Topaz Video AI 7.2.0.2b

Video AI Video AI Beta
1

Topaz Video AI 7.2.0.2b

Hi everyone,

We have a new Beta release with plenty of new updates!

Once you’ve had a chance to try out the Beta, please make sure to fill out the feedback survey here.

See more details below:

7.2.0.2b Downloads

Improvements and Fixes

  • [Pro] Starlight Mini: AMD and MAC support
  • Starlight Sharp: Additional Resolution Outputs
  • Nyx XL
  • Tunable Preset: General Enhancement
  • Seek Render refinements
  • UI touch ups

Pro: Starlight Mini: AMD & Mac support

image
image1500×495 59 KB

The long requested and long awaited Starlight Mini for AMD & Mac is finally here! One of the most talked about and envied AI models can now run locally on a wider selection of machines. Additionally there is some better performance for NVIDIA users. Please note, that the all the same Starlight mini footage requirements need to be met for these iterations. More in the notes below.

Please note that we are aware that the quality will not match the current public released version of Starlight Mini. This Beta variant is not intended as the final output. This is our first Beta version to ensure that we are able to bring this feature to our AMD and Mac users for testing.

We are requesting the following feedback for this Beta:

  • How is the overall performance of this iteration for your machine(s)?
  • For AMD users, can you output a result?
  • For MAC users, can you output a result?
  • For NVidia users, is there any improvement with speed?

Starlight Mini Notes/Reminders

  • Currently listed as “Starlight” on AI Model list.
  • This iteration of the AI model is both complex and resource heavy. Depending on the hardware you are using (AMD, MAC or NVIDIA) the performance can vary and take hours to process pending the media input.
  • Starlight Sharp is a standalone enhancement and can’t be combined with other filters.
  • Recommended to use square-pixel video footage at or under 1920x1080.
  • Please note that if you have text that is too small, blurry or inconsistent, the result may not be processed correctly. The AI model may interpret unclear text as a texture or pattern.
  • If you have interlaced videos, it is recommended that you deinterlace first before enhancing with with this model.
  • Please note that Multi GPU does not work with the current iteration.
  • It is recommended for AMD, to have 20GB of VRAM.
  • The recommendation for Mac is to have 36GB of RAM
  • A requirement for NVIDIA GPU with at least 10GB of video memory, but we strongly recommend a GPU with 16GB or more memory for optimal performance and quality.
  • One of the minimum NVIDIA GPU requirements, is that the GPU needs the compute capability of at least 7.0, which means all the legacy GPUs with compute capability smaller than this version is not supported, including GTX 10 series and Quadro P series.
  • In addition to these NVIDIA GPU requirements, Starlight mini also requires a minimum of 12GB of system RAM.

Pro: Starlight Sharp: Additional Resolution Outputs

image
image1920×680 76.7 KB

Starlight Sharp, can now get a 4K output. This output is currently using a Nyx variant under the hood to assist in the 4K output. This optional method is available until the model can produce this natively. You can decide your own 4K upscaler by inputting the 1080 result as source.

image
image1123×835 61.4 KB

The following outputs are now available:

  • 1x (Native resolution)
  • 1920x1080 (FHD)
  • 3840x2160 (4K UHD)

Nyx XL!

image
image1920×789 101 KB

The original Nyx is primarily a denoise model tuned for preserving detail in high-resolution input footage. Nyx XL takes this to the next level by learning the frames and outputting a clean and sharp output.

Nyx XL is optimized for sharpening and denoising a wide variety of footage in many lighting conditions.

This model is geared for high quality and high resolution media.

Nyx XL lends itself as a great chaining model, when merging with a different enhancement. A great results occurred when coupling Proteus with Nyx XL as the second enhancement. What kind of chaining combo can you make with Nyx XL?!

Additional Nyx XL notes

  • This is a more complex and heavier variant of the Nyx Model series. Please note, that this will currently produce slow render times. This is something that we are looking to make more efficient.
  • Please note that previews can take quite a while.
  • It’s not recommended to use with Seek Render due to long processing.
  • This is a one size fits all type of model, so there are no modifiable settings.
  • Though Nyx XL is not limited by a resolution size, this variation of the model is not an up-scaler. Essentially, it will function at the resolution you input and render out the same. Though you can output at a higher resolution, the details and quality will be based on the original input resolution.
  • Please keep the long processing in mind with your machine’s capabilities/resources when calculating larger resolution sizes.
  • Cloud is currently unavailable for this beta. But will be made available soon!

Tunable Preset: General Enhancement

image
image436×409 18.1 KB

Try the all new Tunable Preset: General Enhancement! The first of it’s kind, this specialized preset is built on three setting into one. As you move along the slider, this will adjust the preset parameters automatically so that you can determine how light or aggressive you want to modify the preset. At default, the modifier is centered between the two extremes.

General Enhancement has the following properties that it’s addressing: sharpness, compression and detail. This preset can also be utilized as an up scaler.

Seek Render: Updates

We have made some additional refinements to the Seek Render tool.
Here are a few key updates:

  • Made several improvements to reduce crashes from the tool
  • Fixed out of sync timecodes
  • Fixed infinite processing status

Known Issues:

  • Videos that are non square pixels may have a slight resolution distortion in the viewer.
  • EXRs and Tiffs are not supported at this time.
  • Nyx XL is not recommended, due to long processing requirements.
  • Though SDR to HDR is supported, there is a little color discrepancy at this time.
  • Seek render does not support the Starlight models. If you have seek render enabled with a Starlight model, this will currently display “Processing” with nothing actually occurring.

UI: Updates

image
image2209×1039 109 KB

We have made some additional touchups to the UI.
Here are a few key updates:

  • Sizing alignment updates
  • Wider Right Panel
  • Presets are now collapsible

We look forward to your feedback! Have fun testing out 7.2.0.2b!

12 Likes
Topaz Video AI 7.2.0.3b (Studio Pre-release)
Topaz Studio | Transition Questions
2

Don’t forget to fill out the Beta feedback survey! Video AI Beta Feedback Survey 7.2.0.2.b

2 Likes
3

Some good-sounding stuff in here, especially the early AMD and Mac Starlight Mini testing! Since the quality isn’t there yet, I’ll hold off on putting my AMD 7900XT back in to try it, but I am curious to see how it eventually compares to my 5070TI. Nice to think I could throw it into another machine when my main one is tied up with a long render… :slightly_smiling_face:

Glad to hear about the Presets panel being collabsible, too!

And, darn, I’m away from home - won’t be able to try it for a couple days’ :face_with_diagonal_mouth:

1 Like
4

Could all the new models (starlight mini/sharp nyx XL) be added to the benchmark output if not already? so comparisons can be made?

3 Likes
5

Unfortunately, my aged Radeon Pro doesn’t have enough VRAM for SL Sharp. I tried running a regular Starlight and got an error on export. Log attached.
logsForSupport.zip (1.8 MB)

6

Seems like many models rhea xl, nyx xl, starlights, do not like videos that are made portrait using the metadata rotation flag, wont process.

7

odd, starlight-sharp works but starlight-mini seems to not change the video in this version

1 Like
8

[quote=“erasmo.topazlabs, post:1, topic:94995”]
For AMD users, can you output a result?
[/quote]Yes, I can! Thank you for making it possible!

Topaz Video AI Beta  v7.2.0.2.b
System Information
OS: Windows v11.24
CPU: AMD Ryzen 9 5950X 16-Core Processor              63.922 GB
GPU: AMD Radeon RX 7900 XTX  23.984 GB - Adrenalin 25.8.1

Using this input file: Apple WWDC 1997 Session 500 - Apple Hardware Roadmap that is a worst-case scenario with 240 x 176 and 8,5 Fps.

I get between 2,2 and 5,3 Fps doing a 4x Upscale with Starlight Mini. GPU sits tight at 100 % utilization and CPU is busy using only 1 thread at it’s maximum 4,5-4,7 GHz.

Seems very computational intense. GPU hits 2500 MHz at it’s stock configuration with 404 watts. Goes up to 2700-2800 MHz while allowing up to 470 Watts. No further attempts of undervolting or so. VRAM consumption is at 23,7 GB of 24 GB available.

Results seem fine. Will give some examples later.

3 Likes
9

I am not done with my first test rendering, but I thought exactly the same when looking at the “preview” (the provisory output file with numbers in it which can you display while the render still takes place), it feels that the quality of Starlight Mini has been degraded compared to the previous version.

1 Like
10

I am starting my tests, here first question for the TOPAZ team:

Starlight Mini has improved in speed. +40 to +70% (the speed changes more often than before too) on my A6000 PRO Blackwell;
Starlight Mini has divided almost by two its VRAM consumption (from 22-24 for a x2 upscale to around 15)

→ did you guys achieve this by switching from fp16 to fp8 or something ? My first render is not yet done, but as noted by jam700, Starlight Mini quality output seems degraded.

VRAM_Temp_SLM_x2
VRAM_Temp_SLM_x2696×329 7.16 KB

1 Like
11

Second question for the TOPAZ team related to the UI:

As previously mentioned, we lost the ‘codecs settings’ in the new BETA version. Can we please have it back, it is very useful to be able to change on the fly the video or the audio encoder as well as the output container, depending on what the purpose of the rendered video is.

I mean this tab:

image
image328×615 28.6 KB

7 Likes
12

Just for the record

Is this added compatibility going to result in less accuracy for Nvidia users?

13

Agreed. Moving the codec settings to the export page is less intuitive. I selected my preset, went to edit the output for a quick export and found it was missing.

6 Likes
14

Very Possible.

I have pointed out in the past that a speedup would result in a reduction in quality due to a reduction in accuracy.

If they don’t manage to raise the output to the same quality than before then my theory is confirmed.

2 Likes
15

That’s also my thoughts

I don’t want the degraded experience, this is the exact thing that happened to every other topaz model years ago when they distilled them down for compatibility sake.

I would RATHER it take longer for a better output.

4 Likes
16

Not worth it at all for degraded quality. I’m already going to spend a TON of time with starlight mini, even making it 2x-3x faster is NOT WORTH a reduction in quality.

Please for the love of god let the Nvidia people have the good models this time. Last time we abandoned CUDA it took years to get back to the quality we previously had

7 Likes
17

If you click on the export button rather than Quick export you can change the codec per clip. As I don’t often change the codec this seems ok to me.

18

Thanks for adding the ability to close the presets section.

19

100% agreed. My render is done now and I confirm the Starlight Mini quality (upscaling x2 an old VHS) is reduced compared to what it was before. It’s much more blurry, the “oh” effect of saving old videos is gone.

1 Like
20

Again, 100% agreed. It feels as us nvidia owners have lost “our” Starlight Mini in the process.

2 Likes