Since the latest update, using the starlight sharp model, it just hangs up mid process. Timer keeps running, but nothing happens. In the logs it is showing at the denoising part each time.
Can you share the logs with the support team so that we can review what is going on and have the devs investigate? help@topazlabs.com
To gather logs, please select Help > Logging > Get Logs for Support and attach the zip file to your reply.
Here is a video to help with the steps of how to collect the logs.
same thing happens here whenever try to upscale to resolution above 1080p. But it works if upscale to 1080p.
I am trying to upscale videos that are in 720p. There is no option for 1080p thru starlight sharp. Minimum is still 720p and the next option is 2x which makes it 1440p.
This worked completely fine numerous times, up until I updated the app and models yesterday.
This has been the case for me ever since the introduction of neuroserver. Starlight āMiniā, āHQā and āPreciseā all run fine, but āSharpā and āFastā both fail with different symptoms. Starlight Sharp starts to run the neuroserver which loads the model and starts processing, but stalls before ever creating an output file. The GPU Bus Inteface is busy but nothing is ever processed.
Starlight Fast, on the other hand, never gets that far. It loads the model and then throws an inference error.
I reported both problems as soon as I tried release 1.4 and supplied the Topaz Video log along with the Direct-X profile file. Since then, I have tried 1.5 and 1.6 with no change in this issue and roll back to 1.3.
Donāt hold your breath waiting for a fix. If you need to run Starlight Sharp, install release 1.3.
Cāmon Kyleā¦how many users with this problem have to send you logs before the devs can figure this out. It has been broken for three releases. I have tried dozens of attempted workarounds, including installing a fresh Windows 11 25H2 PC with NOTHING but the NVidia Studio driver and Topax Video. EXACTLY the same result everytime.
Doesnt work for me upscaling to 1080p. Iām doing 540 to 1080 and it dies the same way. Iāve tried smaller and larger staring resolutions, 1:1, 1:2 and 1:3 upscales. Iāve tried MKV and MP4 input and output containers. Iāve tried RTX5070Ti and RTX5080 GPUs. Iāve tried NVME drives and SATA drives as source and/or target. Iāve tried my main PC and a completely clean Windows installation. They all fail the same way.
I think this is the first report of the problem in early April:
I understand, the more logs and information we at the support level have to pass onto the devs helps with the situations. However, they still have to sort out the cause and how to fix it with the research team. We are also seeing various different situations in logs from different systems and use cases. Not every stall or error are the exact same.
Trying to do what we can from the support side to help the devs be able to replicate these situations and errors for a resolution to be found.
Iām retired from being a product manager of corporate data warehousing and master data management products. I totally get the problem of trying to reproduce the problem. I volunteer to do anything I can to help solve this issue, since it is forcing me to have two PCs available - one for Starlight Sharp (running 1.3) and one running the latest release. I can give them procdump files, watch the processing in procmon and report, collect windows event logs, ANYTHING to help them fix this problem.
Obviously, this worked on some QA machine or else you wouldnāt have released it (at least I hope thatās the case). How about detailing the hardware on which all the models work properly? Iād hate to believe that Topaz never tested each model in each release at least once before release.
Yes we are having basically the same issue. I was also wondering how come not more people reported this issue since the release of 1.4.0 - maybe not so many are using Starlight Sharpā¦
So far the workaround for me as well is to downgrade to 1.3 till the issue is being resolved with a patch. It is not ideal but at least it works.
For now I decided to start rendering my videos with Starlight Precise 2.5 first as I found this is the best method for my 512x384 and 640x480 SD videos and then once again do the second step in Starlight Sharp. I am preparing the material for the second pass right now, hopefully a patch will arrive by June.
You are correct. I have a wrong impression that it works. But I just tested it again, it doesnāt in latest 1.6.0 
My RTX 5060 Ti 16GB works fine with most Topaz AI models. However, when running Starlight Sharp, the process freezes halfway. The GPU clock speed, utilization and power usage all drop to the minimum. What causes this issue?By the way, Iām running version 1.6.0.
By the way, Iām running version 1.6.0.
This is a known problem. I canāt run Sharp under 1.6 either. Release 1.3 will work for Starlight Sharp, but doesnāt support Starlight Precise.
According to many āgeniusesā (like this one Starlight Precise 2.5? More like "Slowlight" Precise 2.5. Oh wait, it's actually "Slowtrash" 2.5, yes precisely - #18 by sdfgsdgssdgdt363fg ), what you experienced is supposed to be the case, and you should only use Starlight models for clips no longer than 3 minutes long, ONE AT A TIME. As they said, Starlight has nothing wrong with it; only the āwrongā users who do not know how to use it āproperlyā.
Depends on how long you can leave your computer tied up. I run Starlight upscales that are 45 to 90 minutes or more long. They take 3 or 4 days, but they work.
Not true.
Iāve used both SL mini and SL precise 2.5 on long clips (1:30 to 2 hours) without issues.
Other than the REALLY long encoding time, of course.
P.S. the hardware here is a RTX 5070TI with i9 12900 and 32GB system RAM - and a Mac Studio M2 Ultra with 64GB.
Not sure if anyone else has reported this, but it seems that the models that donāt run properly under release 1.6, hang or throw an error because the ffprobe job to count frames never finishes.
It has not been reported to my knowledge, I will share that with the team. They are working on testing a resolution for the Sharp model stalling at this time. Should have something ready in the near future.
