The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Prove You're Human: a Halloween captcha that trains a ghost
Built with the Rive CLI and GPU Canvas.
It starts as free Wi-Fi at a guesthouse in the Himalayas. A friendly helper called Momo asks you to prove you're human: select all squares with mountains. Then feet. Then a human. The last picture pulls back, and you see she was standing in every one of them. Whatever you pick is "Correct". The captcha thanks you for the training data, plays your own clicks back in reverse, and every square you picked flips over into a piece of her. Then she ticks "I'm not a robot".
Play again and the check changes. There are five, and each one teaches her one more human thing:
Look: select the squares.
Move: slide the piece to fit. The second slider runs backwards, and she replays your drag.
Talk: type the word you see. She types it back the right way round: OPEN THE DOOR.
Find: click in order, ending with "you". She walks your route, footprints pointing the wrong way.
Final: select all squares with Kichkandi. She is not in the picture any more.
Kichkandi is a ghost from Nepali folklore. You can only recognise her by her feet: they point backwards.
Rive CLI + GPU Canvas
The scene is RML generated by a Python script and built, checked and published with the Rive CLI (--verify, scripted --pointer/--advance playthroughs of all five checks, push, --publish=web). The agent was Claude Code.
GPU Canvas runs two WGSL shaders:
captcha.wgsl draws the captcha board from my drawing: square selection, the pans between rounds, her eyes catching light, the tiles flipping into her portrait, the slide-to-fit puzzle piece, her footprint trail, and the blank paper where she used to stand.
snow.wgsl is the night outside: Himalayan ridges, prayer flags, snow that slows and then rises as the dread grows, and her footprints in the snow coming closer every check.
Luau scripting runs the game (five checks, input, replay of your own clicks and drag) and feeds both shaders.
Data binding: 68 view model properties drive the UI, the shaders and Momo.
A state machine gives Momo six moods (idle, happy, unsure, scared, gone, evil), switched by data.
Human touch
Both pictures are my own hand drawings: the village scene is the captcha photo, and the full figure is her portrait.
Sound is synthesised in code (an original music-box tune, drone, stinger, knock, footsteps). No stock audio. The voiceover is the Windows speech voice.
#rivehalloweenchallenge @rive_app
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Challenges
View all
Related posts
Here’s my submission for the Rive Halloween Challenge: Dusty the Dreamcleaner, a game I’ve wanted to make for 10 years, finally built in Rive.
Type for Treat! Get your Halloween goodies and practice your typing at the same time!
Play as a ghost (I promise it’s not a kid with a blanket over his head) going trick or treating around his neighborhood with his mom and dad.
Collect candies from each house by typing out words then catching them in your goodies bag. But be careful, some of your neighbors are giving away tooth brushes and tooth paste too, yuck!
Upgrade your gear with the goods from the vampire (AKA dad). He only accepts candy, so you better stock up.
If you need help, mom, I mean, the vampire queen is just around the corner with some tips!
-
Had a blast making this game and learned a ton. Super happy I went into this challenge despite barely having enough time. Excited to share more about what I learned in the future when things slow down.
All the designs come from Ivan Dubovik (https://www.instagram.com/vandubovik/?hl=en). All of the game mechanics + Shaders were made using Rive CLI in Cursor. Animations by me.
Hope you have fun with it, and let me know what your high score is!
Great job, Caleb! I gave it a try, but it’s really hard to play, even though I’m familiar with touch typing. I think the candies should fall more slowly, or maybe the bowl at the bottom should be removed to make it a bit easier.
I’m a long-time programmer, and I started in mobile game development in 2011, back in the iPhone 4 era. That experience gave me an appreciation for small interactions that feel good under your fingers. With Peel & Peek, I wanted to capture that feeling in something a little unsettling.
I’d known about Rive for a long time. The Rive CLI and GPU Canvas finally gave me the push to create my first piece with it.
Peel & Peek is a Halloween fidget toy, built with Rive CLI, Claude Code, and GPU Canvas. The grapes, eyes, and slime are drawn by a single fragment shader, without 3D meshes. Luau handles the interactions, a Rive state machine runs the cycle, and data binding lets me tune the details.