I’d say this is a pretty good example, but as you said it really excels with even lower quality inputs
Do u know the processing duration?
The ‘censored’ cigarette is hilarious, btw.
1 Like
Some 15-30min or so for the 10s clip.
Can close the app once it’s been uploaded and started. It will get processed by their cloud service either way once its started. No progress will be lost (unlike local pause) when shutting down the app. It’ll still be there either still processing or ready when you open up TVAi again.
1 Like
I did send a portion of the video to test, but I’m yet to see any results - should I check the mail? 
Scratch that - so that’s how it shows the progress:
Yes, this is something we’ll improve.
We will at least pop open the Cloud exports panel when sending out a render, and probably have email notifications live before the next build.
2 Likes
Definitely understand all of your thoughts regarding local/cloud processing and charging for credits. We’re working on it, but in the meantime wanted to give some context about why we’re releasing it cloud only:
8 Likes
This would have been a good post to start this topic off with.
5 Likes
Oh man. If this is how resource intensive this processing is on a HPC data center, I don’t expect it to be viable on a desktop for at least 10 years…
3 Likes
The process is definitely slow on this model - this is going to take time to develop, I bet (imagining SDXL img2img frame rendering for 1920x1080 resolution).
1 Like
Definitely. Better late than never 
3 Likes
We have line of sight to speeding it up by 10x at least
7 Likes
I hope offline models will get a bit of love, too.
3 Likes
It’s still day 1, we just have a beta post and a roadmap update with the greater vision re: Starlight
Oh, so you can only test Starlight if you have an active Subscription?
Well, then I’m out for this Beta (since other things didn’t change).
1 Like
While trying to not make any promises I may regret: I can confidently say it will be less than 10 years 
4 Likes
Free access for non-active licenses coming in the next round. Server capacity is being monitored and expanded as we’re able to
2 Likes
Rest assured, it is definitely not the end of the story for local processing in Video AI
3 Likes
So, if 10s take 20min. Then 1 hour of video will take 6 * 60 * 20 = 7200 minutes or 120 hours = 5 days
With the mentioned theoretical 10x speed increase 1 hour video would still take 12 hours.
There will be a big need for more optimizations, like frame interpolation where u generate in-between frames instead of upscaling the original frames. Only possible during ‘low change’ scenes ofc.
Yes, quite likely that they share technology between Gigapixel and TVAI for this I would assume.
