I noticed the same.
Before the Starlight Mini download was always around 7gb.
Now for the new beta the download was only around 3,5gb; half of before.
There are no information, what has changed.
I noticed the same.
Before the Starlight Mini download was always around 7gb.
Now for the new beta the download was only around 3,5gb; half of before.
There are no information, what has changed.
Yeah, weāre working on this one ![]()
Thanks for the feedback - yeah weāre tracking these items. We may have a fix for the rotated videos shortly
We are definitely working on speedups for Starlight Mini and also availability on other platforms - but itās an extremely difficult task which is taking some time.
Please be patient - and keep it respectful on the forums ![]()
They will make $0 money from me. Iām not using the cloud. EVER. I wouldnāt use it if it were free.
Thatās not something I was expecting - we will check on this.
Did you download the previous beta? If not, that could be what caused this.
We understand the cloud isnāt for everyone. Video AI will continue to have local rendering as the focus. Much of our cloud work is moving to our web products such as Astra.
However it is way easier to deploy new AI models to the cloud vs local hardware. So the cloud may be getting some of the newer models first, or it may be available only on some platforms (e.g. NVIDIA Windows first).
Thanks for the updates.
We (all) waited for ![]()
From my point of view Topaz Labs leaves the way to do the improvement work on our local hardware under own control, if upcoming models are released for the cloud.
There are meanwhile āhundretsā of online AI providers for video and photo improvement.
I never load my private stuff/footage into a cloud and pay for too high costs per minute.
Thatās great to hear and thank you for the response. I look forward to future updates.
For anyone expecting miracle speed boosts for Starlight mini locally - donāt get your hopes up. From what I gather, SLM basically uses stable diffusion techniques on every frame, and can only run so fast on consumer hardware.
Unless you have a NVIDIA RTX PRO 6000 with 96GB of VRAM, significant speed boosts aināt likely happening.
BTW, anyone interested? ![]()
To be fair, that card can be purchased for $8,300. Certainly not cheap but a $4500 difference would be the cost of an extra 48gb RTX Pro 5000.
Probably USD vs CAD
With 96 GB you could probably load the full Starlight model into memory, no need for SLM
Or you could wait for this with 128 GB of Coherent Unified System Memory for one third of the price:
Nice!
NVIDIA DGX Spark Memory Bandwidth = 273 GB/s
RTX Pro 6000 Memory Bandwidth = 1.79 TB/s
Cheaper yes, but Iād rather have the Pro 6000.
or China mods
NVIDIA's GeForce RTX 4090 With 96GB VRAM Reportedly Exists; The GPU May Enter Mass Production Soon, Targeting AI Workloads
Iāll take one! ![]()
48GB variant is out already on ebay and elswehre, 96gb I have not seen yet but could follow soon. I was close to buy the 48Gb variant but it was to expensive to me I was not willing to pay 3000$ for a refubished card without warranty claims. The seller rejected my offer of 2500$ then i ordered the 5090 ![]()
Ya too risky for no warranty - my 4090 died after one year and I had to use the warranty!