Can we see the original, please?
ok, I just calculated all the little pieces of my whole sequence that I had cut out. So far, out of a total of +/- 54 minutes, Iāve already turned in 33 minutes 42. So the total render time with StarLight is 89h57 and 23 seconds (3 days, 17h57 and 23 seconds) on an RTX 3070. With 720x576 to 1280x1024 resolution files
I still have 10 files to return, for a total of 20 minutes 37.
It still takes a long time hahaha.
Looks great in parts but these typical AI patterns in the vegetation makes star light really unusable for anything with nature imho. If they can fix that, it would be another tremendous step forward.
Thea Detail sharpness 30% looks good also
Since a lot of people are noticing text, etc⦠being changed, I started looking closer at some of the exports that I did and yes⦠I noticed a woman with glasses in a sequence and in the next sequence she magically improved her vision by wearing no glasses,!
Pretty crazy! ![]()
I have an 81 frame video clip @ 1280x730 and Iām using starlight mini to upsize it to 2560x1440 and it seems to truncate at only 64 frames. Iām running local on a rtx 6000 ada, it finishes in about 7 or 8 minutes.
i meant to say 1280x720
Same here. I am by no means poor, but I already have difficulty justifying spending $3,600 on an RTX 5090 (a card which is effectively only ca. 37% faster than my RTX 4090), let alone $7,000 - $10,000 on a RTX A6000 with 48G. So, this is simply not happening for me.
I am mildly annoyed with Topaz for introducing models on an alleged consumer product that require such insanely expensive hardware; but am primarily ticked off at the excessive greed of nVidia. As far as Iām concerned, they can put it where the sun donāt shine.
So a little less than 2 days/episode if my math is correct. Thatāll take about 315 days to do the entire series.
Luckily, for DS9, there are very few episodes with a lot of vegetation (maybe 10%).
What would that amount to in server render cost?
ā¦and thatās 24/7!
oh⦠my neck bahaha! ![]()
If you find 10 people to distribute the encoding work itās one month, so not that unrealistic - at least not for a popular series like that.
I priced up a no frills render only machineā¦.. with the idea of just using it and starlight mini render all this old family home video I have at 640 by 480:
i5 13600
32 gigs DDR5
Z790 mobo
1TB m.2 SSD
5070ti vid card
Gold 850 PSU
Win 11 Home
No Frill case
AIO cooler
And was coming in less than $2k
I could probably go to a less expensive chip set motherboardā¦.. and even a lesser CPU since it appears CPU has nothing to do with the render speed.
But do you need a high-grade CPU at all for Topaz Video AI? I thought that the new āviewerā was CPU related.
And then thereās the additional license cost to put topaz on an additional machine.
I already have it on my main PC and my main Mac.
When they finally release the Mac version of starlight mini, itās going to be even slower than the PC version I would guess.
The idea is to have a balanced rig⦠NO bottlenecks!
But we only need Starlight for the first two seasons. After that, they got a bigger budget and used better lighting. ![]()
Iām looking at 4 months for Evangelion. Iām on episode 2, and just had a GPU crash after 4 days rendering -_-;
Each episode (22-25 minutes) takes 6 days, 18 hours to process through starlight-mini.
For the older models, the CPU most definitely matters. I saw a nice bump in performance going from a 12700k to a 13900k. Also I noticed that AMDās Zen 3 CPUs tend to be outperformed by Intelās 12th-14th gen. Having said that, SL Mini appears to be more GPU bound than previous models. Given the real world price gap between a 5070ti and a 5080, it might be difficult to make up that difference by downgrading the CPU and motherboard.
With a resolution bump or no resolution bump?
Can we please have the program give you an export error at the BEGINNING of an export if it detects an unsupported file type? I just wasted an hour trying to export a starlight video and at the very end it says oops this file is not supported and crashed. Iād rather know that at the BEGINNING of an export, not at the end.