https://livekit.io logo
Join Slack
Powered by
This message was deleted.
# helpdesk
s
This message was deleted.
d
👋 What type of feature are you thinking about?
c
We need very low latency VOD-like streaming for 1:1 content generated on demand by our backend in response to queries from individual users
So it's "live" in the sense that the content doesn't exist yet, we are generating it in real time
Algorithmically, so it is generated by code not a camera feed or recording
The user hits "play" and we need to get them our dynamically generated frames in <5 seconds
But we need VOD playback semantics. Their playback needs to start at the very first frame, not wherever the stream converges
And we ultimately need pause/rewind capability
Mux takes 30+ seconds to start a new HLS stream
We can't keep the user waiting that long
d
yeah for sure. I'm familiar with the use case. The VOD/rewind requirements is actually pretty similar to a feature we are working on.
the latency side is no problem with LiveKit, typical end to end latency is <300ms, even across geographic regions
c
Yes, and you're the only platform that seems to consider that the source might actually be a machine
Every other WebRTC service is P2P only
LiveKit solves the latency, but we would have to synchronize the first frame to the user ourselves
d
what do you mean by synchronizing the first frame? are you saying that you'd only want to start sending when the user has subscribed to the track?
c
Yes, precisely. We are custom generating content for a user so we can't serve it to them incomplete
d
for sure, we are working with a number of genAI companies.. and I'm happy to share techniques for that synchronization
c
Please do, we are also using generative AI so we're in the same problem space
d
yeah! so what folks do is to use our data channels to signal when the user is ready to consume the frames. i.e. when 
TrackSubscribed
is fired on the human user's client, make a call to 
LocalParticipant.sendData(...)
(you'd want to do this several times for redundancy) Then backend could start sending frames when it receives that handshake. We are also building an "Agent framework" to making it easier for load balancing work between these backend workers.. If that's interesting to you, feel free to DM me. I can get you guys early access
Open in Slack
PreviousNext
Powered by