OSSDrop
TrendingExploreForum
Drop a toolSign in

Built something open-source?

Add it to the list — free and open to every open-source project.

Drop your tool
OSSDrop
BrowseBlogForumHow to submitHow it worksFAQAboutContactThe list on GitHubr/OSSDrop
AI & Coding AgentsDeveloper Tools & CLIWeb & APIsData & DatabasesDevOps & Self-HostedFiles & SyncSecurity & PrivacyProductivityNotes & KnowledgeCommunication & SocialAutomation & IoTFinance & BusinessScience & EducationCreator / MediaAll categories →
© 2026 OSSDropFree · open source · community-curated

Blog · Published August 4, 2026 · 6 min read

MLVC: Microsoft's Open-Source Learned Video Codec, Explained

MLVC (Multi-platform Learned Video Codec) is Microsoft Research's open-source neural video codec — a deep-learning approach to video compression, released free under the permissive MIT license on GitHub. Instead of the hand-engineered transforms and motion estimation used by traditional codecs like H.264, HEVC, and AV1, a learned codec trains neural networks end-to-end to compress and reconstruct video. MLVC's distinguishing focus — per its name and paper — is real-world, multi-platform deployment, a practical goal since many learned codecs stay research-only and are hard to run outside a lab.

This is a research-grade project, not a packaged, drop-in codec: you work in Python with deep-learning tooling rather than installing a binary. But it's a credible, openly-licensed starting point from a major lab, with the method documented in an arXiv paper. Below: what MLVC is, how learned video codecs work, honest pros and cons, and its live GitHub star count pulled from OSSDrop.

#ToolBest forLicenseStars
1MLVCresearchers and engineers prototyping neural video compressionMIT279

How we picked

Everything on this page is verified, not guessed. We confirmed MLVC's license (MIT), repository (github.com/microsoft/mlvc), and activity directly from GitHub, and its star count is pulled live through OSSDrop, so the number is current. We describe MLVC by what its authors state — an open-source, multi-platform learned video codec aimed at real-world deployment — and we are deliberate about what we don't claim: no benchmark numbers, compression ratios, or quality comparisons, because those depend on your content, hardware, and settings and should be measured against your own baseline. Where we contrast learned codecs with H.264/HEVC/AV1, we describe the general, well-established difference in approach, not a specific head-to-head result. No placement is paid; MLVC is listed because it is a notable, openly-licensed project, and we feature tools from every source fairly.

1

MLVC

Best for researchers and engineers prototyping neural video compression

MIT · 279 stars · Creator / Media

MLVC (Multi-platform Learned Video Codec) is Microsoft Research's take on neural video compression. Where traditional codecs such as H.264, HEVC, and AV1 rely on decades of hand-engineered transforms and motion estimation, a learned codec trains neural networks end-to-end to compress and reconstruct video — an approach that can capture patterns hand-designed pipelines miss. MLVC's stated angle is real-world, multi-platform deployment, which matters because a lot of learned-codec research never leaves the lab. It ships under the permissive MIT license, with the code on GitHub and the method described in an arXiv paper. Set expectations accordingly: this is a research-grade project, so you'll need Python and deep-learning tooling and some familiarity with video pipelines rather than a one-line install or a packaged codec. If you're studying or prototyping neural video compression, it's a credible, openly-licensed base from a major lab.

Key features

  • ✓End-to-end learned (neural) video compression
  • ✓Explicit focus on multi-platform, real-world deployment
  • ✓Open source under the permissive MIT license
  • ✓Method documented in an arXiv research paper
  • ✓Python / deep-learning implementation

The bottom line

Learned video codecs are one of the more active frontiers in open-source media, and MLVC is a rare example aimed squarely at getting neural compression out of the lab and onto real platforms — with the code openly available under MIT. If you're researching or prototyping neural video compression, it's worth a look. Explore the project on OSSDrop, browse more Creator / Media tools, or drop your own open-source tool.

Frequently asked questions

What is MLVC?

MLVC (Multi-platform Learned Video Codec) is an open-source, MIT-licensed neural video codec from Microsoft Research. It uses deep learning to compress and reconstruct video, with a stated focus on real-world, multi-platform deployment. The code is on GitHub at microsoft/mlvc.

Is MLVC open source and free?

Yes. MLVC is released under the MIT license on GitHub, so it is free to use, modify, and redistribute, including in commercial projects, subject to the license terms.

How is a learned video codec different from H.264, HEVC, or AV1?

Traditional codecs like H.264, HEVC, and AV1 use hand-designed transforms and motion estimation. A learned codec instead trains neural networks end-to-end on video data to perform compression and reconstruction, which can capture patterns hand-engineered codecs miss — at the cost of heavier computation, especially when encoding.

Can I use MLVC in production?

MLVC is a research project, so treat it as a starting point rather than a drop-in codec. Running it requires Python and deep-learning tooling, and you should evaluate quality and compute cost against your own baseline and hardware before relying on it.

Who made MLVC?

Microsoft Research. The implementation is published on GitHub (microsoft/mlvc) and the method is described in a paper on arXiv.

Install

git clone https://github.com/microsoft/mlvc
cd mlvc
# then follow the repo README for Python dependencies and model setup

Pros

  • +Permissive MIT license — free to use, modify, and build on
  • +Backed by Microsoft Research, with an accompanying paper
  • +Aims at real-world, multi-platform deployment, not just lab results
  • +Actively maintained on GitHub

Cons

  • –Research-stage — not a drop-in library or packaged codec
  • –Requires Python and deep-learning expertise to run
  • –No prebuilt binaries or one-line install
  • –Learned codecs are compute-heavy, especially at encode time
Visit arxiv.orgSourceFull details on OSSDrop