@Thrakomania this was also over the course of a month, Always using extra high and max for most of it.
Knowing more about fluids helped, but nothing that claude didn't already know.
i've easily used billions of tokens on both the blood and gore
@Thrakomania lots of testing and feedback about fluid behavior
writing more than 3-4 paragraphs per prompt
encouraging claude not to "pioneer" / it should create something heavily backed by dozens of white papers & research
Telling it not to "go by memory".
Performance last concept first
@cutmyinital a week ago i bought another pro subscription on an alt and swapped between the two. this helped
writing very long prompts back to back
loads of testing. running local lua simulations of the system
Forcing claude to research CFD math and compare to datasets and real sim results
@Darkheart618 just disclaiming: its fully made with AI. i am not smart enough to write 8K lines of calculus. It only took me a month. this could've taken a solo developer or studio years. If this is still really interesting to you anyways, thanks.
@cutmyinital yes its entirely by claude.
I cant imagine the years of shear developmental resources required to ever create such a system from scratch inside of a game engine popular for creating steal a brainrot. Opus 5.5 has been the only model to understand exactly what I wanted to create.
@Chlamedea I did it on roblox because its way more fun and interesting to get around their limitations. In other engines, all i literally have to do is drop in a open source fluid library and optimize it. Roblox doesnt have this, so I wanted to see if it could be made from scratch.
@contextttttt I'm just not sure what to make the discord server about lol.. right now its just a system and tech showcase.
if enough people wanted one, then sure. but it would be a placeholder for a larger idea
@JeanCarmac I definitely want to add this. might be the next thing i will add tomorrow because Its been mentioned a lot in the past. If it works the way i imagine it, i should be able to make a timelapse.
@mo74007668 running it in a full game with players is the goal, but it uses so much client memory and CPU. optimizing it mostly means sacrificing visual accuracy.
less math per frame gives you better performance, but will usually give worse results, so thats the problem im facing
@cj_petrichor@orcartz it was taken down once because i didnt have the right rating so i have to make it 18+. it could get taken down again if it ever became popular.. maybe
the blood and gore aren't uploaded assets though and if i make it an actual game with combat and not a sandbox it could be fine
@Peteralm_ All the scenes in my screenshots should be able to reach 144fps if i pause the simulation entirely. But i will have to check this since i haven't added a way to do this yet. Less blood being spawned and less complex geometry can cost nothing.
@Peteralm_ spraying tons of blood on a clean floor can spike to 17-20fps, 40-70ms luau. as the puddle gets larger, its cheaper, but constant streams of blood is very expensive. as the blood rests or no longer spawns, it literally costs nothing, and you can easily idle at 60fps.