One thing I've noticed with AI image, video, and 3D generation is that people often know what they want, but struggle to describe it in enough detail for a model to produce a good result.

My hypothesis is that the AI should do that work instead.

So I built an agent that starts by asking questions. It keeps digging until it has a clear understanding of the final artifact you're after, then it generates it.

Instead of trying to write the "perfect prompt," you just have a conversation.

At the moment, it can generate:

• Images
• Videos
• 3D models

I'd love to know whether this interaction feels more natural than the usual prompt-engineering workflow, or if you'd rather stay in direct control from the beginning.

You can try it here:
https://3d.agentree.space/

For now I'm keeping the project intentionally small while I learn from early users. Access is handled over the Lightning Network, and there are currently five early-access spots available (starting at 100 sats) simply so I can iterate without opening the floodgates all at once.

Some things I'm genuinely curious about:

• Would you rather answer questions than write a detailed prompt?
• How much questioning is too much before you'd rather the AI just starts generating?
• Are there situations where this approach would be clearly better, or worse, than traditional prompting?

I'm looking forward to hearing what people think. If the idea is flawed, I'd rather find out now than after building more on top of it. 🙂[]([](

))