In this experiment, I used Pikaso to generate 20 images with a provided command, with “AI prompt” on (that means Freepik AI will automatically improve short prompts).
Why? The generative model learns the patterns from the data and assimilates a deep understanding of the underlying structures. Consequently, the output it generates is a manifestation of its accumulated knowledge and insights it learned from the training data. This means that the generated content is a representation of the model’s interpreted understanding of the input data, reflecting the intricate patterns. In fact, in recent years, there have been papers on explaining models using Generative AI like Explaining image classifiers by removing input features using generative models (ACCV 2020).
So, here are some fun conclusions from the experiment:
Vietnamese loves selfie
Command: “cartoon Vietnamese girls wearing traditional ao dai dress by the pond.” Results: 3 of the generated images are about selfies:
This didn’t happen when I replaced the pond with the mountain. Maybe they’re too tired after climbing a mountain to take a selfie, or maybe no one wear Ao dai to climb a moutain ?. (So, maybe you should try ?)
Taiwanese loves dancing
Command: “cartoon Taiwanese wearing traditional dress by the pond” (6 out of 20 images)
Thai people like Zen
Command: “cartoon Thailand girls wearing traditional dress by the pond” (4 out of 20 images)
Discover more from Science Comics
Subscribe to get the latest posts sent to your email.
