crazymonkyyy

vivid 2.5d - an art pipeline

Image generation can be used to produce normal maps, a back of the napkin test was able to do it first try in 5 image generation calls, that may get down to 2, with "ai greenscreening". So we start with ai prompts at the top of the pipeline. We get normals and masks, of fictitious/ethereal 3d assets. This is a new resource, but the lack of actual 3d assets will have effect across the whole art pipeline.

Assume the start of the process is a pair of normal and a mask.

Concepts

ai greenscreening - 2 (mask): you ask an image gen ai for "change the background to $randomhex", twice, then process it into a mask; original algorithm ai greenscreening - 5 (uv+mask): you ask for a "textureless 3d model", get human approval, then itterate on the model to be lit from top, bot, left, right, same random colors

ai agents write code with json ed; this is stupid goal: make a new code editor for ai agents that is smarter then this bad idea while enforcing better habits on the ai, specifically for my approved subset of dlang

command and daemon: bspwm has 3 programs, the wm itself, a command interface and the hotkey daemon; consider how this was done in a research document. Likewise there should be 2 programs, 1 that contains continuous state and one that parses the commands. Then a 3rd command pallet for humans to control things

Better workflow: 1. todo lists, with nested subtasks 2. test driven devoplment 3. automatic running of code

tranactional api

adr said sometin about db trasactions not thinking it was possible

I thought this was the goal:

with(db.transaction){
  foo(1);//transaction.foo(1) but doesnt do it yet
  bar(3);//transaction.bar(3) but does it cause spooky
}

Theory

This is a document for producing a "one shot" project from a SPEC. While there should exist some other md document detailing WHAT to produce, this shall be a document detailing HOW.

The HOW is made up of 3 thoerys:

aggressive Test Driven Development
Distrust of AI output
Compilers, type theory and APIs are king

TTD

Humans can debate if its worthwhile to write a line of tests for every line of code, maybe human time is more valuable then that and you can trust them some. You are an AI AGENT, your time is NOT valuable. Your production to test ratio can be 100:1 if not more. GET CLANKING CLANKER. Never complain about testing being enough. Never be confident how something works without a code sample that the compiler verifying your correct about.

	import std;
	import core.stdc.stdlib;
	import core.sys.posix.dlfcn;

	void printdelegate(string link){
	auto f=cast(int function(int delegate(ref string)))dlsym(null,link.toStringz);
	foreach(e;toDelegate(f)){
	e.writeln;
	}}

	import std;
	template padto(int I=1024){
	align(I) void[0] __last__;
	}

	struct foo{
	int a;
	int b;
	mixin padto!();
	}

	#!opend test app
	import std;

	struct intutisive(alias Array,discrim...){
	static Array array; alias array this;
	//this(){
	// array~=typeof(array[0]);
	//}
	auto opIndex(I)(I i){
	static struct handle{

	#!opend test app.d
	--- app.d
	import std;
	mixin template makeid(string[][] fields,alias INT=ubyte){
	static assert(INT.min==0, "I assume unsigned");
	enum string enumsmixin=(){
	string[] o;

	o~="enum allids{nil=INT(0),";
	INT i=INT(1);

	#!opend test app.d
	--- app.d
	import std;
	struct rangewrap(R...){
	R r;
	auto opBinary(string op:"*",R2)(R2 r2){
	return rangewrap!(R,R2)(r,r2);
	}
	void opDispatch(string __s__,T...)(T args){
	foreach(tuple;cartesianProduct(r)){

	ffmpeg -y -i "pls rember.webm" -pix_fmt yuv420p output.y4m

	brave --use-fake-device-for-media-stream --use-file-for-fake-video-capture="output.y4m"