Marcin Ignac

How will the flower bloom in 2100? We collaborated with artist Annelie Berner to create a generative web tool that visualizes how the Circaea Alpina flower might evolve in response to future climate conditions. I've finally put together a v1 of full case study including info about data, algorithms, behind the scenes and whole look dev process. Video content to follow! #variablenature

@lusionltd @mlperego How much of that is C4D and how much is genAI? (Looks too good to be genAI TBH!)

Happy to confirm that oryzo-1-344b model runs on M1 Max. Early tests positive. Given the legacy support it's a no brainer enhancement for my compute capabilities.

@falcon_ide Will the recording be available? Missed the registration time...

join us for a hands-on shader workshop with falcon gx open studio • march 19 • new stadium attendees get beta access luma.com/5l3z4fl8

@edankwan Nice! Can't wait for my fav 19th month of the year! ;p

I can't wait to share what we've been working on. This is one of the biggest and most ambitious projects our team has worked on, and we have spent countless days and nights on it over the past year. We're getting close to the finish line. Stay tuned.

@_nonfigurativ_ Well there is hope you can WYSIWYG the details (the other 90% of work) with inspector panel, instead of asking agents to "make the logo bigger 3px, please".

Question: why would one need to vibecode a design file representing a web page when one could just vibecode the webpage?

@passivestar_ Which GI is that?

area lights in godot

Sometime I just chill with good old computational geometry and a bit of math. No agents, no AI, just me, sketchbook and trigonometry.

@alightinastorm @W11002Weed @threejs I'm curious here. It sounds like Linux "it's free if the value of your time is zero". Who will do all the admin, accounts, backups, uptime, support, device debugging, distribution, weird file format parsers, data security (not vibe coded open database with admin/12345) etc etc

@W11002Weed @threejs i think you don't even remotely understand that a $20 claude / gpt subscription and threejs can do much more

$799 per seat LMAO you get it for free and open source with my editor and @threejs SEVEN HUNDRED NINETY NINE PER SEAT Not even Adobe would do that

@Riyvir I guess it's a bit like notetaking apps and todo managers. Everybody needs 5 features but it's 5 different features for everybody. Making my own too! x.com/marcinignac/st…

what's your ideal inspiration process? you know i love cosmos, are.na, and pinterest. but i have problems with all of them. pinterest has a great discovery algo. but it's tuned for engagement, so once it starts feeding you a certain type of image you end up in a spiral that's hard to escape. cosmos is great when culture-drive curation is the goal. but heavy curation has a cost: you end up seeing the same references everyone else does. are.na is a wild child. no notes on that. but as much as i love quirky human curation, i also love an algorithm that feels like it understands how i think. i've tried mymind and i respect their commitment to no collaboration but i need collaborative features since i think the real unlocks happen when you combine personal thought, collaborative though, and algorithmic discovery. so this is why i'm building memio. memio is a local-first, creative memory companion. i've been tuning a similarity algorithm that’s been really fun to use. it's looser than pinterest and cosmos, which means it surfaces unexpected connections instead of just showing you more of what you already clicked on. the goal is for it to feel like a little like human curation. here's where memio is going: multi-modal. right now it handles images. text, links, video, and audio are all in the works. cli, mcp, and all the other acronyms too. private by default. your library lives on your computer or in your own cloud storage. the core algorithm runs entirely on-device, so you can do a lot without ever hitting an external api. collaborative when you want. you'll be able to choose which boards you share with other people and llms and that will extend what you can do in the base version. capture whatever. find connections you couldn't find any other way. make cool shit. i'm building this for all of us so tell me what you want from a tool like this!

To all my artist friends and everyone else choosing to spend their time reconfiguring matter and energy into beauty and meaning: thank you ❤️ The world needs it [Pictured: The Weather Project by Olafur Eliasson]

@ekimroyrp @threejs What machine are you running this on? Chrome/Safari on M1 Max are not doing very well with growth sim.

Differential Growth web app - Try it out: ekimroyrp.github.io/260308_Differe… - #differentialgrowth #threejs @threejs

In an alternate timeline we’d be using Evangelion GUI designs rather than CLIs

@edankwan But is there such thing as AI demo without the prompter? Somebody has to instruct the agents, no?

Would AI hold up in demoscene competition? Like demoscene battle AI vs human.

@alex_barashkov You could at least use different colors than ksawerykomputery.com/works/generati…

We’re shipping an insane number of mini apps these days just to deliver branding + website work. Here are the prompt keywords I keep reusing to build them fast : Intro For one recent project (one client, branding, and a few-page website), we built: - 3 small apps just to generate different parts of the hero motion reel - 2 different ASCII generator apps (custom look + custom behavior, not the usual "ascii effect" from publicly available tools) - an OG image template generator I see this trend continuing, and most likely every new project we build will be stuffed with 5-10 personalized software solutions. Prompt keywords (with what they do): "Create React app, use Vite" - quick starting line: sets up the frontend stack + build tool. If it sounds scary, ask AI what it means. "Use shadcn components, prefer Base UI over Radix UI. Install components from the official registry (not random snippets), check the latest version. If you can’t find a good match, try alternatives." — gets you clean UI fast with solid base styles. "Use react-three-fiber" - for 3D scenes and visual-heavy work (effects, complex hero visuals). "Write a WebGL-based animation without using any additional libraries" - good for small custom shaders: mesh gradients, noise, swirls, waves, pixelated effects. "Use motion dev for …" - great library for UI animation, my default choice when it’s not complex VFX work.

@cobaltdigital33 @alex_barashkov Here you go ksawerykomputery.com/works/generati…

@alex_barashkov could I peep the code/prompt that was used to create the software showcased in the post? it's very beautiful

@ekimroyrp @graftoverflow @threejs Default on the web. Click and drag. Every 3d desktop tool is different so I have no hope there.

@graftoverflow @marcinignac @threejs Well default where? Rhino and Zbrush for example is RMB (Rhino what I use most) 😃

Custom Bismuth Spaces generation web app - Try it out: ekimroyrp.github.io/260301_Bismuth… - #bismuth #threejs @threejs

@mrdoob How do you manage leaks?

Exploring irradiance probes... 👀 #threejs

@involvedenvi Yes, it was left for interpretation : What do YOU see?

@marcinignac Absolutely. That was one of my favorite. There was so much mysterious beauty in it, maybe more like machine level.

genAI was so much more fun when it could not do realistic things. Midjourney 2022-03

@CedricKiefer @techno0ptimist @variable_io If you look at many of image / references web apps they are all "upload your images to our cloud, we will analyse them for you". 300GB, sure, yeah, yhm.

@CedricKiefer @techno0ptimist @variable_io Sorry for confusing so many people. Screenshot database and analysis is local. CC is normal SaaS.

Vibecoding myself the ultimate screenshot manager. 108k images from over 14y of @variable_io practice. Color extraction, features extraction, similarity search. Done in Claude Code in 2h spanned across today. 🤯