Even the same “slice of lime” can be interpreted so differently depending on the AI.
4 models: Gemini 3.7 Flash, Kimi K3, Claude Opus 5, GPT-5.6 Sol, “Give them a piece of lime and draw a pose of a human being integrated with the lime.” and enter the exact same visual prompt. All four models produced completely different expressions.
Updated the DBZ runner game by giving Goku the flying ability and more cleanup!
Stack: usecrayon new pro harness, Threejs, Opus 5! Disclaimer: This is a fan project made out of my love for DBZ, no intention of harming the IP.
Gemini 3.7 Flash, Kimi K3, Claude Opus 5 and GPT-5.6 Sol each got a slice of lime and the same prompt
Draw a human pose integrated with the object. Four completely different results. Clear window into how each model sees.
Opus 5 vs. Kimi K3
OpenAI Kimi_Moonshot
You have reached the end of the archive
All of Claude Opus 5