Blog Post

Linux and Open Source Blog
4 MIN READ

Announcing the Open-Source Release of ML Video Codec (MLVC)

Naba_Kumar's avatar
Naba_Kumar
Icon for Microsoft rankMicrosoft
Jul 24, 2026

We are excited to announce that ML Video Codec (MLVC) — a neural-network-based video codec developed at Microsoft — is being released to the open-source community under the MIT License.

  Video codecs compress video for transmission or storage, reducing bandwidth and storage requirements. MLVC is a modern machine-learning-based codec that uses substantially less bandwidth than con...
Updated Jul 24, 2026
Version 6.0
almalinux
azure linux
centos
debian
ebpf
linux on azure
monitoring
observability
oracle linux
red hat
BlueRaja's avatar
BlueRaja
Copper Contributor
Jul 25, 2026

This article doesn't answer many of my biggest questions

  • Is the codec able to handle arbitrary resolutions/framerates, or only those two? If so, how do the compression rates compare for more realistic resolutions?
  • What is the compute cost for decoding? What sort of hardware is required to decode in real time?
  • What is the size of the codec itself compared to existing solutions? 
  • If the codec is updated with new model weights in the future, will it still be able to decode video that was encoded using an older model? Or does this technology require keeping copies of every model ever on our PCs?
ClemCa's avatar
ClemCa
Copper Contributor
Jul 26, 2026

If the NVIDIA one is anything to go by, the embeddings will stay relatively consistent between versions, so it shouldn't lead to a lot of artifacts between different versions, but there's no guarantee.

Decoding is supposedly not very expensive, though we'll have to see in application.