document.Write("<img src=width.png?" + document.innerWidth + ">")

That's a problem.

The W3C should probably create a new, rich spec hundreds of pages long so that frontend developers may instead declare images as a unitless set of point relationships to be rendered at any resolution without digital artifacts.

For example, instead of working on the pixel level, the developer would be free to simply declare, "an arc may exist in one of these four locations." Then, merely by declaring two further "flag" values, the developer can communicate to the renderer which three arcs not to draw, except for the edge case of no arc fitting the seven previously-declared constraints.

Just imagine-- instead of a big wasteful gif for something as simple as an arc animation, the developer would simply declare, "can someone just give me the javascript to convert from arc center to svg's endpoint syntax?" And someone on Stackoverflow would eventually declare the relevant javascript.

The browser can also ship with a pretrained GAN, so the site just asks for a picture of a cat and then the GAN creates one as needed, but nobody will know exactly which cat you saw.

That will be remembered as day when "Catwoman" went from being sexy to being weird and disgusting.

Now I want to train a GAN that takes images' alt text and reproduces the image.

I can see where this is heading. LOL

Yes, but how do you keep the client from then sending that data to the server?

You can't. Even if you block JavaScript, they can still get it from CSS media queries, where the can say which file to download given a certain screen size.

The best solution is to use the same screen size as everyone else so you don't stand out. And that's what this does.

Right, that’s what I’m saying. Keeping the data client side as ggp suggests is not possible.