Avid DNx is a lossy mezzanine codec specifically engineered for compositing and post-production workflows, offering reduced storage and bandwidth requirements compared to uncompressed video. The codec is based on SMPTE ST 2019-1 (VC-3). Originally named Avid DNxHR ("Digital Nonlinear Extensible High Resolution"), it extends and generalizes the Avid DNxHD codec's HD scope to also cover UHD, as well as custom resolutions and color spaces. Unlike other popular codecs like H.264 or AV1 it is not intended to be used as a distribution codec. In 2025 "Avid DNxHD" and "Avid DNxHR" were unified by Avid under the "Avid DNx" acronym.
History Avid DNx was originally introduced as Avid DNxHD in 2004 as a proprietary Constant Bitrate (CBR) codec, supported in Avid products like Avid MediaComposer. It was standardized by SMPTE (Society of Motion Picture and Television Engineers) in 2008 as SMPTE ST 2019-1 (VC-3), covering HD resolutions in YCbCr color encoding and specifying encapsulation of the resulting streams in MXF and MOV (QuickTime File Format) containers.
The 2014 revision of SMPTE ST 2019-1 (VC-3) added encoding directly in the RGB color model, allowing the codec to store graphic content without the need to convert to YCbCr representation, either externally or internally, as typically required by other codecs such as Apple ProRes.
On September 12, 2014, Avid Technology, Inc. announced the DNxHR (High Resolution) codec as part of a broader "Avid Resolution Independence" announcement at their Fall 2014 Avid Connect event, which was held during the IBC 2014 conference in Amsterdam, Netherlands.
DNxHR was formally standardized with the publication of the 2016 revision of SMPTE ST 2019-1 (VC-3), where DNxHR corresponded to the newly introduced Resolution Independent (RI) profile. It extended the resolution coverage from HD to support any resolution between 2x2 and 16384x16384 pixels, most importantly including support for the emerging 2K, 4K and 8K resolutions, as well as adding support for an optional alpha channel (lossy or lossless) and Variable Bitrate (VBR) encoding.
In 2021 Avid DNxHR was added as production codec to the SMPTE ST 2067 suite of standard documents covering the IMF (Interoperable Master Format), allowing the codec to be used in modern post-production environments.
In 2022 SMPTE published an amendment to SMPTE ST 2019-1 (VC-3), which extended the support for the alpha channel into the HD profile.
At the IBC 2025 conference in Amsterdam Avid announced a rebranding of DNxHD and DNxHR to "Avid DNx", to eliminate the confusion around supported resolutions and bit depths arising from the continued and often misleading use of the HD and HR acronyms .
The publication of the 2026 revision of SMPTE ST 2019-1 (VC-3) adds two new RGB compression levels (HQ and SQ), which were previously released by Avid under the acronym "Avid DNx GX". This revision also renamed the "RI" profile to "HR", bringing this in line with industry practice, which is based primarily on use of Avid DNx. It also generalizes the custom bitrate support to CBR mode (previously only supported in VBR mode, to stay compliant with 2016 revision of SMPTE ST 2019-1).
In 2026 the Academy of Television Arts & Sciences awarded Avid Technology an Emmy Award for the development of Avid DNx.
Specifications Source:
Avid DNx is a fully compliant implementation of SMPTE ST 2019-1 (VC-3). It supports 5 primary compression levels, corresponding to different bitrate ranges and compression quality levels, as found in SMPTE ST 2019-1 (VC-3).
Avid DNx can carry any color coding and supports all bit depths between 8 and 16 bits for filler and lossy alpha.
444: 4:4:4 and RGB, highest quality and bitrate, lossless alpha support (VFX work and Cinema delivery). HQX (High Quality eXtended): 4:2:2 and 4:2:0, less aggressive quantization than HQ if needed for multi-generation workflows (High-end broadcast production and delivery) HQ (High Quality): RGB, 4:4:4, 4:2:2, 4:2:0, high end format for standard broadcast usage (none or few multi-generation work, specifically sports and advertising) SQ (Standard Quality): RGB, 4:4:4, 4:2:2, 4:2:0 , standard broadcast production usage (everyday broadcast production) LB (Low Bandwidth): 4:2:2 and 4:2:0, Proxy format for post-production work, replaced during finishing with a higher quality content representation (rough cuts and non-VFX work)
+LevelBit DepthColor CodingResolutionSub-samplingBitrate Range (Mbps @ 30 FPS)4448 through 16Anyup to 16384 x 163844:4:4(:4), RGB(A)116-440HQX4:2:2(:4), 4:2:0(:4)72-220HQ4:4:4(:4), 4:2:2(:4), 4:2:0(:4), RGB(A)36-220SQ4:4:4(:4), 4:2:2(:4), 4:2:0(:4), RGB(A)36-145LB4:2:2(:4), 4:2:0(:4)36-45 Table reproduced from , showing format support in Avid DNx. The bitrate range reflects the range available when using custom bitrates. The upper limit corresponds to the compressed size of the Level specified in SMPTE ST2019-1.
Users first select a Level based on the high-level quality aspects (aggressiveness of the quantization matrix). Within this Level they can then select any target bitrate within the specified Bitrate Range.
Up to the 2026 revision of SMPTE ST 2019-1 this adjustable bitrate feature required the use of the VBR flag. The 2026 revision generalized this to also cover CBR bitstreams.
The codec is available in a free version at developer.avid.com
Trademarks and Patent Situation "Avid DNx", "Avid DNxHD", "Avid DNxHR", "DNxHD" and "DNxHR" are trademarks owned by Avid Technology.
The last of the patents associated with SMPTE ST 2019-1 (VC-3) runs out on January 18, 2027. However, as the codec has been continuously evolved, Avid holds several other patents which are either specific to the Avid DNx encoder (not covered by the standard) or to processing Avid DNx bitstreams. These will not run out before 2045.
Support Avid DNx is primarily supported through the Avid DNx SDK, published and licensed by Avid Technology. The SDK is the only complete implementation of the standard. Reference decoder software (un-optimized) is also available through SMPTE . The SDK supports all major platforms (Windows Intel, Windows ARM (Neon), Windows ARM EC, Linux Intel, MacOS ARM (Neon)).
In software VC-3 is also supported by FFmpeg. However this implementation is known to be incomplete, supporting primarily the HD resolutions of the original 2008 version of the VC-3 standard, frequently refusing operations for options introduced later (including most of the DNxHR options) and does not comply with the Avid DNx encoder specifications.
Hardware implementations of Avid DNx are available from several vendors, like for example Atomos.
Avid DNx is supported by all major editing applications, including Avid Media Composer, Adobe Premiere Pro, Blackmagic Design Resolve and Apple Final Cut Pro. Avid DNx is currently licensed to a wide range of application providers.
Use as an Editing Codec Avid DNx is widely used in cinematic and high-end broadcast productions, which have specific post-production requirements. It was the first video codec specifically engineered to serve the needs of editing/post-production workflows.
Most video codecs today are used for consumption/distribution (streaming). This drives their primary architectural traits: They are designed to deliver the best possible quality at the least transfer cost (bandwidth) for sequential access by many viewers at the same time. They go to great technological lengths to achieve these generally adversarial goals, making them so complex that sustained use can only be maintained by using dedicated hardware, and usually only one stream at a time.
What they are not designed to deliver are fast random access, sustained high quality through many encoding/rendering generations, and the ability to update individual frames. They are designed for consumption, not further processing. Distribution codecs represent an end-point of a delivery infrastructure, while editing codecs need to serve the exact opposite end, the start and the entire length of the production pipeline.
In 2004, developers Katie Cornog, Dane Kottke and Michel Rynderman realized that the traits that made codecs a good distribution choice made them a poor editing choice. Production/Editing codecs should provide near instantaneous random access, be able to support many streams in parallel for compositing scenarios, offer superb quality over many editorial generations as the input gets more and more refined in the production process, and be easily implemented and supported in software on standard computing platforms without the need for specialized and expensive hardware. They decided to define a codec which would explicitly put production needs before efficiency considerations, overcoming the performance limitations presented by distribution codecs when used in an editing environment.
The DNxHD codec built on the comparatively simple foundation of the original JPEG image codec. The codec is the archetype of lossy, DCT-based codecs, which forms the mainstay of modern, MPEG-based codec technologies. It uses a single quality parameter which defines how much signal quality loss is deemed acceptable. The parameter makes the quality predictable, but not the resulting compressed size (bitrate): The Variable Length Coding which controls the compressed size depends strongly on the complexity of the content, resulting in each frame varying widely in size after compression. When applied to a stream of images (called frames in video terminology) this results in what is known as a Variable Bit Rate (VBR) bitstream.
While this is perfectly acceptable in photographs, it presents a massive obstacle for a production video codec, which needs to deal with long streams of such frames. If each frame has a different compressed size, this necessitates the use of a frame index, which lets the decoder know exactly where within the stream the corresponding frame can be found. Not only does this make the handling complex and expensive, it also imposes strong inhibitions on the resulting video files:
During live event recording, an editor can access a frame only once the corresponding index is available. This introduces long latencies until the captured content can be accessed and processed further.
The content of a specific frame gets frequently altered during editing: Content may be cropped, scaled, rotated, color-corrected or generally be subjected to effect processing. All these operations will result in a different compressed frame size if the content is re-compressed (rotoscoping), making it impossible to replace the frames in an existing file (in-place edits): To exchange a single frame, the entire file needs to be re-written.
Typically an edited program consists of many short segments called clips or cuts, which are assembled on a timeline that controls the serial playback. During the editing process these clips/cuts, which come from different source files, are cut, trimmed and re-arranged in their playback order.
When finished, the assembled result usually needs to be written into a single output file. With linear media like film and tape, this put a time-consuming finishing step at the end of the editing process, as the material also had to be assembled linearly, one frame after another. VBR-coded media inherit this time-consuming linear finishing characteristic, as the position of a frame depends on the (compressed) size of all preceding frames in the program.
DNxHD changed this fundamentally by replacing the single JPEG quality parameter with many such quality parameters across a single frame. This allows the encoder to control the compression on 16 x 16 pixel areas known as macroblocks. The Avid DNx encoder controls both the local quality and the local bit cost of each macroblock using the local quality parameter. The sum of the local bit costs then forms the compressed frame size.