Any news about new Starlight model?
2 Likes
We can only hope. Last time there was months of silence, then starlight dropped out of nowhere. Would be nice to get a better model at some point in the near future.
1 Like
Well, Iâm not a negative person, but for me the Topaz Video AI project is about to be buried; absolutely nothing has happened for over six months.
5 Likes
They have certainly killed off my enthusiasm for this product with the lack of updates and no more beta releases. I used to visit these forums every day, now maybe once a week at most just to check if there has been a release or not.
I donât care about Topaz anymore and will not be renewing any products, unless there is a shift back towards development of the desktop products. I donât care for their cloud offerings.
17 Likes
I completely agree with you.
2 Likes
Their new subscription model instantaneously disenfranchised a large segment of their customer base, namely the home user and hobbyist. Same thing happened to Adobe when they went to cloud based subscription model, but at least with Creative Suite you are allowed up to 99 days of offline use before you are required to call home.
2 Likes
Do you know how to control the batch/chunk size? Thanks.
Strange is they also donât really care to their cloud product. There are still aspect ratio problems and you get the output h264 CFR23 compressed.
1 Like
Add to that those still quite long processing times of cloud rendering and the crazy restriction to 5 minute chunks - which makes the product too clunky for me to be usable (similar as with SeedVR2).
I do not care that much if the encode goes for 2 days or one week - but Iâm not willing to split a 2 hour video into 5 minute chunks - or even less, depending on where you find a fitting scene cut in the original video where you actually can reasonably split the source. And still wqith such an approach you will get continuity issues.
So: why canât you simply upload a whole video at once? As long as this is not possible, thereâs no way Iâll use that service. Not to speak about the still quite insane costs of the cloud rendering.
3 Likes
2 Likes
Well, there IS a BETA section for TopazVideo here.
But until now it has been empty - which is quite a pity as there seems to be little to no testing on anything other than NVidia at Topaz themselves. Otherwise releases as those 1.0.3 and 1.0.4 would be hardly possible with the totally broken Starlight on AppleSilicon.
Also, an issue like the non-working user presets button (on Windows and Mac) wouldâve been found really early if there had been a Beta.
The only parameter I know of that might affect batch/chunk size is --max-gpu-mem. I set it to 31 because I have an RTX 5090.
FYI, there is one other useful parameter that does not appear in my example. You can tell runner.exe to run ffmpeg preprocessing filters prior to running the SM filter. For example:
âffmpeg-preproc-filters âscale=854:480,setsar=1â
Would scale each frame to 854x480 and set the pixel aspect ratio to 1:1 prior to passing the frame to SM. I believe you should be able to use any of the filters that are complied into the Topaz version of the ffmpeg executable.
1 Like
Is SeedVR2 also limited to 5min?
Itâs great that you can select different codec settings with the --ffmpeg-encoding parameter, which arenât otherwise available in the GUI.
By the way, why are there no new Beta versions? The last one was on August 28th, almost three months ago.
Betas only matter if the feedback received is acted on. I have submitted many bug reports and feature suggestions for past betas, but the issues do not get fixed in the final release.
3 Likes
If you execute runner.exe with no command line arguments, it displays the syntax:
usage: runner [-h] [âstart-frame-idx START_FRAME_IDX]
[âend-frame-idx END_FRAME_IDX] [âupscale-factor UPSCALE_FACTOR]
[âffmpeg-encoding FFMPEG_ENCODING] [âdevice DEVICE] [âmax-gpu-mem MAX_GPU_MEM]
[âffmpeg-preproc-filters FFMPEG_PREPROC_FILTERS]
input_path output_path
That, I donât know. Iâm just a Topaz user, like you.
Itâs a good question, though.
This is not fixed, you can select longer videos.
But from what Iâve read itâs the way around OOM errors that seem to often occur otherwise.
Donât have first hand experience here, though.
1 Like
Iâve had 1.0.4 working Ok , exporting a 4x upscale SLM on my Win11 desktop here (exporting to ProRes422 HQ)
However, I just tried exporting to image sequence, TIFF 8 bit, and it didnât work. The in [ and out ] were set 10 minutes apart so it was going to be a long render, but it said it finished after only one minute.
I looked in the export folder, and there were no files, so it didnât export anything.
Anyone else finding this bug in TIFF export in 1.0.4?
1 Like
Does having two matching GPUs improve Starlight Mini, either by speed or quality? I was thinking of getting another card with a pcie 4.0 extension cable etc
Thatâs not going to help much unless youâre paying for a Pro subscription. Thatâs required for multi-GPU support.
And a question comes to mind: Do you have the power supply capacity to run two GPUs?
1 Like
