Video Compression Artifacts: 10 Types, Examples & Fixes

Updated November 2026: how to tell compression artifacting from GPU and display artifacting, YouTube's AV2 plan for early 2027 and its AI Super Resolution, VMAF v1 in libvmaf v3.2.0, current H.264 (06/2026) and H.265 (01/2026) editions.

What are compression artifacts? The short answer

Compression artifacts are visible errors that lossy compression adds to an image or video: blocks, stripes, halos, smears or stutter that were not in the original. They appear when an encoder such as H.264, HEVC or AV1 discards more information than the content can afford, usually because the bitrate is too low for what is on screen. In everyday speech, "video artifacting" means the same thing: the picture shows structure the camera never recorded. The most common types are blocking (macroblocking), banding, ringing, mosquito noise, blur, color bleeding and, in motion, judder, stutter, freezing and tiling. Each one has a specific cause, so once you name the artifact you know where to look for the fix.

Why this matters

Anyone who encodes, streams, or QCs video will sooner or later stare at a frame that looks wrong and need to answer two questions quickly: what am I looking at, and what caused it? Guessing costs a debugging afternoon and sometimes lets the bug ship. This guide gives you names for every common artifact, a map from symptom to cause to fix, and, since this is a measurement section, an honest table of which metric catches each one. It is written for the video engineer, encoding lead, or QA engineer who can tell that "something is off" and wants to name it, measure it, and trace it properly.

What a compression artifact actually is

Start with the word. An artifact is any visible structure in the picture that the camera did not record and the director did not intend. It is an artifact of the process, not of the scene. Lossy video compression discards information the encoder bets you will not miss: fine detail, subtle color, precise motion. Usually the bet pays off and you see nothing. When the encoder is pushed too hard (too few bits, too aggressive a setting, difficult content), the discarded information leaves a visible scar, and that scar is the artifact.

Think of compression as packing a suitcase. If you fold the clothes carefully they come out fine; if you sit on the lid to force it shut, everything comes out creased in predictable places. Compression artifacts are those creases. They are not random noise. Each one has its own mechanism, shows up in a particular place, and has a recognizable look, and that is why a field guide like this can exist at all.

One distinction is worth making first, because it changes how you hunt. Not every artifact comes from the encoder. Some originate earlier, in a noisy or already-compressed source. Others originate later, in delivery: a network drop, an adaptive-bitrate switch, a decoder bug. The encoder sits in the middle, so it gets blamed for all of them, but the fix belongs wherever the artifact was introduced. Tracing that path is the job of the capstone article on tracing an artifact to its cause; this guide covers recognizing the artifact in the first place.

The four families of artifacts

No single official taxonomy of video artifacts exists, but the research literature has settled on a split between spatial artifacts (errors inside a single frame) and temporal artifacts (errors that appear only in motion, across frames). The classic survey by Yuen and Wu in Signal Processing (1998) cataloged the spatial set (blocking, ringing, the mosquito effect, blurring, and color bleeding), and later perceptual studies added the temporal set. For a working engineer, sorting by origin is more useful, because the origin tells you where the fix is. This guide uses four families.

The first family is block-transform artifacts. Nearly every codec (H.264/AVC, HEVC, AV1) splits each frame into blocks and compresses each block with a Discrete Cosine Transform (DCT). This math step represents the block as a sum of patterns, then quantizes (rounds off) those patterns to save bits. If the rounding is too coarse, the block structure becomes visible. Blocking, ringing, mosquito noise, and blur all come from this one mechanism.

The second family is tonal-quantization artifacts, mainly banding, also called contouring. Banding appears in smooth gradients such as a sky, a fade, or a studio backdrop when there are too few distinct brightness levels to render the gradient smoothly, so it breaks into visible steps.

The third family is color artifacts: color bleeding and the errors caused by chroma subsampling, the standard practice of storing color at a lower resolution than brightness. Saturated edges and colored text suffer most.

The fourth family is temporal and delivery artifacts: judder, stutter, flicker, and problems on the delivery side such as bitrate switches, freezes, and tiling from packet loss. These exist only in motion or in transit, and per-frame metrics cannot see them at all.

Figure 1. The four families. Sorting artifacts by where they originate shows you where the fix belongs.

The block-transform family: blocking, ringing, mosquito noise, blur

All four come from the same source, the block-based DCT and its quantizer, so they tend to appear together and get worse together as bitrate drops.

Blocking (often called macroblocking) is the most recognizable artifact: the picture breaks into visible squares, as if built from coarse tiles. The codec processes the frame on a grid of blocks, and at low bitrate each block is quantized so coarsely that neighboring blocks no longer match along their shared edge. The grid is real and you can count it. A 1920×1080 frame split into 8×8 transform blocks gives 240 × 135 = 32,400 blocks, which means 32,400 small edges where a visible seam can open. Modern codecs counter blocking with a normative in-loop deblocking filter, defined in the H.264/AVC and HEVC standards, which smooths block boundaries before the frame is displayed and reused as a reference; HEVC adds a second stage called Sample Adaptive Offset. If blocking is still visible, the encoder has run out of bits to hide it. Blocking and the DCT grid covers the full mechanism and how metrics behave.

Ringing is the rippling or ghosting that clings to a sharp, high-contrast edge, such as the outline of text or a building against the sky. It is a textbook example of the Gibbs phenomenon: when the quantizer truncates the high-frequency DCT coefficients a crisp edge depends on, the reconstructed edge overshoots and oscillates, like a struck bell that keeps ringing instead of falling silent. Ringing is visible in a still frame.

Mosquito noise is ringing that moves. The encoder quantizes each frame a little differently, so the ringing around an edge shimmers from frame to frame, a cloud of busy dots hovering around hard edges like mosquitoes around a porch light. It is worst on the content real products show most often: sharp text, logos, and computer-generated graphics over a flat background. NIST research on mosquito noise traces it directly to block-DCT quantization. Ringing and mosquito noise around edges examines ringing and mosquito noise together in detail.

Blur is less conspicuous: fine detail and texture disappear when the quantizer discards the high-frequency information that carries them, or when the video was downscaled and never recovered. Grass becomes a green smear, skin loses its pores, brick loses its mortar. The difficulty is that blur is not always an artifact. A shallow depth of field is intentional softness, so the real skill is telling a creative choice from a compression failure, which is the subject of Blur and detail loss.

The tonal family: banding

Run a smooth gradient (a clear evening sky, a slow fade to black, a softly lit studio wall) through a hard enough encode and it stops being smooth. It splits into visible bands of flat color with sharp steps between them, like a topographic map. This is banding, also called contouring, and its cause is precise: there are too few distinct code values to represent the gradient, so many pixels round to the same value, form a flat plateau, and then jump to the next one.

The arithmetic points straight at the fix. Standard 8-bit video has 2^8 = 256 brightness levels per channel. Suppose a sky ramps gently from code value 90 to code value 110, only 20 distinct levels, across 400 pixels of height. That gives 400 ÷ 20 = 20 pixels of identical brightness before the value is allowed to change, so you see 20-pixel-wide flat bands with a visible step between each. Encode the same gradient in 10-bit, which has 2^10 = 1024 levels (four times as many), and the same brightness range spans about 80 code values. The bands shrink to roughly 400 ÷ 80 = 5 pixels, and the steps drop below the threshold where the eye picks them out. This is why the standard fixes for banding are higher bit depth (8 → 10-bit), dithering (adding a little calibrated noise to break up the steps), and dedicated debanding filters. The cause-side mechanics of bit depth live in the Video Encoding section's bit-depth article, and the measurement side is Banding: when smooth gradients break into steps.

Banding also distorts your metrics, as covered below: it is the artifact the popular scores miss most completely.

The color family: bleeding and chroma subsampling

Color gets compressed harder than brightness on purpose, because the eye is far more sensitive to luminance (how bright) than to chrominance (what color). Codecs exploit this with chroma subsampling, most commonly the 4:2:0 scheme, which stores a full-resolution brightness image but only a quarter-resolution color image. This is usually invisible. On a saturated, high-contrast color edge, such as red text on a gray background or a bright logo, the low-resolution color can smear past the sharp brightness edge it is supposed to sit inside. That smear is color bleeding, where the color leaks outside its lines.

Yuen and Wu traced the mechanism to the quantizer zeroing out the higher-frequency color coefficients. What remains is a coarse, low-frequency version of the color that cannot keep up with a sharp edge. It shows most on saturated reds and on the colored text in screen content and presentations. The recognition-and-measurement treatment is Color artifacts: bleeding, shift, and chroma subsampling, and the encoding-side mechanics of color spaces and subsampling belong to Video Encoding's color article.

The temporal and delivery family: judder, stutter, freezes, tiling

The last family exists only in motion, and the standard metrics are structurally unable to see it.

Judder is uneven motion: a slow camera pan that should glide instead lurches in a repeating hitch. The classic cause is frame-rate conversion. Showing 24-frames-per-second film on a 60-Hz display forces a 3:2 pulldown (also called telecine), which displays one film frame for three refreshes and the next for two. That 3-2-3-2 cadence is uneven, and on a steady pan the eye reads the unevenness as a stutter. Stutter proper has a different cause: frames arrive late or get dropped, so motion that should be continuous jumps. Both are temporal, and neither changes how any single frame looks.

Delivery-side artifacts complete the family. An adaptive-bitrate switch drops to a lower quality mid-stream and snaps the picture softer for a few seconds; a freeze stops the picture while the player rebuffers; and tiling or corruption appears when lost network packets leave smeared blocks or a half-updated frame. The network and the player cause these, not the encoder, which is why they get their own article, Streaming-specific artifacts, and connect forward to the streaming-experience block. Judder and stutter themselves are covered in Judder, stutter, and frame-rate artifacts.

Figure 2. Five artifacts up close (synthetic frames). Each has a signature look that is easy to recognize once you have seen it.

Which metric catches which artifact — and which it cannot

A quality metric is only useful for artifacts it can detect. The three popular full-reference metrics each have hard blind spots: PSNR (Peak Signal-to-Noise Ratio, raw pixel error in decibels), SSIM (Structural Similarity, a 0–1 structural match), and VMAF (Video Multimethod Assessment Fusion, Netflix's 0–100 perception-trained score).

Blocking and blur fall squarely within what these metrics measure, so they catch them reasonably well. They catch ringing and color bleeding only weakly, because the error is small in pixel terms but still visible to the eye. Banding is the worst case: Netflix's 2021 CAMBI post says that VMAF, PSNR and other common metrics "don't detect banding artifacts properly", which is why Netflix built a separate detector, CAMBI (the Contrast-Aware Multiscale Banding Index). The entire temporal family (judder, stutter, flicker) is invisible to all three for a structural reason. PSNR, SSIM, and VMAF are computed frame by frame and then averaged, so an artifact that lives in the timing between frames never enters the calculation.

What changed in 2026: VMAF v1. On June 20, 2026 Netflix open-sourced VMAF v1, which closes part of this gap. It ships in libvmaf v3.2.0 as the v1.0.16 SDR models, so HDR content still needs another check. The release adds CAMBI as a built-in banding feature, plus a chroma feature (the standard v0.6.1 model used luma only, so color bleeding barely moved the score). A new feature, AIM, sits beside the existing detail-loss feature and penalizes blockiness the v0 model under-scored. Motion handling also improved, with a capped motion feature and an option to measure motion over a longer window, but Netflix itself lists film grain and high frame rates as still open. Most production pipelines still run the v0.6.1 models, so check which model your dashboard uses before you trust it on banding or color. The model details are in VMAF explained.

Figure 3. Which metric catches which artifact, with VMAF meaning the v0 models most pipelines still run. The popular metrics go dark on banding and the whole temporal family. Color and glyph both encode the verdict, so neither carries it alone.

Table 1. The field guide in one table: every common artifact, its signature, its cause, and whether each metric actually catches it. Pair every metric with the blind spot for the content in front of you.

Common mistake: trusting a green dashboard over your own eyes. The most expensive error in this topic is concluding "VMAF is 95, so the picture is clean." A VMAF v0 score of 95 can sit on top of obvious banding in the sky, color bleeding on the logo, and judder on every pan, because none of those enter the score. When the content is prone to a blind-spot artifact (smooth gradients, saturated text, slow pans), add the right detector, such as CAMBI for banding or a temporal check for judder, and confirm the hard frames by eye. A metric is a proxy that works within a limited domain, not a clean bill of health. The full catalog of these failures is Where objective metrics lie.

New artifact sources in 2025–2026: synthetic grain and ML enhancement

Two recent changes in video processing have introduced artifacts that the classic taxonomy above does not name.

Film grain synthesis. AV1 can strip film grain before encoding and re-create it in the decoder from a compact model, so the encoder does not spend bits on random noise. In July 2025 Netflix reported that its AV1 film-grain-synthesis streams cut the average bitrate by 31.6% and the starting bitrate by 24%. This brings a new failure mode: synthetic grain that looks too uniform, too coarse or stuck to the wrong regions, and banding that becomes more visible once the denoiser has flattened a gradient. A 2026 paper by Netflix's Andrey Norkin adds a debanding offset to the grain model for that reason. Pixel-aligned full-reference metrics such as PSNR and VMAF misread synthesized grain, because it is statistically similar to the original but sits in different places; VMAF v1 lists film grain as an open problem. Exposure is growing. In December 2025 Netflix said AV1 carried about 30% of its streaming, with film grain synthesis in production since July 2025.

ML "enhancement" smearing. In August 2025 YouTube confirmed it had been running an experiment that used "traditional machine learning to unblur, denoise, and improve clarity" on some Shorts during processing. Creators described the result as smeary, like an oil-paint filter, with skin and texture that looked artificial. The effect is over-smoothing plus hallucinated detail, a cousin of blur that per-frame metrics can even score as an improvement. On October 29, 2025 YouTube went further and announced "Super Resolution", an AI upscaler for uploads below 1080p that starts with SD to HD. Creators can opt out, the original file is kept, and viewers see the option labeled in quality settings. If your pipeline runs AI upscaling or denoising, check the hard frames by eye, as you would for banding.

The codecs are changing too. The Alliance for Open Media finalized the AV2 specification (version 1.0.0, dated May 28, 2026) and announced it in June 2026, and in October 2026 YouTube said its AV2 streams will be available on the web as soon as early 2027. The core mechanisms on this page (block transforms, quantization, in-loop filtering) largely carry over, so the field guide still applies while AV2's own artifact profile is being characterized.

Is it compression artifacting or a GPU problem?

Video artifacting from compression lives in the file or the stream, so the same blocks, bands or smears appear in the same frames on every device that plays it. GPU artifacting is a hardware or driver fault on your own computer. It shows up as random colored pixels, flickering triangles, checkerboard patterns or lines, and it appears in games, menus and the desktop as well as in video. So a search for "what is artifacting" returns two different answers; this guide covers the first kind.

You can tell them apart in about a minute. Play the same clip on a second device: if the artifact follows the clip, the cause is compression or delivery. Take a screenshot or screen recording; compression artifacts appear in it, while many display faults do not. If the glitches show up across unrelated content, get worse as the machine heats up under load, or change after a driver update, check the graphics card, cable or display rather than the encoder. Still images and audio have their own compression artifacts as well: JPEG blocking and ringing around text, or pre-echo and "swirly" sounds in low-bitrate MP3 or AAC audio. The mechanism is the same quantization of transformed data.

The decision aid: you see X, the cause is likely Y, the fix is Z

When an artifact shows up, start from the symptom and work back to the cause. The flow below is the short version; the tracing article is the full one.

If you see visible squares, that is blocking. The bitrate is too low or the quantizer too aggressive, so give it more bits, a better encoder preset, or stronger deblocking. If you see stripes in a smooth sky or fade, that is banding, caused by insufficient bit depth or quantization of a flat region; fix it with 10-bit encoding, dithering, or a debanding filter. If you see halos or shimmer around text and edges, that is ringing or mosquito noise from high-frequency truncation, and the fix is more bits, a lighter sharpening pre-filter, or a denoise pass before encode. If you see mush where texture should be, that is blur: confirm it is not intentional, then raise the bitrate or check the scaling chain. If you see color smearing past an edge, that is color bleeding from chroma quantization or subsampling, so use a higher-quality chroma setting or 4:2:2 for critical work. If motion lurches on a pan, suspect judder from a frame-rate mismatch. If it jumps or freezes, suspect dropped frames or rebuffering on the delivery side; the encoder is not involved.

Figure 4. Symptom, cause and fix. Read left to right: name what you see, work out the stage where it started, and apply the fix at that stage.

Compression artifacts FAQ

What does compression artifacts mean?

Compression artifacts are distortions that lossy compression adds to an image or video, ones that were not in the original. In video they appear as blocks, stripes in gradients, halos around edges, blur or jerky motion.

What is artifacting in video?

Artifacting is the everyday word for video compression artifacts: blocks, smears, banding or flicker in the picture because the encoder or the network lost information. Usually the bitrate was too low for the content, or packets were lost in delivery.

What causes video artifacts?

Most video artifacts come from quantization. To hit a bitrate, the encoder rounds away detail inside each block, and that shows up as blocking, ringing, blur and banding. The others come from the source (noise, a file that was already compressed), frame-rate conversion (judder) or delivery (rebuffering freezes, packet-loss tiling).

What are the types of video artifacts?

The common types are blocking (macroblocking), ringing, mosquito noise and blur, which come from block-transform coding; banding, from too few code values; color bleeding, from chroma subsampling; and judder, stutter, freezing and tiling, which show up in motion or delivery.

What is macroblocking?

Macroblocking is another name for blocking. The frame breaks into visible squares because each block was quantized so coarsely that its edges no longer line up with its neighbors. It means the scene got too little bitrate, or data was lost in a live stream.

What are visual artifacts?

Visual artifacts are visible structures in an image or video that were not in the scene, whatever produced them. In video the most common cause is lossy compression. After that come delivery faults (freezes, tiling), frame-rate conversion (judder), AI upscaling or denoising, and, on a computer screen, GPU or driver faults.

Is GPU artifacting the same as video artifacting?

No. Video artifacting from compression is baked into the file or stream and looks the same on any player. GPU artifacting is a hardware or driver fault that puts random pixels, lines or shapes over everything the computer displays. Playing the clip on a second device tells you which one you have.

Are encoding artifacts the same as compression artifacts?

Mostly yes: encoding artifacts are compression artifacts that the encoder introduced. But not every visible artifact is an encoding artifact. Freezes and tiling come from the network and the player, and judder comes from frame-rate conversion.

How do you fix compression artifacts?

Fix each one at the stage that caused it. For blocking and blur, give the encoder more bits or a slower preset. For banding, encode in 10-bit or add dither. For ringing and mosquito noise, denoise or soften sharpening before encoding. For freezes and tiling, fix the network or the player. Post-processing filters can hide artifacts, but they cannot bring back detail that was lost.

Can VMAF detect banding?

The widely used VMAF v0 models are poor at detecting banding. Netflix's separate CAMBI detector handles it, scoring from 0 up to about 24, and viewers start to find banding annoying at around 5. VMAF v1, open-sourced in June 2026, includes CAMBI as a feature.

For a one-line definition, see the compression artifact glossary entry.

Where Fora Soft fits in

Fora Soft has built video software since 2005, including streaming, WebRTC conferencing, OTT, e-learning, telemedicine and surveillance products, and every one of them ships frames that have to get through an encoder and a network. We recognize artifacts and trace them to the stage that caused them as part of our daily work. A banding complaint on a telemedicine feed is a question about bit depth and pre-processing, while a tiling complaint on a live stream is about packet loss, and the two get fixed in completely different places. We use the artifact gallery as a diagnostic tool and pair it with the right detector for each blind spot, because a metric dashboard on its own will pass several of the artifacts in this guide without flagging them. Where it helps a decision, we point to our own benchmark methodology so you can check the method yourself. When artifacts show up in your product, our video processing software developers trace them to the encoder settings, and our live and VOD streaming team fixes the delivery side, from packet loss to ladder design.

The handbook

Fora Soft Video Engineering Handbook

We wrote a seven-volume handbook that takes the whole video stack apart across 9,000+ pages - encoding, delivery, audio and AI. Volume 1 goes through the block transform, the quantizer and the in-loop filters that make every spatial artifact on this page, and the metrics that score them; volume 3 covers the delivery side where switching, freezing and tiling are born; volume 6 covers the models that now catch what a per-frame metric cannot. Written by Nikolay Sapunov, CEO at Fora Soft.

Browse the handbook on Amazon →7 volumes · Kindle & paperback

See our case studies

References

1. M. Yuen and H. R. Wu, "A survey of hybrid MC/DPCM/DCT video coding distortions," Signal Processing, vol. 70, no. 3, pp. 247–278, 1998. Tier 5 (peer-reviewed, foundational). The standard taxonomy of block-DCT coding artifacts (blocking, ringing, the mosquito effect, blurring and color bleeding), with the mechanism behind each. https://www.sciencedirect.com/science/article/abs/pii/S0165168498001285
2. ITU-T Recommendation H.264 (06/2026, in force; checked 2026-11-11), Advanced video coding for generic audiovisual services. International Telecommunication Union. Tier 1. Defines block-based transform coding and the normative in-loop deblocking filter that mitigates blocking. https://www.itu.int/rec/T-REC-H.264
3. ITU-T Recommendation H.265 (01/2026, in force; checked 2026-11-11), High efficiency video coding. International Telecommunication Union. Tier 1. Defines the HEVC 8×8 deblocking grid and Sample Adaptive Offset (SAO) in-loop filtering. https://www.itu.int/rec/T-REC-H.265
4. Netflix, CAMBI (resource/doc/cambi.md), Netflix/vmaf GitHub repository, accessed 2026-10-11. Tier 1 (metric-author). Defines CAMBI as a banding detector that fits the VMAF full-reference API; maximum observed score 24 ("unwatchable"), around 5 where banding starts to become slightly annoying. https://github.com/Netflix/vmaf/blob/master/resource/doc/cambi.md
5. P. Tandon, M. Afonso, J. Sole, L. Krasula, et al., "CAMBI: Contrast-aware Multiscale Banding Index," Picture Coding Symposium / arXiv:2102.00079, 2021. Tier 5 (peer-reviewed). Banding arises from quantization of smooth regions; the detector models contrast sensitivity and handles 8- and 10-bit content and dithering. https://arxiv.org/pdf/2102.00079
6. C. G. Bampis, Z. Li, K. Swanson, et al., "VMAF v1: Good Is Not Good Enough," Netflix Technology Blog, June 20, 2026. Tier 1 (metric-author). Open-sources VMAF v1: CAMBI as an elementary feature, a chroma feature, AIM alongside DLM for blockiness, a capped motion feature. Confirms VMAF v0's blind spots (banding, which CAMBI now covers in v1, plus chroma and the motion/frame-rate limits) and says grain and high frame rate are still unsolved. https://netflixtechblog.com/vmaf-v1-good-is-not-good-enough-60d7e4244ea8
7. Netflix, VMAF Frequently Asked Questions (resource/doc/faq.md), Netflix/vmaf GitHub repository, accessed 2026-06-24. Tier 1 (metric-author). VMAF is a per-frame proxy aggregated by arithmetic mean by default; this is the basis for the rule that per-frame metrics cannot see temporal artifacts. https://github.com/Netflix/vmaf/blob/master/resource/doc/faq.md
8. F. Yuan (NIST/ITS), "Mosquito noise in MPEG-compressed video: test patterns and metrics," SPIE, 2000. Tier 5 (peer-reviewed/institutional). Traces mosquito noise to the block-based DCT quantization process and characterizes its appearance around edges. https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=28452
9. A. Unterweger, "Compression Artifacts in Modern Video Coding and State-of-the-Art Means of Compensation," in Multimedia Networking and Coding, IGI Global, 2013. Tier 5 (peer-reviewed). Survey of artifacts in modern (H.264/HEVC-era) codecs and the in-loop tools codecs use to compensate. https://www.wavelab.at/papers/Unterweger13a.pdf
10. L. Krasula, et al. / Netflix, "CAMBI, a banding artifact detector," Netflix Technology Blog, 2021. Tier 4 (deployer engineering). States that "VMAF, PSNR, and other commonly used video quality metrics don't detect banding artifacts properly"; practical account of why banding evades standard metrics and how dithering and bit depth change its visibility. https://netflixtechblog.com/cambi-a-banding-artifact-detector-96777ae12fe2
11. ITU-T Recommendation P.910 (10/2023), Subjective video quality assessment methods for multimedia applications. International Telecommunication Union. Tier 1. The standard procedure for assessing impairments (including artifacts) with human viewers; this is the ground truth a metric is validated against. https://www.itu.int/rec/T-REC-P.910
12. Netflix, "AV1 @ Scale: Film Grain Synthesis, The Awakening," Netflix Technology Blog, July 2025, as reported by Broadband TV News, July 4, 2025. Tier 4 (deployer engineering). AV1 film grain synthesis streams: −24% starting bitrate, −31.6% average bitrate, ~3% fewer playback errors. https://www.broadbandtvnews.com/2025/07/04/netflix-film-grain-synthesis-saves-on-the-bitrate/
13. A. Norkin (Netflix), "Film Grain Synthesis with Debanding Feature," 2026. Tier 5 (author preprint). Denoising before encoding can make banding more visible; proposes a debanding offset in the grain model for AV1/AV2. https://norkin.org/pdf/DCC_2026_FGS_with_debanding.pdf
14. T. Mehta, Android Authority report on the YouTube Shorts processing experiment, August 26, 2025. Secondary (trade press quoting YouTube support and YouTube's creator liaison Rene Ritchie). YouTube describes an experiment using "traditional machine learning to unblur, denoise, and improve clarity"; creators report a smeary, oil-paint look. https://www.androidauthority.com/youtube-shorts-upscaling-experiment-3591230
15. Alliance for Open Media, "Alliance for Open Media Releases AV2 Codec," press release, June 2026. Tier 1 (standards body). Release of the AV2 specification. https://aomedia.org/press%20releases/Alliance-for-Open-Media-Releases-AV2-Codec/
16. Netflix, Releases: Netflix/vmaf (libvmaf v3.2.0, June 20, 2026), GitHub, accessed 2026-11-11. Tier 1 (metric-author). Initial public release of VMAF v1 with the v1.0.16 SDR models, CAMBI updates and new chroma and temporal feature extractors. https://github.com/Netflix/vmaf/releases
17. TV Tech, "AV1 Open Video Codec Now Powers 30% of Netflix Streaming," December 3, 2025, reporting the Netflix Technology Blog. Secondary (trade press). AV1 at about 30% of Netflix streaming; film grain synthesis productized in July 2025. https://www.tvtechnology.com/news/av1-open-video-codec-now-powers-30-percent-of-netflix-streaming
18. 9to5Google, "YouTube adds AI-upscaled 'Super Resolution' for low quality videos," October 29, 2025, reporting YouTube's announcement. Secondary (trade press). AI upscaling for uploads below 1080p, starting with SD to HD; creator opt-out; originals kept. https://9to5google.com/2025/10/29/youtube-adds-ai-upscaled-super-resolution-for-low-quality-videos/
19. M. Corrigan, "YouTube to Commence AV2 Streaming in 2027," TV Tech, October 27, 2026. Secondary (trade press quoting YouTube Engineering Director Des Keane). YouTube AV2 streams "available for Web as soon as early 2027"; over 90% of YouTube's catalogue (by watch time) available in AV1 as of March 19, 2026. https://www.tvtechnology.com/platform/streaming/youtube-to-commence-av2-streaming-in-2027