10 Things I found Decoding Netflix's AV1 Streams
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
The Stranger Things Netflix Problem is Killing TV
Philip DeFranco
479.8k views
AI Coding Sucks | Prime Reacts
ThePrimeTime
245.8k views
WE ARE SO BACK - Day In The Life at Netflix
ThePrimeTime
330.5k views
Final stretch
ThePrimeTime
34.3k views
Things are breaking down
ThePrimeTime
48.7k views
I'm switching languages again...
ThePrimeTime
165.4k views
Coding Should Be A Vibe
ThePrimeTime
154.3k views
Day 6 - VIBE CODING A GAME IN 7 DAYS
ThePrimeTime
119.7k views
Day 5 - VIBE CODING A GAME IN 7 DAYS
ThePrimeTime
71.3k views
Day 4 - VIBE CODING A GAME IN 7 DAYS
ThePrimeTime
67.1k views
Top Comments (10)
Respect for taking the feedback about the music!
Liking and commenting to repay you for taking the music out.
A major reason for cartoons being extremely hard is AV1, like most video encoders, weighs heavily in compressing chroma on a block or frame basis, as well as fudging edges. For most TV shows, and even minions, there's a huge delta in luma across a frame, which allows perceptual crushing of chroma (we have a harder time distinguishing dark colors, so you can comrpess the dynamic range (bit depth) at high and low luminosity. with live action, heavy use of color grading leads to further optimizations of dynamic range. A cartoon has a broad palette of bright colors and usually extremely contrasty edges, so you can't compress away color along a black line, shadows are marginally darker *and flat*. AV1's motion algorithms cause ghosting/blocking along chroma and contrast edges, which aren't noticeable in motion with live action or even 3D animation (which often uses cinematic blur). Cartoons are extremely hard to compress because every frame of a block in an animated sequence is distinct, but also irregularly related - this is why frame interpolation bugs out with animation. TLDR: AV1 is a multi-pass temporal algorithm which relies on narrowing the dynamic range per sample block and prioritizing luma detail over chroma detail. A profile which looks great on live action will look disastrous on cartoon/2D animation with solid colors.
13:50 Regarding HTTP/3 and QUIC, UDP is entirely *not* the point if you want to understand QUIC. QUIC is effectively a "next-gen TCP replacement", with all the bells and whistles, and architecturally it ideally would belong right next to TCP, UDP, etc. The reason it was layered on top of UDP is for ease of deployment (think OS support, firewalls, etc), not because QUIC conceptually has any inherent relation to UDP or some kind of "UDP behavior". (And UDP was chosen specifically because UDP introduces pretty much no behavior of its own, it essentially just adds a concept of ports to basic IP) Technically TCP could very well have been designed as a layer on top of UDP as well, with no difference in behavior, that would have been the exact same concept as what was done with QUIC.
I've done video encoding professionally for a long time for one of the large streaming services. The best explanation for why Bojack didn't really see a big efficiency gains is probably because H.264 and subsequently HEVC are *extremely* good at anime-style animation encoding and I don't think AV1 really brought any even moderate improvements specifically for animation encoding.
I'm glad to have article readings back! For some feedback, I feel like the webcam is too large too often and covers too much of the article, which makes it hard to read along and look at the charts and graphs.
I was in the middle of watching the music one and it went private as I was watching it lol
17:20 Fun fact, AV1 *does* support film grain at the encoder level
When your A.I writes AVI instead of AV1
Man HBO needs this, the night battle would be zero bytes for the video for the whole thing.
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
Respect for taking the feedback about the music!
Liking and commenting to repay you for taking the music out.
A major reason for cartoons being extremely hard is AV1, like most video encoders, weighs heavily in compressing chroma on a block or frame basis, as well as fudging edges. For most TV shows, and even minions, there's a huge delta in luma across a frame, which allows perceptual crushing of chroma (we have a harder time distinguishing dark colors, so you can comrpess the dynamic range (bit depth) at high and low luminosity. with live action, heavy use of color grading leads to further optimizations of dynamic range. A cartoon has a broad palette of bright colors and usually extremely contrasty edges, so you can't compress away color along a black line, shadows are marginally darker *and flat*. AV1's motion algorithms cause ghosting/blocking along chroma and contrast edges, which aren't noticeable in motion with live action or even 3D animation (which often uses cinematic blur). Cartoons are extremely hard to compress because every frame of a block in an animated sequence is distinct, but also irregularly related - this is why frame interpolation bugs out with animation. TLDR: AV1 is a multi-pass temporal algorithm which relies on narrowing the dynamic range per sample block and prioritizing luma detail over chroma detail. A profile which looks great on live action will look disastrous on cartoon/2D animation with solid colors.
13:50 Regarding HTTP/3 and QUIC, UDP is entirely *not* the point if you want to understand QUIC. QUIC is effectively a "next-gen TCP replacement", with all the bells and whistles, and architecturally it ideally would belong right next to TCP, UDP, etc. The reason it was layered on top of UDP is for ease of deployment (think OS support, firewalls, etc), not because QUIC conceptually has any inherent relation to UDP or some kind of "UDP behavior". (And UDP was chosen specifically because UDP introduces pretty much no behavior of its own, it essentially just adds a concept of ports to basic IP) Technically TCP could very well have been designed as a layer on top of UDP as well, with no difference in behavior, that would have been the exact same concept as what was done with QUIC.
I've done video encoding professionally for a long time for one of the large streaming services. The best explanation for why Bojack didn't really see a big efficiency gains is probably because H.264 and subsequently HEVC are *extremely* good at anime-style animation encoding and I don't think AV1 really brought any even moderate improvements specifically for animation encoding.
I'm glad to have article readings back! For some feedback, I feel like the webcam is too large too often and covers too much of the article, which makes it hard to read along and look at the charts and graphs.
I was in the middle of watching the music one and it went private as I was watching it lol
17:20 Fun fact, AV1 *does* support film grain at the encoder level
When your A.I writes AVI instead of AV1
Man HBO needs this, the night battle would be zero bytes for the video for the whole thing.