Left: Normal ASCII shader
Right: Shape-aware ASCII shader
ASCII characters are chosen based on their shape, so you get clearer edges
Vectors are calculated and compared with WebGPU compute shaders
It's based on an altered version of Alex Harri's approach. I'll link below
Left: Normal ASCII shader
Right: Shape-aware ASCII shader
ASCII characters are chosen based on their shape, so you get clearer edges
Vectors are calculated and compared with WebGPU compute shaders
It's based on an altered version of Alex Harri's approach. I'll link below
The way I build the vectors is slightly different:
The screen is divded into cells. Then I divide those cells into a subgrid. The average luminance of each cell in the subgrid is what I use to build the vector
I think the math also differs in some places because it all runs in compute shaders
alexharri.com/blog/ascii-ren…
Just to be extra clear this is not my blog post. This is written by Alex Harri. I couldn't find him on X so I can't tag. This is why the original tweet says "based on Alex Harri's approach"
I should add it's not necessarily because of the AI. It's because of the users who don't know what to ask for
Once or twice is okay, but repeat that little mistake thousands of times and you have an unfixable mess
The reality is that if you don't know what you're doing, you won't know how or when you're digging yourself into a deeper hole
And Claude doesn't help, it responds to everything you say with "Great idea! That's actually really clever! You're so smart!" so you can't even be saved
another plus is that if you combine a shadow border (i.e. `ring` in tailwind), with a normal shadow, they'll blend
it looks a lot nicer than a normal css `border` with a shadow
LLMs try to make their own lives easier by adding "use client" to the top of every file and preferring to use hooks in client components
they've seen this hooks-based code in their training data way more than RSC code
this is destroying SEO for new sites because nothing is SSR'ed. the client receives nothing in the HTML and content only renders via JS
some teams are unaware that this is even happening to them because they're not reviewing LLM-written code
seeing many web apps move away from nextjs and switch to react-router
I think many teams will switch to tanstack start too, but it's a little too early for me to have found any examples in the wild