Gaussian Splats as a Video Codec?

For most of digital media history, compression has been treated as an engineering problem. It is better understood as an industry control point.

Every major media transition creates a data problem. DVD needed a way to fit movies onto optical discs. Blu-ray needed higher-quality HD playback. Streaming needed video that could survive consumer broadband, mobile networks, and global distribution. Digital cinema cameras needed high-resolution image capture without unmanageable storage. Each generation answered with new codecs, new standards, new workflows, and eventually, new licensing structures.

That pattern is now moving into volumetric media.

Point clouds, meshes, neural representations, and Gaussian splats may feel new to the startup world, but the business mechanics around compression are not new. The media industry has been refining those mechanics for decades.

The media industry has always been codec-driven

By the late 1990s and early 2000s, digital video had already become a codec economy.

DVD helped establish MPEG-2 as a mass-market video format. Blu-ray expanded the stack with MPEG-2, MPEG-4 AVC/H.264, and SMPTE VC-1, all of which Sony lists as supported Blu-ray Disc video codecs. MPEG itself describes its work broadly as developing standards for the coded representation of digital audio, video, 3D graphics, and other data. (Sony)

The camera and post-production world developed its own parallel codec culture. Apple ProRes became a dominant professional editing format, with Apple describing it as a codec family designed for multistream, real-time editing performance, high image quality, and reduced storage rates. Sony developed XAVC for professional cinema and television workflows, including 4K and HD production. Panasonic’s AVC-ULTRA family includes AVC-Intra, AVC-LongG, and AVC-Proxy codecs based on H.264/AVC. RED built REDCODE RAW, also known as R3D, as its proprietary compressed RAW video codec. (Apple Support)

These codecs solved real technical problems. They made content recordable, editable, distributable, and playable. But they also became commercial infrastructure. Once a codec becomes central to a workflow, it affects who can build products, who can ship devices, who can distribute content, who can interoperate, and who may eventually owe licensing fees.

That is the part of the codec story that startups often underappreciate.

Patents are not incidental to codecs

The media industry does not treat compression as a neutral utility. It treats compression as intellectual property.

That is why codec history is also a history of patent pools and licensing administrators. MPEG LA, one of the most recognized names in codec licensing, is now part of Via Licensing Alliance. Via describes itself as a patent-pool administrator that aggregates patents around key innovations and offers licensing solutions for standardized technologies. Its current licensing programs include AVC/H.264, VC-1, MPEG-2, MPEG-4 Visual, and HEVC/VVC. (Via Licensing Alliance)

Blu-ray has its own licensing structure. One-Blue was established as a one-stop-shop license for essential patents covering Blu-ray Disc, DVD, and CD products. Premier BD, formerly tied to the BD4C licensing program, also administers Blu-ray-related patent licensing. (One Blue, LLC)

The broader codec licensing ecosystem includes Access Advance, which administers HEVC and VVC patent pools, and Sisvel, which manages licensing programs around VP9 and AV1 patents. Access Advance says its HEVC Advance pool licenses patents essential to the HEVC standard under fair, reasonable, and non-discriminatory terms, while Sisvel lists its VP9/AV1 video coding platform as a patent licensing program with current licensees. (Access Advance)

This is not just old history. Codec licensing disputes remain active. In November 2025, Reuters reported that InterDigital sued Amazon over video-compression, HDR, and streaming-efficiency patents. In February 2026, ASUS publicly stated that it had temporarily suspended its German website and online store after a Munich court ruling in a Nokia patent dispute affecting selected products using HEVC. Streaming Media also reported in March 2026 that Via’s new AVC/H.264 streaming license fee structure for new licenses beginning in 2026 replaced the old single-cap model with a tiered system that can reach $4.5 million annually for very large OTT platforms. (Reuters)

That is the important lesson for volumetric media: when compression becomes commercially necessary, IP pressure follows.

Volumetric media is entering the same machinery

Volumetric video is often discussed as if it sits outside traditional media infrastructure. It does not.

A volumetric system still has to solve the same basic problems as video: representation, compression, storage, synchronization, transmission, decoding, rendering, device support, and interoperability. The difference is that volumetric media may include geometry, attributes, motion, view dependence, lighting behavior, topology, and scene structure. That makes the compression problem larger, not smaller.

The standards bodies have already recognized this.

MPEG’s Visual Volumetric Video-based Coding work, including V3C and V-PCC, targets lossy compression of 3D point clouds, including geometry and attributes, scalable or progressive coding, sequences captured over time, and random access to subsets of a point cloud. ISO also describes ISO/IEC 23090-5 as specifying syntax, semantics, and decoding for visual volumetric media using video-based coding methods. (mpeg.org)

MPEG’s Geometry-based Point Cloud Compression standard, ISO/IEC 23090-9, specifies geometry-based point cloud compression. MPEG also has ongoing work on video-based dynamic mesh coding, with MPEG-I Part 29 focused on encoding dynamic meshes with time-changing attributes. (ISO)

The telecom ecosystem is also engaged. 5G-MAG, the Media Connectivity Association, describes itself as an international nonprofit industry association focused on standards and open-source software for connected media experiences. Its standards materials include volumetric video experiences using MPEG V3C, including ISO/IEC 23090-5 for V3C and V-PCC and ISO/IEC 23090-10 for carriage of V3C data. (5G MEDIA ACTION GROUP)

3GPP is now tracking beyond-2D media directly. Its Series 26 specification list includes TR 26.956, “Evaluation and Characterization of Beyond 2D Video Formats and Codecs,” and TR 26.958, “Study on 3D Gaussian splats for mobile.” (3GPP)

This is the signal. Volumetric media is not escaping the codec economy. It is being pulled into it.

The familiar codec players are already in the room

The organizations around this space are not random.

InterDigital describes its research and innovation work as contributing to global wireless, video, telecom, AI, streaming, and XR standards, with a Video Lab focused on production quality, efficient compression, streaming, and immersive experiences. Fraunhofer HHI describes HEVC/H.265 as a widely deployed ITU-T and ISO/IEC video-compression standard and VVC/H.266 as the most recent international video-compression standard from ITU-T and ISO/IEC. (InterDigital)

Ofinno describes its Advanced Media Lab as contributing to MPEG and JVET, including next-generation video coding, 3D point cloud compression, 3D mesh compression, and 3D Gaussian splatting. Nokia, Ericsson, and Fraunhofer HHI also announced a 2025 collaboration around next-generation video coding standardization for the 6G era, explicitly connecting future video codecs to mobile and immersive media experiences. (Ofinno)

These organizations understand the codec economy. They understand how standards, patents, licensing, and product dependency interact. They know that compression is not only a way to reduce bandwidth. It can also be a way to build standards-essential patent portfolios and long-term licensing leverage.

That is the context startups need to understand before assuming that a compression breakthrough is only a technical moat.

The startup compression trap

Volumetric startups are understandably drawn to compression.

Without compression, volumetric media is difficult to store, transmit, and render at scale. Bandwidth, latency, storage cost, mobile power consumption, GPU load, and cloud compute all become commercial barriers. So the pitch becomes familiar: lighter assets, faster streaming, higher fidelity, lower bandwidth, better compression.

There is nothing wrong with that pitch. The problem is assuming that compression is only a product feature.

In media, compression is where IP concentrates.

Large companies and research labs often have the resources to file patent families across multiple jurisdictions, participate in standards meetings, contribute technical proposals, track IPR disclosures, and shape terminology around future essential claims. Startups often do not. They may build quickly, publish demos, ship SDKs, and win early customers without a serious freedom-to-operate strategy.

That creates an asymmetry. One side may be building product traction. The other may be building claim coverage.

Both can be happening at the same time.

Gaussian splats changed the urgency

Gaussian splats have accelerated this entire discussion.

The original 3D Gaussian Splatting work, “3D Gaussian Splatting for Real-Time Radiance Field Rendering,” presented a method that starts from sparse points produced during camera calibration, represents scenes with 3D Gaussians, optimizes anisotropic covariance, and uses a fast visibility-aware rendering algorithm to support real-time rendering. Inria later described the method as enabling real-time rendering of photorealistic scenes learned from images, with reported rendering above 100 frames per second in its discussion of the breakthrough. (Repo SAM)

That combination made Gaussian splats attractive very quickly. They are visually compelling, relatively fast to render, and easier for many teams to reason about than fully implicit neural representations. They also sit close enough to point clouds, radiance fields, game-engine assets, and video pipelines that many different industries can project their own product strategy onto them.

Game engines see assets. Capture companies see scans. XR companies see telepresence. E-commerce companies see product visualization. Mapping companies see spatial reconstruction. Streaming companies see a delivery problem.

Codec companies see a compression frontier.

The standards world is already responding. In February 2026, Khronos announced a release candidate for KHR_gaussian_splatting, a glTF 2.0 extension for storing 3D Gaussian splats. The draft extension defines 3D Gaussian splats using position, rotation, scale, opacity, and spherical harmonics stored as attributes on a point primitive. (khronos.org)

MPEG is also moving. At its 153rd meeting, MPEG advanced video-based Gaussian Splat Coding through an amendment to ISO/IEC 23090-5 V-PCC, using V-PCC mechanisms to map Gaussian splat attributes into video representations suitable for compression by standard video codecs. MPEG also lists Gaussian splat coding as Part 45 in its explorations work, with documents including common test conditions, draft use cases, draft requirements, and lightweight GSC requirements. (mpeg.org)

This is no longer just a research trend. It is becoming standards activity.

Open repositories do not equal patent clearance

The industry needs to be more precise about what “open” means.

A public GitHub repository is not the same thing as freedom to operate. A research implementation is not the same thing as a commercial patent license. A file-format extension is not the same thing as a representation being free of patent risk.

The official Gaussian Splatting GitHub repository says it contains the authors’ implementation of “3D Gaussian Splatting for Real-Time Radiance Field Rendering.” Its license materials state that Inria and the Max Planck Institut for Informatik hold ownership rights in the software and that the licensor’s goal is to allow the research community to use, test, and evaluate the software. Other files in the repository state that the software is free for non-commercial research and evaluation use under the license terms. (GitHub)

That does not mean every Gaussian splat implementation has the same software license. It also does not answer the full patent question. Software copyright, repository licenses, standards contributions, and patent rights are related but distinct issues.

The dangerous shortcut is saying:

“Gaussian splats are open, so a commercial compression and streaming business around splats is safe.”

That is not a serious IP position.

The patent-publication window matters now

There is also a timing issue.

In the United States, patent applications are generally published after 18 months from the earliest relevant filing date, subject to exceptions. The USPTO’s Manual of Patent Examining Procedure states that, with certain exceptions, nonprovisional utility and plant applications are published after 18 months from the earliest filing date for which benefit is sought. (USPTO)

That matters because Gaussian splats entered broad industry awareness in 2023. If companies filed patent applications around splat compression, streaming, mobile rendering, level-of-detail selection, spherical harmonic reduction, asset packaging, dynamic updates, or hybrid video-plus-splat delivery during that early commercial wave, some applications may already be public or may become public soon, depending on filing strategy and exceptions.

There may also be earlier filings around related concepts: point-based rendering, anisotropic splatting, radiance-field compression, point-cloud video coding, neural rendering, view-dependent attributes, and hybrid geometry-video systems.

This is why casual patent searching is a bad operating model. Product teams should not run ad hoc searches, skim abstracts, and treat the result as business guidance. A patent search without counsel can create false confidence, miss relevant claims, and produce a record of awareness without a competent legal analysis behind it.

The responsible path is counsel-led freedom-to-operate work, an internal IP strategy, and a standards strategy.

Where the Gaussian splat risk will concentrate

The most important Gaussian splat patent risk is unlikely to be the broad idea of “using splats” in isolation. The more likely risk zones are the implementation layers that become commercially necessary.

Those include:

• splat quantization

• attribute compression

• spherical harmonic compression

• progressive transmission

• spatial tiling

• level-of-detail switching

• mobile decode

• GPU memory layout

• view-dependent streaming

• dynamic splat updates

• real-time sorting

• browser playback

• cloud rendering

• hybrid video-and-splat transport

• asset packaging

• codec integration

• 
  

Those are exactly the kinds of areas where codec history says patents accumulate.

A company does not need to own the entire concept of Gaussian splatting to create commercial pressure. It may only need claims around one indispensable piece of a production pipeline.

Standards participation is not bureaucracy

This is why standards participation matters.

Standards participation gives companies visibility into how a representation is being formalized, which companies are contributing, which constraints matter, what implementation assumptions are forming, and where licensing issues may emerge. It also creates a disciplined record of technical contributions and helps companies avoid building products around assumptions that the market later rejects.

Participation does not eliminate risk. But non-participation leaves a company reacting to decisions made elsewhere.

That is not a strong position for any company building the future of spatial media.

Our position on Gaussian splats

We are not against Gaussian splat research.

We are not against experimentation, academic work, open-source tools, or new scene representations.

We are against building a commercial product whose core dependency is Gaussian splat compression, streaming, or delivery while treating the technology as automatically patent-free because public repositories and demos exist.

That is not how the media industry works.

The history of codecs shows that once a compression technique becomes useful enough to scale, it becomes valuable enough to patent, license, standardize, and enforce. That was true for legacy video. It was true for modern streaming. It is becoming true for volumetric media.

Gaussian splats may become a major representation for real-time 3D scenes. They may become part of e-commerce, digital twins, XR, spatial video, telepresence, gaming, mapping, simulation, and mobile media. If that happens, the codec economy will arrive with them.

The companies that understand this may not be the loudest in the market. They may not be the ones releasing the flashiest demos. They may be filing, contributing to standards, refining claims, and waiting for commercial dependency to build.

That is why our recommendation is simple: do not build a Gaussian-splat-based commercial product without an IP strategy, a standards strategy, and a licensing-risk strategy.

Compression is not just a performance feature.

In media, compression is often where the business model begins.

Why the Volumetric Format Association matters

This is the role the Volumetric Format Association was created to play. Other standards bodies are studying pieces of the volumetric problem, including point clouds, meshes, Gaussian splats, mobile delivery, and 3D asset representation. The VFA is focused on something more complete: treating volumetric video as a new media format with an end-to-end framework for capture, compression, server-side streaming, playback, decoding, rendering, and interoperability. Its Volumetric Player Format Specification defines how a volumetric playback application connects to a backend server so that volumetric data can be streamed, how the stream is organized, and how the player communicates with the server. The same public specification summary also describes HTTP streaming from server to device, trick-play requirements for volumetric players, and mesh- and patch-based compression standards. (VFA Website)

That distinction matters. A new media format cannot scale if every company ships its own closed capture format, its own compression method, its own server protocol, and its own playback stack. The industry needs interoperability, but it also needs a fair licensing environment that respects the IP behind the server-side specification, the compression technologies, and the player specification. The VFA’s founding materials state that membership allows companies to share intellectual property within the organization while protecting the value of that IP, and the association was launched to build end-to-end specifications across capture acquisition, interchange of data, decode and render, and persistent metadata. (LAPPG)

That is why VFA participation is not a side issue in the codec conversation. It is one of the most important ways the volumetric industry can avoid repeating the worst parts of the traditional codec economy. The goal should not be a market where every promising volumetric representation becomes a fragmented licensing fight. The goal should be a new media format with clear specifications, fair access, responsible IP treatment, and enough interoperability for creators, platforms, device makers, and technology companies to build real businesses. The VFA was launched by major companies across the ecosystem, including Verizon, ZEISS, RED Digital Cinema, Unity, Intel, NVIDIA, MediaPro and Canon, with the stated goal of ensuring interoperability across volumetric video. (TV Tech)

For this industry to move beyond demos, volumetric video needs more than another compression technique. It needs a complete media-format strategy. That means streaming specifications, player specifications, compression support, licensing policies, and standards participation that are designed for a real ecosystem, not just a single product cycle. That is the future the Volumetric Format Association is working toward, and it is the reason this discussion should not be framed as Gaussian splats versus everything else. The larger question is whether volumetric media becomes a fragmented collection of proprietary pipelines or a scalable media format that the whole industry can build on.

---

Books in the thumbnail include and are worth reading:

• Programming HD DVD and Blu-ray Disc Authors: Michael Zink, Phil Starner, Bill Foote Publisher: McGraw-Hill Publication date: 2007 ISBN: 978-0071496704 / 007149670X Amazon: Programming HD DVD and Blu-ray Disc Source notes: Amazon lists the title and authors; B&H lists McGraw-Hill, ISBN, and a September 10, 2007 publication date.

• Blu-ray Disc Demystified Authors: Jim Taylor , Michael Zink, Charles Crawford, Chris Armbrust  Publisher: McGraw-Hill Publication date: November 3, 2008 ISBN: 978-0071590921 / 0071590927 Amazon: Blu-ray Disc Demystified Source notes: Amazon lists the authors and publication date; Barnes & Noble confirms the publisher and ISBN.

• Compression for Great Video and Audio: Master Tips and Common Sense, Second Edition Author: Ben Waggoner  Publisher: Focal Press / Routledge Publication date: November 2009 ISBN: 978-0240812137 / 0240812131 Amazon: Compression for Great Video and Audio Source notes: Amazon lists the second edition, publisher, and ISBN; ACM lists the November 2009 publication window.

• Interactive TV Standards: A Guide to MHP, OCAP, and JavaTV Authors: Steven Morris, Anthony Smith-Chaigneau Publisher: Focal Press / Elsevier Publication date: 2005 ISBN: 978-0240806662 / 0240806662 Amazon: Interactive TV Standards Source notes: Amazon listings confirm the title and authors; library records confirm Focal Press/Elsevier and 2005 publication.

• Video Encoding by the Numbers: Eliminate the Guesswork from Your Streaming Video Author: Jan Ozer Publisher: Doceo Publishing Publication date: January 2, 2017 ISBN: 978-0998453002 / 0998453005 Amazon: Video Encoding by the Numbers Source notes: Library and Amazon-market listings confirm Jan Ozer, Doceo Publishing, ISBN, and 2017 publication.

• QuickTime for .NET and COM Developers Author: John Cromie Publisher: Morgan Kaufmann Publication date: January 31, 2006 ISBN: 978-0127745756 / 0127745750 Amazon: QuickTime for .NET and COM Developers Source notes: Amazon and VitalSource confirm the title, author, publisher, and ISBN; Barnes & Noble lists the January 31, 2006 publication date.

• Rendering with Radiance: The Art and Science of Lighting Visualization Authors: Gregory Ward Larson, Rob Shakespeare Publisher: Morgan Kaufmann Publication date: March 15, 1998 ISBN: 978-1558604995 / 1558604995 Amazon: Rendering with Radiance Source notes: Amazon lists the Morgan Kaufmann edition and publication date; ACM confirms the January/March 1998 publication period and publisher.

• Time-of-Flight and Structured Light Depth Cameras: Technology and Applications Authors: Pietro Zanuttigh, Giulio Marin, Carlo Dal Mutto, Fabio Dominio, Ludovico Minto, Guido Maria Cortelazzo Publisher: Springer Publication date: May 24, 2016 ISBN: 978-3319309736 / 3319309730 Amazon: Time-of-Flight and Structured Light Depth Cameras Source notes: Amazon lists the Kindle edition; Google Books confirms the full author list, Springer, May 24, 2016, and ISBN.

• Making Things See: 3D Vision with Kinect, Processing, Arduino, and MakerBot Author: Greg Borenstein Publisher: Make / O’Reilly Publication date: March 6, 2012 ISBN: 978-1449307073 / 1449307078 Amazon: Making Things See Source notes: Amazon lists the publication date; O’Reilly and Processing references confirm the author, publisher family, and book details.

• Rendering With Radiance: The Art and Science of Lighting Visualization Authors: Greg Ward Larson, Rob Shakespeare Publisher: Space & Light / later reprint listing Publication date: August 1, 2007 on the Amazon listing ISBN: 978-0974538105 / 0974538108 Amazon: Rendering With Radiance, 2007 listing Source notes: The thumbnail appears to show a second copy/edition of the Radiance book. Amazon describes this ISBN as a replacement for the out-of-print first edition.

• High-Definition DVD Handbook: Producing for HD-DVD and Blu-Ray Disc Authors: Mark R. Johnson, Charles Crawford, Chris Armbrust Publisher: McGraw-Hill Publication date: April 2007 ISBN: 978-0071485852 / 0071485856 Amazon: High-Definition DVD Handbook Source notes: Apple Books and McGraw-Hill list the April 2007 release; VitalSource confirms the authors, publisher, edition, and ISBN.