The broadcast squeezeback, rebuilt with CSS Grid and WebVTT | Hacker News Reader

The broadcast squeezeback, rebuilt with CSS Grid and WebVTT

mux.com

26 pointsmmcclure10 comments

> I'd seen that effect a bajillion times without really looking too hard at it. It's just what television does. But during one of my infamous 2am can’t-sleep iPhone research sessions, I learned it actually has a name: a squeezeback.

reminded me why i love hanging out with media folks: TouchDesigner helped me bridge tech and interactives in the best way, letting them tell me exactly whats going on in their head using their flows and lingo, and i can fast-feedback their designs iteratively while theyre in the room

I didn't know it was called a squeezeback either. The highlight of the AMPEX ADO 1000 just put me down a rabbit hole into some history.

sorry about the nerd snipe

author here! this was a fun one. the idea had been noodling around my head for a while- still would love to figure out a workflow for live streams for this, I'm sure it's doable but maybe not with native-only tech like it is right now

Just use OBS and arbitrary HTML web pages, hooked up to a small backend server, in OBS's embedded browser layer type. Easy peasy and a TON of fun!

Are these transitions supposed to be animated, i.e. the video changes size over a few seconds? Because it just jumps instantly to the new size for me on both Firefox and Chrome.

can you share the os and browser versions? they're animated in Chrome 150 for me

Maybe prefer-reduced-motion?

Yep, that was it. This is not the first time a website failed to perform as advertised because of this. To be clear, I'm blaming the websites or their creators; this is not a bug, but rather a feature working exactly as intended. But I would recommend that if you're showcasing a feature that depends on a certain setting, and that setting isn't set, then you should show a warning of some sort (implementation details left as an exercise for the developer).

> Apparently, the hardware could do it by the early eighties. Ampex's ADO and the digital video effects boxes that followed could scale and reposition a live signal in real time

Fun history: The very first digital squeeze effect box was called the Squeezoom by Vital and released in 1977 (https://patents.google.com/patent/US4266242A). It made big waves when first seen on national TV during a live Grammy awards ceremony putting camera shots of all four nominees into a 'quad-split'. Prior to the Squeezoom, quad-splits were faked using an analog video switcher to simply crop four full-screen images into quarters. No pixels were shrunk or even moved, instead, each of the four images were framed by the camera operator in their respective corners with the camera's optical zoom lens.

Digitally scaling a video frame in real-time was quite a feat with late 70s TTL chips which were barely fast enough to keep up with full bandwidth 6 Mhz video rates by using costly delay lines and buffers pushing the Squeezoom's price to $250K (over $1,000,000 today), However the trickiest part was implementing high-quality multi-tap filters to smooth the scaled pixels in real-time. While the Squeezoom was first, won an Emmy for technical achievement and established the term "Squeeze", it was also legendary not only for its cost but for being cantankerous and unreliable.

The Ampex Digital Optics (ADO) shipped in 1981 and got even more attention because the Squeezoom could only scale images in X and Y but the ADO could simulate 2.5D perspective with rotation and affine transforms. It's worth reading the patent just to appreciate how the hell they did it with 1980 TTL chips (https://patents.google.com/patent/US4472732/). Here's the demo video for the 2nd gen ADO 100 (https://www.youtube.com/watch?v=Zw1JrNBroF8&t=160s).

A year later later Quantel introduced their $300k Mirage effects unit which could warp live video onto an arbitrary 3D geometric shape, however the trick was it pre-rendered all the geometric remapping coordinates to a massive RAM disk using a Pascal program running on a workstation (which could take an hour or more to calculate) and only played them back on cue. The real-time part of the Mirage's hardware just digitized a live video input and mapped it to the fixed stream of pre-generated addresses by mapping the 2D video X/Y pixels to pre-calculated 3D X/Y/Z coordinates. Thus the Mirage could warp video into complex geometries like cubes, spheres, cylinders and even page peels but those geometries were canned, not interactive. The Ampex Digital Optics hardware did generate addresses in real-time entirely in logic-gate based hardware but throughout the 80s the geometries became increasingly complex and the desire for real-time video effects with shadows, trails, reflections, specular highlights and multiple colored light sources forced more of a true 3D rendering pipeline to be implemented in custom ASICs. The vast budgets to develop this bleeding-edge custom hardware were funded by selling astronomically expensive video effects gear. The allowed video engineers to solve the challenges of real-time 3D address generation, texture mapping and lighting - and that led directly to early Graphics Processing Units.

The ultimate real-time 3D video warping effects box was Sony's legendary $350,000 System-G which shipped in 1992 and reportedly had worldwide sales of... three units. I saw one in mid-90s at The Post Group in LA, just a few months after I'd seen a 1983 Quantel Mirage literally being used to prop a closet door open at Broadway Video in NY. The incredible pace of innovation in real-time effects was driving equally extreme rates of depreciation. In the late 80s, a young engineer at Sony named Ken Kutaragi saw the early System-G prototype in a research lab and imagined that racks of million dollar hardware could someday be amazing for video games. A few years later Kutaragi launched Sony's Playstation development effort with some former System-G engineers. In fact, the history of real-time digital effects for TV is full of notable names from early computer graphics and video games. Ed Catmull and Alvy Ray Smith were at Ampex in the late 70s and Lucasfilm in the early 80s before co-founding Pixar. Two young Ampex engineers named Nolan Bushnell and Ted Dabney 'borrowed' parts from work to prototype early video game graphics hardware but Ampex management saw no market, so the pair left and founded Atari.

Over the rest of the 90s, the visual magic of the ultra high-end System-G trickled down into lower-cost television production gear and eventually desktop-based real-time video and graphics hardware like the low-cost Video Toaster, Play Trinity, Targa frame buffers, etc. By the mid-2000s, desktop CPUs and GPUs started becoming as powerful as even the most advanced custom ASICs and FPGAs in broadcast video gear and the amazing 30 year journery that began in the late 70s with the 2D Squeezoom came to an end. Today, the GPU descendants of those early video effect ASICs have completely replaced their ancestors and the real-time virtual environments you see mapped into stadiums on Monday Night Football are rendered by desktop GPUs in rack-mounted PCs.

Reply on news.ycombinator.com