Image models generating partially-eaten burritos over time
14 hours ago
- #AI-models
- #benchmark
- #image-generation
- The benchmark for image generation models was inspired by the horse riding astronaut meme from 2023.
- Simon's Pelican benchmark keeps the idea alive, despite testing different modalities.
- Burritos are humorously deemed more important than pelicans and equestrian absurdism.
- Initial surprise that the model couldn't replicate the image well, assuming similar examples existed in training data.
- The concept is weird because ingredients in a burrito get smushed and congealed.
- All images were generated using fal defaults, with the prompt: 'A partially eaten burrito with cheese, sour cream, guacamole, lettuce, salsa, pinto beans, and chicken.'
- Better prompting is possible but considered high effort (HIL) and feels like cheating.