$0. That’s what it cost to build something studios charge $15,000 for.
Watch the clip closely. The camera tracks the player and his opponent on its own, a live court radar sits at the bottom like an NBA broadcast, and the second the ball drops through the rim, the system writes “MADE. BUCKET!” by itself.
A tracking overlay like this for a real sports broadcast runs $10–20k and a team of engineers. The sports computer-vision analytics market is already worth $6B and growing every single year.
This guy built the same thing in one weekend. On free tools. From a laptop.
And here’s the punchline — the basic version doesn’t take a weekend. It takes 5 minutes. Three libraries, a dozen lines of code, and you’ve got a full pipeline: see the player → track the ball → log the shot.
I broke the whole thing down step by step.
Want the “zero to working tracker in 5 min” walkthrough? Drop “COACH” in the comments and I’ll send the breakdown in a reply.
Save this post so you don’t lose it. A year from now this is a skill people pay for — not a “wow, neat” scroll. 👀
EVERY posture app fails for the same reason — and it's not detection.
you already feel your back hurting. you don't need AirPods to tell you. you ignore it anyway.
building a sensor to catch the thing you're already ignoring doesn't fix the loop. it adds a notification to it.
the problem was never awareness, it was friction
an alert mid-task gets dismissed like every other alert
now your AirPods nag you AND drain context-switching focus
in a week it's muted, like every health app you've downloaded
"noticing what your hardware can do" is a cool demo. shipping something people use at day 30 is the actual skill.
the hard part was never the build. it was retention. and a posture buzz doesn't have one.
xitter.cf/om_patel5/status/20681…
Only 3 seconds until 115,000 is drained.
xitter.cf/0xalpha__/status/20668…
I taught my webcam to stretch my face like Luffy from One Piece.
No AI. No After Effects. No filter app. Just Python.
Here’s exactly how it works 👇
The stack is free and runs in real time on a regular laptop — no GPU:
• Python
• OpenCV (webcam feed)
• MediaPipe (face + hand tracking)
First, the face. MediaPipe Face Mesh gives you 468 points on your face, 30 times per second. That’s the “skeleton” you deform — cheeks, lips, eyes, every pixel mapped to a coordinate.
Then the hand. MediaPipe Hands gives 21 points per hand. You measure the distance between the thumb tip (point 4) and the index tip (point 8). Small distance = a pinch. That’s your “grab” trigger.
When the pinch fires, you find the nearest face point under your fingers and lock it as the origin. Now your hand is physically “holding” your skin.
Here’s the magic — the warp. You build a displacement map: pixels near the origin get pulled toward your hand, with the strength falling off as you move away from the grab point (feather / falloff). Then cv2.remap() applies that map to the frame, and the face goes full rubber.
The detail that actually sells it: snap-back. When you release the pinch, the face springs back with an elastic ease-out. That’s what makes it feel like rubber instead of a broken photo.
A few sliders to dial it in:
• over — stretch strength
• feather — edge softness
• region — grab radius
• back — return speed
Tweak until it looks straight out of the anime.
That’s it. Luffy in one evening 🏴☠️
If this was useful — ❤️ + 🔁 so it reaches more people, and 🔖 bookmark it before it’s gone. Follow for more CV builds like this.
