18-36-22-Main.tzlog (474.8 KB)

17

Mac Studio M1 max: apollo soft and sharpen give error “out of memory” after 5secs of processing.
What is the difference between the 2 models? Dont see any tooltips.
Anyhow, no Apollo frame interpollation possible in this version on mac

18

Could you test the 4090 with the TR?

19

This a known issue, see the first post:

Also see the first post.

1 Like
20

After installation of 3.1.0.2.b RTX30 version it does not preview nor process any video loaded, also no preset can be selected.

Than I installed ALL version and it works perfectly. Processing speed improved from 0.53spf in 3.0.12 to 0.20spf.

1 Like
21

What CPU and Memory (Ram) do you have?

22

Edit: This issue doesn’t appear to be related to AV1. I’m personally having issues with some video clips but not others. Some are AV1 while others are H264.

I am personally unable to get the Apollo Soft AI network to work if the input video is encoded with AV1. The network gets stuck on the Model Loaded set.

Steps to reproduce:

  1. Open Topaz Video AI Beta 3.1.0.2
  2. Load a video encoded with AV1 (My videos are encoded with libsvtav1 from ffmpeg)
  3. Increase the target frame rate and use Apollo Soft as your interpolator.
  4. Start processing. Notice how it gets stuck on Model Loaded.
  5. Try again with a video encoded with something else (E.G. H264) and you won’t experience this issue.

Note: Other AI models like Apollo Sharpen and Proteus do not have this same issue with AV1. It is only Apollo Soft from my testing. (Although I haven’t tested every AI yet)

I should also note that I tested both the Win All and Win RTX 30 Series builds and neither of them worked.

System information:
GPU: RTX 4090 with driver 528.02
OS: Windows 11 (21H2 - OS Build 22000.1455)
CPU: Ryzne 9 5950X
RAM: 64GB