Low-latency HLS without throwing away your player
Standard HLS trades latency for robustness: six- to ten-second segments mean a viewer is always 20–30 seconds behind live. For a lecture that's fine. For an auction, a sports feed or interactive content, it isn't.
Chunked CMAF, not a new protocol
Low-latency HLS gets there by delivering partial segments — CMAF chunks — over HTTP chunked transfer, so the player can start rendering a segment before it's fully written. Crucially, it degrades gracefully: a player that doesn't understand the low-latency extensions just plays the normal segments.
Glass-to-glass latency drops to roughly 3 seconds, without forcing every viewer onto a bespoke player.
Turning it on
POST /v1/jobs
{ "input": "...", "outputs": ["hls"], "latency": "low" }
The trade-offs are real: shorter chunks mean more requests and slightly less coding efficiency, and your CDN has to support chunked transfer end to end. We enable it per asset so you pay that cost only where it matters.
What we tuned
The defaults that worked best in testing: 4-second segments, 1-second parts, and a hold-back of three parts. Enough buffer to survive a jittery mobile connection, tight enough that "live" feels live.