@noahmorris Built an aggregator version back in 2022.
Kept it for private and uni students use, though discord was not added at the time.
Then this guy came forward and sold a basic app for 50mil
https://t.co/PyyUwyJt8o
@vidrush This is what I've been telling the moat is. The direction of video, the editing decisions. If editor experience can be reliably put into the "director" module as a skill where it's deciding what goes where , for how long in what way, that's what is the secret to be kept
@primalrobin Why would do it local when you could do it in cloud, 10x the video quantity, price: free to ~$ 0.20 per 15min video (for cloud runs, e2e)
Exact same quality, almost unsupervised
P.s: for anyone reading, can't say more than that ๐ค
@eknoor01 I'd be interested for this thing. If these are not templates then deciding on num of assets reliably, their placement, etc is a tricky thing.
I tried a FSM based layout and then created tens of templates on top of it.
Lacks creativity in the end. Meh at best
Gemini watching your videos out of the box or you fixing a hundred things for cheap models to reliably look and respond based on frames.
The added advantage would be to include Jeb to predict the subject of the scene itself which you then feed to the "watcher".
@AaryanPrateekX Online marriage BGV as a service
You give the email/handle and they do the rest
Website: https://t.co/EVrc1IteNT
(see what I did there ?)
@ajeetbharti@airtelindia I had similar problem with JioFiber
Did a deep dive into app layer, turns out they probably have china supplier for the router, standard build which sends all the data to Jio cloud.
Then put a sniffer into the router and now I have all the info needed (and more)
@sonerabay I'd first define concrete characteristics of what is "natural", maybe an ever growing list.
Then study OG face channels that do things from studio, prepare notes.
Finally, work with SOTA models on it and try replicating.
Might just try it this weekend (rip my wallet ๐ข).
@sonerabay Looks really great !
Did a similar thing for a Yoder channel
Imo next low hanging fruit is doing R&D around how to make these appear more natural. I'm sure team has worked at some levels on it.
Had some success at making them a bit more natural but just the talking faces seem ๐ค