ChatGPT Images 2.0 generates valid Rubik’s Cube on mirror
ChatGPT Images 2.0 (Pro) produces physically accurate image of Rubik’s Cube resting on horizontal mirror after one move from solved state. Riley Goodside at Google DeepMind engineered the prompt. Output depicts three visible cube faces plus reflections of two more, with consistent legal color configurations on cube and mirror image, passing known complex test for image models.
First post
![ChatGPT dialog screenshot
User:
Create a photo of a Rubik’s Cube resting on a mirror placed horizontally on a table, in a state after exactly one move from its initial solved position. Three faces are visible, as well as the reflections of two of them. Create and use your own reference images via code, considering carefully the implied 3D geometry of this scene.
ChatGPT:
[Generated photo showing a validly positioned cube with green, orange, red, and white stickers visible]](https://pbs.twimg.com/media/HGr-eDZWUAAwwVp.jpg)
Why it matters
Magician Andrew Mayne invented a stage illusion for walking through a mirror years before he became OpenAI's first prompt engineer.
Riley Goodside (Google DeepMind prompt engineer) began publicly sharing targeted GPT-3 prompt demonstrations in 2022.
Riley Goodside reported that adding a second move from solved causes the mirror prompt to produce invalid color states.
All earlier Rubik's Cube mirror variants posted on X displayed illegal color configurations on the cube or its reflection.
5 more posts
AN
