We are excited to announce that we have reworked authentication to make using the app a more pleasant experience. We have relaxed restrictions on offline app usage to match our other apps, allowing up to 30 days (or until the end of your subscription) of offline usage after authentication.
Other fixes:
fixed incorrect “Invalid username or password” issues.
prevented seeing unexpected paywalls popping up while the app is open.
reduced delay for seeing credits show up for cloud exports.
Added support for the denoise filter in the cloud.
Added more error messaging for out of memory issues.
Resolved odd timeline behavior for image sequences.
loosened restrictions for encoding to Material Exchange Format (.mxf).
Yet another useless update for my system. Renders are slower and crash after 12min. Havent had a stable update where I could do long renders or do batch processing since version 1.0.4. Least I got something for paying to renew my license this year. But this is a joke. I cant try any new models or features I’ve been looking forward to.
So, for almost a year now we’ve been waiting for Starlight Mini to be optimized for more speed, but to this day they haven’t done it, at least not for the desktop version. I’m now absolutely certain that they’re deliberately not doing it so that people will use the cloud version; it’s simply all about profit, nothing else. We’re standing at the bus stop, waiting for months for the bus that will never come again. For me, this app is history.
Ran Starlight Mini (locally) on a small 360p video of ~29s (644 frames) in 2032s, 3.15s per frame on M4 Max 14 Core CPU 32 Core GPU with macOS Tahoe 26.4 beta. This is consistent with previous runs.
One positive difference is the export queue preview works much better, the # of fps is (almost) correct, it varied from 0.3fps and 0.8 fps during the run. I was able to see the frames in real time (it seemed to get stuck on the last frame, but was able to navigate easily to previous frames)
On asitop, CPU P-cores was ~50% utilized, GPU 100%, 50W power consumption, ANE not used (neural engine). RAM was about 23-28GB out of 36GB, so quite comfortable.
Would love to see an optimized Starlight Mini (without sacrificing quality).
An optimized starlight mini would be nice, and if you look at ai progress, it improves over time. I really doubt starlight mini is the best upscaler that will ever be released. My dream would be if they can optimize starlight mini for faster speed without sacrificing quality and also introduce a new model that beats starlight mini and starlight sharp in temporal consistency and quality. You could even call it starlight mini 2 .
If you look at the image/video generation space, better tech is coming out at least every 3-4 months. It seems a little strange that nothing better has been put out after starlight. I prefer the look of starlight sharp, but it doesn’t have the same quality as starlight mini. Starlight mini is just better at reproducing a complex image.
If you are having crashes occur, can you pull the logs and send them along with your system profile report to the support team so we can try and see what is going on? help@topazlabs.com
I am holding out hope they are working on another revolutionary upscaler that leverages current AI tech but with temporal consistency which is their bread and butter. Upscaling images with something like Qwen Image Edit is currently fairly mindblowing - it is now only a matter of time before someone cracks the image/video interpretation and consistency needed to do successful video processing, even adhering to and following user prompts. This is coming one way or the other, and probably sooner than we think. I am still hoping it will be Topaz that really makes it happen.
.
