Conversation

How would you generate prompts for a relatively coherent metaverse of dalle2 objects? Eg: maybe your theme is “fruits and vegetables” and you do stuff like: Tomato office building Root vegetables highway Strawberry dungeon This is clearly a new kind of job. Wonderland Conjurer
8
17
Replying to
I’d be curious to see if it understands perspective enough to generate images of the same scene from different angles. Could eventually try to turn those into 3D models.
1
1
Replying to
That doesn’t sound like a significant extra step. You wouldn’t use GPT though… probably just trad GOFAI and 3D algorithms. I can imagine rotating the wireframe from one perspective and getting GPT to fill in the missing details.
Replying to
I’m more just curious about dall-e’s ability to maintain a consistent representation. But one could prompt engineer something that outputs some number of perspectives and and 3D reconstructs the scene from them. Some form of pg possibly. Just depends on how good dall-e is at that
1