November 25, 2025
Guac or yuck? Burrito-gate begins
Image models generating partially-eaten burritos over time
AI serves half-eaten burritos—community split between cravings, creepiness, and 'cheating'
TLDR: An AI tried to show burritos getting eaten over time using default settings, splitting reactions between queasy and hungry. Accusations of model “cheating” and love for a standout called “nano banana” spotlight how messy food reveals AI’s limits and sparks fierce trust debates.
An AI experiment tried to generate photos of burritos getting eaten over time—yes, a literal timeline of bites. Inspired by the old astronaut-on-a-horse meme and a pelican test, the creator insisted burritos are more vital to civilization than birds or space cowboys. Using default settings on a tool called Fal (no fancy tricks), they asked for a messy, partly devoured burrito loaded with cheese, sour cream, guac, lettuce, salsa, beans, and chicken. The twist? The AI struggled with the chaos of smushed ingredients—and the crowd went wild.
The comment section turned into burrito‑gate. One camp felt queasy: “This is spooking our appetite,” cried namegulf, while another confessed hunger pangs, with ilaksh nearly DoorDashing on the spot. Then the drama: skocznymroczny accused a pic of looking “way too good” for plain Stable Diffusion 1.5 (a popular image tool), hinting at secret finetunes—cue the cheese conspiracy. Meanwhile, the star of the thread became nano banana, a model some said was the only one to nail “partially eaten,” with adammarples demanding to know its “secret sauce.” Jokes flew about guac-and-order, and fans debated if better prompting is “cheating.” Verdict: this messy food test exposed how AI handles real-world goo—and how fast the internet can turn salsa into scandal.
Key Points
- •The experiment tests image models on rendering a partially eaten burrito.
- •It was inspired by the 2023 horse riding astronaut meme.
- •“Simon’s Pelican benchmark” influenced the approach, though it covers different modalities.
- •Complex, mixed food textures made accurate replication challenging.
- •All images were generated using fal’s default settings; no extensive prompting was used.