May 2, 2026
Collapse now, arguments later
Bitmap and tilemap generation from a single example
This old image-making trick is back, and the comments are split between awe and eye-rolls
TLDR: WaveFunctionCollapse turns one small sample image into bigger matching pictures and game maps, and it’s still widely used years later. Commenters were torn between admiring its staying power and mocking its dramatic quantum-sounding name, while others debated whether it’s an early cousin of modern AI image tools.
A quirky old program that can grow whole pictures and game maps from just one example image is having another moment online, and honestly, the real spectacle is in the comments. The tool, called WaveFunctionCollapse, starts with a blank grid and fills it in piece by piece so the final result only uses tiny visual patterns found in the original image. In plain English: show it a little sample, and it tries to make a bigger thing that feels like it belongs to the same world. That’s why game developers love it for textures, tiles, and level ideas.
But the community instantly turned this into a mini drama about branding, age, and AI comparisons. One veteran commenter was downright sentimental, noting it was posted a decade ago and celebrating that it’s still alive and useful in real workflows. Meanwhile, the spiciest eye-roll came from people side-eyeing the name itself: “Wave function collapse” sounds wildly futuristic, but critics called it a fancy label for something much simpler and not actually quantum. Ouch.
Then came the inevitable 2020s discourse: is this a baby cousin of image generators like Stable Diffusion, or a totally different species? That question lit up the thread with curiosity more than combat, alongside practical nitpicks about whether the app needs a big .NET runtime and a delightfully nerdy request to take the whole thing into video. So yes, the algorithm is clever—but the comments turned it into a reunion, a roast, and a brainstorming session all at once.
Key Points
- •WaveFunctionCollapse generates bitmaps or tilemaps from a single example by enforcing that output NxN patterns must already exist in the input.
- •The algorithm alternates between observation and propagation, selecting the region with the lowest nonzero Shannon entropy and collapsing it according to input pattern frequencies.
- •WFC can encounter contradictions when no valid coefficients remain for a region, and the article states that deciding whether valid nontrivial outputs exist under its constraints is NP-hard.
- •The article provides a step-by-step algorithm, including counting input patterns, optionally augmenting them with rotations/reflections, initializing the wave, and repeating observation-propagation until completion.
- •WFC has been implemented in many programming languages, adapted to major creative and game-development tools, and used in games such as Bad North, Caves of Qud, Dead Static Drive, Townscaper, and Matrix Awakens.