Zimo

watching AI rewrite the rules.

[Zimo @Zimovis]

SOMEONE TOLD HIM HIS PC LOOKS OLD AND UGLY. HE OPENED IT UP TO PROVE THEM WRONG.

From the outside it’s a beige tower straight out of 1998, dusty vents, a DVD tray that hardly opens, screams garage sale find

Then he cracks the side panel open and this is what’s actually inside…

A modern high-end GPU, a full liquid cooler, and a tiny LCD screen right on the cooler pump showing live temps in real time. Ancient shell, brand new guts.

The screen isn’t just for looks either — that kind of live monitoring is what lets people fine-tune cooling curves and catch overheating before it throttles performance, especially on GPUs running local AI models now.

This is the ultimate sleeper build: judge nothing by the case.

[Zimo @Zimovis]

THIS GUY BUILT A REAL MARS ROVER SUSPENSION SYSTEM IN HIS OWN LIVING ROOM.

Same rocker-bogie design NASA uses on actual Mars rovers, built with a soldering iron, some scrap metal, and a lot of trial and error.

Then he pushes down to test the suspension, and watch what happens…

Every wheel moves independently, keeping the chassis level even when the ground underneath is completely uneven. That’s the exact same trick that lets Curiosity and Perseverance climb rocks on Mars without tipping over.

The two glowing blue modules up front aren’t just LEDs — they’re camera housings, likely feeding an AI vision system to help the rover navigate on its own later.

Space-grade engineering, home workshop budget.

[Zimo @Zimovis]

I ASKED AI TO CODE A GALAXY SIMULATION IN JAVASCRIPT, JUST FOR FUN.

Just a prompt describing particles, randomness and a glowing black hole at the very center. No 3D modeling, no manual placement of any single star.

Then I hit run, and this is what actually rendered on screen…

Thousands of glowing particles spinning into a spiral galaxy with a black hole core, all generated with Three.js and pure math — random X, Y, Z positions pulled toward a center point with a randomness curve.

No artist touched a single pixel. The AI wrote maybe 15 lines of code, and the GPU did the rest, rendering it live in the browser.

This is what happens when code replaces the paintbrush entirely.

[Zimo @Zimovis]

THIS GUY USED AN AI TO DESIGN A PC CASE, THEN BUILT THE WHOLE THING BY HAND.

Hand-engraved metal panels wrapped in filigree. A glowing spiral core slowly spins at the center. Purple light pours out of every gap.

Then you find out where that elegant engraving pattern came from…

He generated the entire filigree design with AI, then spent hours carving every line into aluminum by hand. The spiral core spinning inside isn’t decoration either — it’s the GPU and water loop, wrapped in a rotating LED strip like a mini reactor.

AI handled the art, a human handled the metalwork. That split is becoming the new normal in custom PC builds — AI drafts the pattern, hands execute it.

Function meets fantasy, and somehow it still boots.

[Zimo @Zimovis]

I BUILT A LASER TURRET THAT LOCKS ONTO ANYTHING THAT MOVES IN THE ROOM.

Two motors, a webcam, an Arduino, a breadboard full of wires. Nothing fancy, just random stuff I had lying around, put together over one weekend.

Then I put my hand up to test it and things got weird fast…

Both turrets snapped onto my hand instantly and just… followed it. Every twitch, every finger movement, tracked in real time. No delay, no jitter, just pure lock-on.

It’s basically computer vision plus two servo motors talking to each other 60 times a second. The camera finds you, the code calculates the angle, the motors move — that’s the whole trick.

This is the same base logic behind auto-tracking camera gimbals and turret sentries in games.

[Zimo @Zimovis]

I TURNED A RANDOM CCTV CAMERA INTO A SPEED RADAR USING AI.

No lidar, no sensors, no special hardware. Just Python, OpenCV, and object tracking watching live traffic in Bangkok, in real time.

Then one car blew through the busy intersection so fast the code instantly flagged it…

100 km/h. In a zone packed with motorcycles and pedestrians. The AI didn’t need a fancy sensor — it tracked each car frame by frame, measured how far it moved between frames, and did the math to get real speed, all from a normal video feed.

This is the same core trick self-driving cars and traffic cameras use. Track the object, measure the pixels, convert to real-world distance, repeat 30 times a second.

The hard part isn’t the math — it’s keeping track of the same car when it’s half-hidden behind another one for a second.

Would you trust an AI camera like this to catch speeders on your street?

[Zimo @Zimovis]

WE GAVE 4 AI MODELS THE SAME CODING TEST.

One prompt: build a self-solving 3D Rubik’s Cube, single HTML file, Three.js, no other libraries, no help, no hints, just raw code from scratch.

Three of the four models nailed it. But one model’s cube literally never assembled itself…

The one that broke? Claude. ChatGPT, Gemini and DeepSeek all rendered a clean cube and solved it move by move. Claude’s version just floated apart — glitched panels, no structure, never came together.

This is the real AI test. Anyone can write a nice paragraph. Rotation matrices and 3D geometry don’t lie — either the logic holds or the whole thing falls apart on screen.

That’s why “vibe coding” benchmarks are getting more brutal than chat leaderboards. One wrong transform and your cube turns into confetti.

Which AI do you trust for actual coding work? 👇

[Zimo @Zimovis]

ONE OF THESE AI MODELS IS BLOWING MONEY FOR NO REASON.

Fable 5 and Opus 4.8 were tasked to code complex physics, double pendulums, and particle collisions live in real time.

Both simulations look almost identical on screen, until you actually look at the price tag at the top...

Opus 4.8 solves double pendulum trajectories, Galton board probability distributions, and fluid particle simulations for under a dollar.

Fable 5 burns over $3.30 on the exact same complex physics tasks with almost identical visual accuracy.

Is Opus 4.8 currently the most cost-effective LLM for code-driven animation?

[Zimo @Zimovis]

THIS ANIMATION ENGINE TRANSFORMS CODE INTO A GLOWING HEART SHAPE.

A developer uses pure JavaScript to build a clean particle effect that renders seamlessly in real time on screen.

It looks like standard 3D web graphics, until you see how AI generated the whole math matrix...

The script blends Three.js vector coordinates with GSAP timeline curves to control each particle's movement.

AI coding assistants generated the vector calculations and easing parameters automatically from a text prompt.

Will AI completely replace manual 3D particle animation?

[Zimo @Zimovis]

THIS INSANE AI VIDEO GENERATION SIDE-BY-SIDE SHOWS THE FUTURE OF ANIMATION.

Two massive AI video models generate complex samurai physics, ink transitions, and camera movement in real time.

Both models render stunning anime visuals, until you closely check every single motion...

GPT 5.6 Sora/Sol and Kimi K3 process fluid dynamics and stylized camera turns with impressive temporal consistency.

The models maintain character armor details and stylized ink effects across rapid frame transitions without distorting the scene.

Which AI model created the cleaner cinematic animation in this side-by-side test?

[Zimo @Zimovis]

THIS INSANE PHONE FARM IS USING AI TO CONTROL HUNDREDS OF DEVICES AT ONCE.

An entire room packed with massive server racks full of real smartphones running non-stop 24/7.

It looks like a regular datacenter setup, until you see what the automated AI software is actually doing...

It’s an advanced hardware farm where custom AI scripts simulate human actions across hundreds of Android screens simultaneously.

By pairing physical devices with AI vision models, operators bypass emulator bans for account automation and ad engagement.

One operator with an AI controller can now simulate an army of real active users.

Will AI-powered phone farms permanently break how we measure real internet traffic?

[Zimo @Zimovis]

THIS DIY MACHINE IS LITERALLY DEFYING PHYSICS IN REAL TIME.

A creator built a custom rail setup that catches a falling metal rod much faster than a human brain can process.

It fights gravity continuously right on screen, until you step back and see what's actually driving it...

It uses an optical encoder and a custom AI control loop to track the pole's angle and move the motor hundreds of times per second.

The control script is running live on the monitor in the background to adjust the position and keep it perfectly vertical.

Is self-balancing AI hardware the most satisfying DIY engineering project to build?

[Zimo @Zimovis]

THIS LOOKS LIKE A REAL SCI-FI ROBOT, BUT THE TRUTH IS MUCH CRAZIER THAN YOU THINK.

The entire crowd was convinced this was a real autonomous AI walking around.

Everyone was filming in shock, until they saw what was inside...

It’s Titan — a giant robot suit controlled by a human inside using hidden electronic joints.

Now, garage creators are using simple 3D printers to build the exact same giant robot suits for cheap.

Would you wear a suit like this or wait for real AI robots?

[Zimo @Zimovis]

an autonomous mobile robot (AMR) using computer vision and edge AI to slide under a pallet, lift it, and navigate the warehouse.

real-time spatial AI handles pallet alignment, obstacle avoidance, and dynamic pathing without any human input.

the future of industrial logistics is entirely vision-guided and AI-driven.

[Zimo @Zimovis]

this guy hooked up an IMU sensor to his finger to control a mecanum wheel robot with hand gestures.

he used a lightweight ML model on the chip to filter out hand jitter and recognize smooth gesture patterns in real time.

tilt forward to drive, roll sideways to strafe, rotate your wrist to spin.

no bulky controllers — just tiny sensors paired with edge AI for seamless motion tracking.

[Zimo @Zimovis]

this creator just built a physical AI companion named Kirin on an $8 microcontroller.

a tiny breadboard setup, a small speaker, and an OLED display showing animated eyes — running a local voice pipeline with real-time response.

we went from microcontrollers running simple sensor loops to putting fully vocal AI agents on raw edge hardware.

you don't need a $1,000 phone or a monthly subscription to run a personal desktop agent.