Topaz Video 1.7.1

Hi All,

The team has worked on a few great updates and improvements!

See more details below:


1.7.1 Downloads

Windows | Mac
Please note, with Neuroserver, DMG versions are no longer available.


Improvements and Fixes

Features

  • SDR-HDR: Hyperion 2.5
  • Topaz Panel Update
  • Model Manager Update
  • First Start Neuroserver Model Manager
  • Stabilization Preview: Disable Option
  • Extra Metadata: Settings and Export
  • Add Watermark to Exports
  • Hardware Decode Options

Bug fixes:

  • Encoding/decoding errors
  • Thumbnail artifacts with .mov sources
  • LUT issue
  • Starlight Precise 2.6 processing speed fix

SDR-HDR: Hyperion 2.5

Hyperion 2 has been making a splash with it’s SDR to HDR enhancement that is being greatly enjoyed. Since then, we have been refining the model and now we have Hyperion 2.5. Hyperion 2.5 for SDR to HDR is our best version yet. This addresses the occasional ghosting-like artifacts and keeps the integrity of the original source intact. This update includes AMD/NVIDIA/MAC.

Model Recap

Hyperion 2.5, is a great update to our previous Hyperion 2, so many of the previous conditions are still relevant. This model expands your color and nit value range for VFX Compositing and for modern viewing on HDR supportive displays!

About

  • Converts 8-bit GenAI or archival video into HDR, high bit-depth media suitable for professional finishing pipelines.

  • Expands tonal range and color precision to provide significantly greater grading latitude in highlights, shadows, and midtones.

  • Prevents highlight and shadow clamping often seen in standard range and native GenAI video sources.

  • Expands existing color opportunities/steps for professional compositing and color manipulation workflows.

  • The additional color opportunities greatly reduces visible banding and compression artifacts common in 8 bit renders.

  • Has great details with text and textures even during high motion blur areas.

  • Export ProRes video or EXR (Pro) image sequences.

  • Support ACES color output with EXR (Pro) renders for better integration into industry-standard film and television pipelines.

  • This works best for medium to high quality content.

  • This model has some built in enhancements to revitalize and improve content.

  • This version is an easy “one click” solution. There are no parameters to adjust.

  • This is a 1x model. This is not an up-scaling model.

  • Max input and output resolution for this model is 3840x3840.

Technical things to know

Nvidia Requirements:

  • Requires an absolute minimum of 12GB VRAM with 32GB CPU RAM, though 16GB VRAM is recommended.
  • Targeted for GPU compute compatibility of 8.9. This is also the recommendation.
    • Please note, though there might be some success with a CC of 8.6 cards, this is not technically supported and some may encounter a less ideal experience.

AMD Requirements:

  • Requires an absolute minimum of 12GB VRAM with 32GB RAM, though 16GB VRAM is recommended.

Mac Requirements:

  • Requires an absolute minimum of 16GB though 24GB is recommended.
  • All Apple Silicon should be supported.

Outputs:

10-Bit ProRes (Cloud & Topaz Video Local Rendering)

  • ProRes Primaries: BT.2020 PQ
  • ProRes Transfer Function: BT.709

16-bit EXRs (Cloud Rendering)

  • EXR Image Sequence: BT.2020, Linear

16-bit EXRs (Topaz Video Local Rendering -PRO)

  • EXR Image Sequence: Aces 2065-1

Updated Topaz Panel for Premiere: Version 1.1.1

We’ve updated the Topaz Panel for Premiere with multiple fixes to enable an improved batch processing experience when enhancing from the Adobe Premiere timeline.

Simply drag video or image clips from your bin or timeline into the panel, choose an AI enhancement model, and upscale to HD, 4K, or even 8K. Better yet, this process does not take up system resources, allowing you to continue editing. Rendered files are automatically imported to your project and placed in timeline. Learn more here


Fixes & Improvements:

  • Improved the ability to render with handle media. Found in preferences, Add handles will add the desired handles to a clips in/outs when enabled. This is a useful feature for upscaling a timeline with transition effects; also enabling more editing, trimming capabilities after upscaling.

  • Fixed an issue where the panel did not treat sub-clips or clips from the same source as unique renders. Previously if portions of a clip was cut into a timeline OR user marked in/out on a clip, dragged to panel and repeated, the panel would only render one instance. Now the a user can drag multiple portions of a timeline that reference the same source clip and render each clip independently. This vastly improves the ability to enhance from a mastered source. Use scene detection to segment your clip, then choose an independent enhancement for every clip in a timeline.
  • Fixed an issue where a rendered clip is inserted into the wrong sequence after switching sequences. Now a user can render from the timeline and confidently switch to another timeline while the panel is rendering.
  • Fixed an issue where timeline clips would return missing 1-2 frames.
  • Fixed an issue where returned timeline clips shift right or get stacked in vertical video tracks.
  • Fixes an issue where the rendered file resolution did not always match the selected upscale preset setting in the panel.
  • Fixed an issue that would allow an unsupported output resolution for specific models. Now Precision models (Proteus, Iris, etc) properly support up to 8K, while Generative (Starlight) models support up to 4K.
  • Improved the panel’s ability to display accurate credit usage for cloud credits.

Known Issues / Limitations

  • Users may experience some unexpected behaviors when sending from timeline:
    • We recommend the timeline is set to match the source clip resolution and then changed after upscaling. This is not an issue when upscaling from a bin.
    • When enhancing from timeline, be aware of these settings:
      • Use In/Out range. When you have a clip that has been cut into the timeline, if this setting is off, the clip will render ALL the frames.
      • Add Handles (found in preferences). This setting will apply the desired handle media beyond the selected clip’s in/out range.
  • When enhancing clips from the timeline with audio, there is a known issue with audio potentially overwriting in the timeline. We recommend unlinking your audio and sending only the video tracks to the panel.
  • Vertical video. To upscale vertical video formats, choose 2x or 4x. The HD, 4K, and 8K presets will place your vertical video into a 16x9 aspect file.
  • Vertical video that is than 4k or larger will switch to VP9 codec and give an error when completed. To render vertical video larger than 4k, choose ProRes.
  • Slow-motion is a video only process and will not effect an audio track.

Additional Updates to Note!

Model Manager Update: Neuroserver Models

The Neuroserver Model Manager has been added below the traditional Model Manager. This has a built in hardware and version checker. The intention is to give clarity with the size of these models that need to be stored locally. Also, if you have installed neuroserver models that are not relevant for your next few projects, then this is a great place to uninstall them.

  • In the example above, there are four items already installed. There is nothing to download. Click on the “No, I’ll do it later” button.

  • If an installed model is unchecked, then you can remove that installed model.

  • If a new model is checked, then you have the option to install the model at this time.

  • Download progress will show in the upper screen right side of the app.

Please note that if your hardware does not meet the requirements or if there is a Pro-only model and your version is standard, then the model will be greyed out.

First Start Neuroserver Model Manager

When you open the app for the first time, you will be met with the Neuroserver Model Manager. This is to give you some transparency on the neuroserver models that you may want to download with clear indications of the space that these models will require. You can use this opportunity to install or uninstall a model. This step is optional. You can always manage your neuroserver installs and removals at a later time.

Stabilization Preview: Disable Option

The Stabilization Preview workflow may not be the right choice for your project. We have added the ability to turn the “stabilization analysis” off via Preferences. It can be found at the very bottom of the Advanced section. Be sure to “save” if you wish to retain any changes made.

Extra Metadata: Settings and Export

  • Project Setting source metadata have been updated significantly with the following: file name, file path, duration (time code), duration (frames), audio, pixel format, file type, resolution, aspect ratio and file size.

  • Export Meta data simply didn’t exist Aside from adjustments and Codec. This now includes output: name, destination path, duration (time code), duration (frames), audio, pixel format, file type and resolution


Add Watermark to Exports


You now have the option to add a Topaz watermark to renders. This is accessible in the Preferences menu under the Advanced category. Be sure to “save” if you intend to keep the change.


Hardware Decode Options

You can now decide to enable or disable hardware decoding. This can be beneficial for H.264, H.265 and AV1 processing. This can be done by going to the Preferences menu under the Application category. Then navigate down to the Processing section to find it. Be sure to “save” any changes you make.


Bug fixes

  • Fixed encoding/decoding errors with some 10-bit sources.
  • Addressed thumbnail artifacts of digital noise with .mov sources.
  • Resolved LUT issue with “colorspace error”.
  • Starlight Precise 2.6 processing speed fix, experienced by some, for inflated render times outside of model expectations.
  • Refined slower thumbnail generation.
  • Addressed the “Retry” button from restarting failed videos that have already gone through the stabilization analysis process.
  • Addressed neuroserver URL install issues experienced by some.
  • Fixed access issues for Starlight Mini on compatible AMD GPUs

Known Issues

  • Updating to Mac OS 27 will cause issues with Topaz Video. The team is working on a solution for the recent OS release.
  • Some may experience Starlight Sharp “freezing” during export. The team is working on a solution for this.

Don’t Forget

Is your system compatible?
Click Here

Documentation
Click Here

Encountered an Issue?

  • If you experience any issues, please ensure that your drivers are up to date. Next to a proper restart or an app reinstall, this is usually the first step to ensure you have the correct drivers running for this application.
  • Be sure to do a full reinstall if you are having issues making the upgrade from a previous version.
  • Sometimes a firewall can prevent the app from starting. If this occurs, be sure to add Topaz on the exception list.

Otherwise, you can easily reach out to our fantastic Support Team: Click Here


Enjoy the new updates!

9 Likes

Just wanted to express thanks for the wonderful 2.6 model. This is stunning.

13 Likes

Does this obsolete 1.7.1.2b, or is there anything in that beta that we need to keep looking at?

Thank you for mentioning the compatibility issue with macOS 27. A workaround is to remove the audio from the video file to avoid crashes when you connect/turn on a monitor.

THANK YOU it works again!

2 Likes

This release has all of the patches from the beta included.

1 Like

Once again, please have (or at least have the option to have) program startup to the Projects screen instead of automatically loading the most recent project.

I hope Hyperion 2.5 will work on RTX 3xxx, too, as any purchase of RTX 4090 or 4080 is out of the question right now… :slight_smile:

Meme drawing as an animation (very bad quality of the OG clip) looks great with SLP 2.6. :sparkles:


The preemptive “Running Stabilization Pass” the second stabilization is selected continues to be an abomination. It makes it impossible to efficiently queue up multiple conversions quickly for unattended processing. I don’t understand why this suddenly showed up in 1.7 and continues into 1.7.1.

And stability on MacOS continues to be as bad as 1.7.0 was for the most part. Any time I queue up more than a couple of videos, it crashes while I’m gone. Without exception.

1 Like

Any improvements with SLP ghosting?

5 Likes

Has anyone notice a difference in StarlightPrecise 2.6 in this version. Seems darker than in the 1.7.02b beta version.

In the source metadata of Project Settings, I hope to add “frame rate” information. In addition, when Topaz Video processes video frame interpolation for variable frame rate (VFR) videos, the exported frame rate is inconsistent with the setting (the actual exported frame rate is approximately the set value / 2.5).

An error(?) with Starlight and a few notes and suggestions.

Firstly, I am very impressed with Starlight 2.6, it does very well ,especially with VHS footage and other older viorecordings like old DVD movies and such, but it does have a recurring problem ,which I’ll get back to shortly.

Firstly, though, when choosing the “Output resolution” when using Starlight. It simply won’t do more than 4K. Like if I have a FullHD 1920x1080 video, I can choose 2x upscale no problem, but if I choose 3x upscale or 4x upscale, it still only says and outputs 4K (which is 2x upscale).
Am I missing some information? I’m guessing it doesn’t do more than 4K?

Okay. Now to the recurring problem:
Whenever I use Starlight (2.6) all the people get sort of Benjamin Buttoned. Any imporfection in (and sometimes shadow lines on) someone’s skin gets Starlight-interpreted as scars or deep wrinkles.
A 20 year old will end up looking ling a hybrid between a 20 year old and a 70 year old.
I’ve added a photo of this. A photo I feel shows one of the better performances by Starlight.

It would be great to be able to adjust certain perameters like “Face improvement” and “Blur improvements” and such.

Also I think what would really make this product amazing, would be the ability to upload higher quality photos of a person’s face and tell the program, that THIS is that person. A sort of tracking system for persons faces and the ability to link a more high res photo to it, so that it knows what it’s supposed to actually look like.
This would ESPECAILLY be a game changer when upscaling old VHS video and such.

I’ve done this with photo enhancements via AI, where I’d give it a close-up photo of the person’s face and tell it to use it as a detail-reference, and that has yielded insane results.

9 Likes

I canceled my subscription to Topaz Video due to issues mainly associated with facial restorations. My tests were all done using local processing and the results outside of the face problems were really good. Allowing a photo for each of multiple people in a video would be needed, and for use with Starlight Mini and Precise 2.6 (often did well on faces but the failures were unacceptable) for those who process locally.

EDIT: Maybe allow two pictures of person: front view and side view. It’s easy for me to make work for other people no matter how difficult the task might be… :slight_smile:

The new Starlight Precise 2.6 is a godsend! I’m remastering a 1950s B&W TV show that was 480x360. After a lot of experimentation (and failures with Starlight Mini), I’ve come up with the following workflow:

  1. Linear upscale in FCP by 3X to 1080p.
  2. Enhance with Starlight Precise at 1X.
  3. Linear upscale in FCP by 2X to 4K.

This process seems to eliminate the upscaling steps in Topaz Video that introduce random hallucinations and plasticization. And the result is indistinguishable from a modern 35mm B&W print.

2 Likes

This is almost what I was targeting (not looking for 4K upscale)! I found 70 episodes of the “Topper” series at archive.org and worked on the first episode of season one for testing Topaz - the video is 494x360. I did not get FCP. (just looked up - only for MAC? I have Windows).

I did not do a good job looking to find non-facial issues that might have occurred but was “wowed” by how well the items I saw were improved. It was the facial issues that resulted in canceling my subscription.

After running short tests where no problems were seen, I ran the full 22min epsiode through both Mini and Precise: Mini took 10hrs6mins while Precise took 10hrs58mins. I decided I could live with that as the machine could be dedicated to the task. When I watched the two full length upscaled videos, I saw hallucinations in areas my short tests had not used.

EDIT: Went to Apple website - FCP is Apple only :slightly_frowning_face:

And frame interpolation with the scene detection option turned on does not work properly, and still does not work properly, even though Topaz had a few weeks to fix it.
Is anyone even testing this app?

I have installed 1.7.0 and try to update to 1.7.1 but does not work .. I’m on the lates Win11 Pro - 25H2

The Starlight models are all limited to a maximum resolution of 4K, so this is not a bug. Honestly, I even think this limitation makes sense, as the hardware requirements at this resolution are really massive.

Regarding your recurring problem:

The issue occurs when too much sharpness is used with the SLP model. Since version 1.7, there has been a slider that lets you adjust the strength from 1 to 3. By default, the slider is set to 3, which in my opinion is far too high and only somewhat suitable for already very high-quality 720p or 1080p material.

I now use the SLP sharpness only at level 1, and with some footage I still notice an ugly oversharpening, which is why I would honestly like an even wider adjustment range.

Another thing is that SLP, or the Starlight models in general, are not “finish” models. I always use another model afterward, such as Proteus, to give the material the final polish and to remove additional (over)sharpening.
With grain and noise, the material immediately looks much more natural and better again.

3 Likes

If your hardware allows it, you should definitely use the x4 mode for 360p footage. Your issue with faces will improve significantly as a result. In my experience, the x3 mode only really becomes useful starting from 480p material.

In my tests, I have found that all Starlight models benefit greatly from a higher upscaling factor, which is especially noticeable in faces and fine details.

For anyone who wants to try this themselves, you should take 720p footage and upscale it once with a factor of x2, and then compare it to a 480p video upscaled with a factor of x3. You can easily downscale the 720p video to 480p using HandBrake.

But the same applies here as well: the Starlight models are not “finish” models. They only help bring very poor source material up to a reasonable level, and then it should be refined further using another model. Grain and noise in particular are very important to avoid plastic-looking final results.

1 Like