"That sounds amazing. How did you do that? I'm still super stoked to see that compiler. When will you release it? Do you have some example data, maybe a short video?"
I have mostly coded the decoder at this stage - I told you that. It is pretty simple at less than 1k lines of lua. I also described a first pass algorithm to encode an image. Divide it in to blocks and set each block to the most common colour, then patch the remaining pixels with set pixel and jump relative. With a reduced colour space, that seems like it will work reasonably well and perform similar or better than png. That is not a very hard thing to program in lua. I have not written this yet, but probably will write it to test the draw engine.
This is not a general purpose codec like Netflix uses. It is a draw engine that uses simple commands to draw primitive images in a pixel map. Maybe it is a bit like WMF, but a lot simpler. There is no SVG, just a pixel map. It might be possible to write a compiler that takes a video sequence and encodes it into an instruction sequence. That would be pretty hard and probably require some mpeg like tricks with reusing previous frames. I don't care if I or anyone else makes that work - it is a possibility.
The goal for this draw engine is to be able to quickly draw images from a compressed data stream, control the operation frame by frame and be able to set the viewport frame by frame. In 1024x1024 you can load 96 small 128x85 images and then make a short repeating video by switching the viewport on a frame by frame basis. It is not meant to be Starwars in 4K video. In fact, if you look at the type of programs people like to make with love2d, most of the graphics is quite simple and stylized - it is not photorealistic. Pixel graphics with a reduced colour space often looks better. This draw engine is designed for simple pixel graphics.
Initially I have some simple applications for the draw engine. Static images, video noise and scrolling/panning images. For example, suppose you want to show a simulation of a security camera in a game. Just load the full 360 camera view into the pixelmap, then issue viewport commands to pan the image back and forth. That can easily be encoded into a string and interpreted. You could have hundreds of simple animations like this. You don't need a photo realistic netflix codec for this.
@grump In your efforts to make yourself angry, you seem to have imagined that I claimed to have written a compiler that can replace a codec. No. I was pointing out that draw engines work in a different way to modern codecs. They build an image algorithmically, as opposed to using signal processing techniques to compress colour space in 3D. Codecs also exploit the physiology of the human vision system to eliminate redundant data. Given a system that can draw graphics primitives, you could make a Turing machine that keeps going until the destination pixel map exactly matches the source pixel map. This is not practical, but it illustrates the difference between draw engines and codecs. One sets pixels, the other compresses colour space.
Here is an interesting thing about codecs. It is only in the last 20 years or so that dsp's and SIMD have made the number crunching possible to let us compress 4K video. Before that, draw engines were the way to go. Like X server. Or, OpenGL. The idea of making images or sounds by specifying commands or parameters is not new, but it is useful.