I tried upscaling a full movie using Starlight Mini for the first time. Short renders went fine. Kicked off a 6 day render (that’s how long it takes my 3080 to finish an hour and half of film). Got 4 days in and it failed! I had checked in on the rendering a few times, including last night and it was going smoothly, I even checked the preview on the export queue and everything was looking great.
I was really excited about Starlight mini, but now I’m just disappointed with Topaz Video AI in general. I’ve wasted so much money and resources on this export (4 days of leaving my office Air Conditioner on and my PC running at full steam). I paid for Topaz for stability and ease of use, if the app is going to fail after 4 days of a 6-day job with no resume option, I don’t know what I’m paying for anymore.
Seeing an out of memory error as processing an entire film is likely too resource intensive to pass it in one go. I recommend breaking it up into smaller chunks (~5 mins or less).
Could be it’s about special characters in the file names. What I also see in the log is that the audio track is not readable. Break video into pieces at least two here makes sense. What you also can do is making copies from the running output video, then not everything is lost when it fails ..and do not disturb much the system. I recently killed my 3day job as another program used my gpu to much.
This happens to me like 50% of the time. Even the 10 minute chunk I did the other day. I left it running overnight and it still hadn’t finished “loading model”. Reboot and tried again and it finished “loading model” this time, but as reported above it still failed after 23 hours or so…
I do see there is an update today to Topz 7.0.1 that includes two Starlight Mini fixes including “loading model” hanging and downloader reliability.
Overall, it seems Starlight mini needs to go back into Beta.
When I broke up the files into 10 minute segments I used this for file names: Suzane.19950XX.mkv where XX went from 00 to 09. Still failed near the end of exporting. Not sure why the audio should matter, Topaz shouldn’t be doing anything to it. And I always just mux the original source video and audio together with the AI upscaled video and leave off the attached audio track to the AI upscaled video anyways. I guess maybe Topaz AI has a bug with audio, so I might as well rip the audio out before rendering anything. So much guesswork. I’m just giving up for a while. My 6 day export, I ran nothing else the entire time. I was on vacation for 4 of those days. So it wasn’t competing for GPU. I even disabled 2 of my 3 monitors to reduce the strain on my GPU.
If you don’t have the bug in question, don’t bother chiming in with “works for me”. Also no need to gaslight people by saying something is “impossible” and then showing that it worked. You are the only person that said it’s “impossible”, so you’re only fooling yourself.
Did you even try what was suggested (lowering VRAM usage in the prefs)?
The endless „loading model“ issue was in many cases due to VRAM usage set too high and also Margaux suggested this to be a out of RAM error, …
And the part with the „impossible“ video was meant as “impossible to save as it’s that bad quality“, just to show that also really long renders can go through.
In any case I‘d suggest exporting to PNG, so you don’t lose progress after an error and can continue from where it failed. Or, copy the temporary output file from time to time if you haven’t enough space for PNG.
I did not try your suggestion. I guess I need to make it clearer:
I’m not willing to try multiday long exports just to be an unpaid beta tester for a product I paid for. I am willing to troubleshoot Topaz devs if they have questions, but they should already have everything they need from my logs.
Not your fault jo.vo. I appreciate you trying to help!
I broke up a short film into 5 minute chunks. Processed one overnight. Woke up and Starlight Mini failed. It’s never worked for me. Is there an official support system? I’m wasting so much time, both my time and GPU/CPU time.
I had an 3090-24GB, this gpu also often failed with SLm and at the same time my 4070 (i have two pcs) had no problems, but sometimes on the 3090 system my renderings went trough simillar you have now, so works, but then not, strange …overall the card performed very badly with SLm + my gpu temps were constantly 85°C, the cooling solution of 30x cards are on a lower level as it’s today …so in short, it could be your gpu.
Tried lowering max memory. Doesn’t work. If it did work, wouldn’t it be the application’s job to figure out that memory is maxed out and to scale back? To a certain extent, the operating system handles that for you.
I have a 3080, if it’s not a supported card, then Topaz should just mark it as such.
I tried another render yesterday. Nearly 1 full day in and it gave me a “Recoverable error” but when I resumed, it lost 1 days worth of render time. lol. I guess it’s recovering, by starting over. Not sure what the point of that is. At this rate a 1 hour video will finish in 400 years.
Hi, I can’t say how good supported 3090 is today, because I don’t have my 3090 anymore. My test were with first release of Starlight. Play around with gpu driver, also try game ready driver (even Topaz says Studio is the right one), try a different Topaz release and output try other output formats.
So it wasn’t competing for GPU. I even disabled 2 of my 3 monitors to reduce the strain on my GPU.
