This message was deleted.

It’s certainly possible, but complex and more compute-intensive. You’d need to decompress, decode, process/modify, reencode, then recompress each stream you’re interested in. An easier path for the specific use case you mention is a client-side processor which grayscales the video frames before they’re transmitted to the SFU.

Unfortunately in the use case that I am solving, the clients do not do any processing on their side (cameras). I am aware of the extra compute requirements per stream and we have figured out that it is acceptable for current stage. Any pointers you can give where the decoding/processing/reencoding should happen?

cc @fancy-wire-61616 @dry-elephant-14928 @polite-kilobyte-67570

@ripe-lifeguard-63895 I guess the camera is running LK client and directly talking to the LK server. Given the processing that would be required. a different service might be the best option. We have an upcoming RTMP-in based service that is designed to take RTMP input and bridge it to WebRTC and distribute the streams via LK server. Adding some processing pipelines to that service might be another option, but we will have to think through that. Do you have the option of using such a service?

You can also ingest via WebRTC with Pion, then have a server process that receives and processes the streams on your backend. once processed, you can use our Go SDK to publish to LiveKit for distribution

@fancy-wire-61616 Yes, that would be an option.

@ripe-lifeguard-63895 just saw a post you wrote on HN! Really cool stuff. 🙂 Is this what you were building when this thread started? https://news.ycombinator.com/item?id=34221287