Topaz Video AI 6.1


Today’s release is a very important moment in the history of Topaz Video AI. We are releasing Project Starlight, the first-ever diffusion model for video enhancement.

This is a significant evolution over our existing models, and it’s the single largest increase in model capability since Video AI launched.

Even the most challenging footage can be enhanced with Project Starlight. So please, try out the “bad” footage—footage that may have previously produced artifacts with other models.

Because of its size, this model is significantly slower and more expensive to run than previous models, and requires cloud processing on server-grade graphics hardware. But this is just the beginning. Our mission as a product and research organization is to advance visual quality to the furthest extent possible—it just takes a little more processing power to get there right now.

We see a clear path towards expanding access to Starlight and other models of this generation to very high end desktop GPUs, and Project Starlight is just the start of a new series of models.

6.1


Project Starlight Research Preview

Active Video AI users can currently process three 10-second weekly previews of Starlight for free. Results will be viewable through unlisted, shareable links with the new “Compare” button in Video AI.

These previews take about 20 minutes to render if servers are available immediately, and all exports are delivered at 1080p.


Starlight Early Access

Video AI users can also render up to 9000 frames of video using Starlight (5 minutes at 30 fps).

The current introductory under-cost pricing of 90 credits/minute will change over time as our team expands access and server capacity.


Our research team is eager to receive feedback and discuss this step-change in AI video upscaling. We appreciate your involvement on this journey to Video AI’s next frontier.


Project Starlight announcement on X

Topaz Labs CEO Eric Yang has also written a new blog post in our Roadmap section to talk more about Project Starlight.


Want to learn more about Enterprise deployments for Starlight and other models? Our new page details different options for bringing Video AI to your team.


Project Starlight is also available in the Video AI web app.

8 Likes

We’d love to see some Starlight examples shared here using the new Compare button too, here are some of our favorites:

3 Likes

If it’s 90 credits per minute, how much is 90 credits going to cost?

Does the fps affect the cost of credits as it’s processing double the amount of frames but the same duration ?

And since you mentioned the introductory price is going to change, does that mean it will cost more or less later on?

One last question, if Starlink is done through cloud, why do we need to upgrade the software ? Can’t we use the feature through the web browser?

3 Likes

Credit pricing (for all products that use Cloud Credits) is available on this page

No - higher framerate videos will allow for less time to be processed. For example, you can process up to 10 seconds at 30 fps, but 12.5 seconds at 24 fps.

Less :slight_smile:

You’re 100% correct! Try out our new web app.

2 Likes

Someone else said in the previous version with 1200 to 1500 credits you can only process 6.4min video W T F. No body is paying $100 to process one 45min video mega ripoff S M H. There are people on Fiverr that will upscale a full 2hrs to 2 1/2 hr movie for $10 no idea of there quality at that price it is worth it to take a chance.

17 Likes

So, an overpriced cloud service that I will never use on a desktop app that I’ve 1.) already paid the purchase price on and 2.) already paid the annual renewal for; despite this, I’m stuck on a much earlier version, because basic functionality (e.g., the always-promised, never-fixed VFR bug) is lacking in the newer versions. This is extremely frustrating.

34 Likes

This is the key takeaway we want desktop users to understand about Project Starlight:

Because of its size, this model is significantly slower and more expensive to run than previous models, and requires cloud processing on server-grade graphics hardware. But this is just the beginning. Our mission as a product and research organization is to advance visual quality to the furthest extent possible—it just takes a little more processing power to get there right now.

We see a clear path towards expanding access to Starlight and other models of this generation to very high end desktop GPUs, and Project Starlight is just the start of a new series of models.

4 Likes

Also re: VFR bug – we have multiple confirmed cases where VFR output is solved, but there are still some file types and specific hardware/setting combinations causing this issue.

Definitely recommend contacting support (help@topazlabs.com) for more support with VFR outputs.

3 Likes

As mentioned, we are planning to reduce pricing as quickly as server capacity allows:

The current introductory under-cost pricing of 90 credits/minute will change over time as our team expands access and server capacity.

2 Likes

Another Sony RX10 IV 960 fps shot comparing original, one processed through Artemis LQ and one processed through Starlight with a second pass of Proteus with Dehalo at 60 to tone down the over sharpening/crunchiness. Aliasing/Moire be gone (or massively reduced)! Starlight reduced a lot more of the problematic aliasing/moire that Artemis LQ could not.

7 Likes

Starlight is also highly capable of improving detail and sharpness in GenAI videos:

4 Likes

In multiple threads with in the past 24 hours alone, I see user reports of this bug continuing to be an issue for them. Whether it’s with some settings or all of them, I’m not willing to budge from v.5.3.6 until it’s universally fixed with all settings. I’ve already wasted a lot of hours troubleshooting newer versions of TVAI and re-encoding video with v.5.3.6. This is a real disconnect. I could care less about any new feature until this deal-breaker is fixed.

10 Likes

How about giving users what we used to have as part of the standard Video AI package… Multi-GPU capabilities? Probably more profitable for topaz to charge for cloud rendering I guess. Less important for users to simply use the advanced hardware they already have without over charging for “pro” version.

16 Likes

With Starlight, both Personal and Pro license users will have the same level of access and processing speed.

For our other models, we made the decision when launching Pro to improve multi-GPU for large deployments and focus on single GPU performance on the desktop. This shift is mirrored in the decline of SLI, CrossFire, and other methods of using multiple physical cards for processing.

That’s right, and we have worked with several users to identify possible causes.

Some users have even reported this issue in 5.3.6, so it’s something that has been around to varying degrees for some time. We’re adding more and more ways to prevent this with each update since 6.0

Yeah I’m gonna check to see if I can get some samples of some older VHS stuff I have and see what it looks like with Starlight since I’ve mainly been showing Sony RX10 IV slowmo footage.

Yes, the model handled those interlaced looking lines in front of the car really well. :+1:

1 Like

Another example from our Starlight web page:

Tony you think it would be possible in future to have an alternative to the three 10 seconds weekly preview renders with Starlight? I was thinking maybe the option of having a one 30 second weekly Starlight preview render.

why not let users run this model on their own machine? even if the compute costs are massive locally, this is a desktop app we paid to use locally, not a SaaS web app.

17 Likes