006657b2ec

bitcoin

videos

lightning

Last week I shared an experiment: an AI that interviews you before generating images, videos, or 3D models, instead of expecting you to come up with the perfect prompt.

The feedback here was incredibly helpful. A few of you also took the time to try it, and watching how people interacted with it made one thing obvious:

It still felt too much like a chatbot.

The AI asked questions, but users were still expected to lead the conversation. That wasn't the experience I was aiming for.

So I went back and redesigned the interaction.

The AI is now much more proactive. It drives the conversation, asks better follow-up questions, and keeps digging until it has a clear picture of what you're trying to create.

The interface changed just as much. Instead of a traditional chat, you're only focused on the current step. Previous messages stay out of the way, the AI suggests answers you can select with a click, and you can always type your own response whenever the suggestions aren't quite right.

Only once the AI is confident it understands your intent does it propose generating the final artifact.

The goal is to make creating something feel less like prompt engineering and more like working through a creative brief with someone who's asking the right questions.

For now, I'm focusing on getting this workflow right for **text-to-image generation**. Video, 3D, and image-to-image generation are temporarily disabled while I finish integrating them into the new experience, but they're coming back very soon.

I'd really appreciate another round of feedback.

A few things I'm curious about:

* Does this feel better than a traditional chat interface?
* Are the suggested answers useful, or would you rather type everything yourself?
* Does the AI ask enough questions, or should it be even more proactive before generating?

You can try it here:

[https://3d.agentree.space/](https://3d.agentree.space/)

Thanks again to everyone who commented on the previous post. Several of the changes in this version came directly from your suggestions, and I'd love to hear what you think of this iteration.



The GIF below shows the flow from the first question to the generation proposal.


![](https://m.stacker.news/147218)

After your feedback, the AI that interviews you finally feels competent

devs

What if AI generators interviewed you instead of expecting better prompts?

yeah I was thinking about MCQ-style too. Its much easier to just click and select an option than  to write it from scratch

I'll take care of it before the next iteration, thank you for the suggestion

thank you for testing!

The issue should be fixed now

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/](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. 🙂[]([](

![](https://m.stacker.news/146595)

![](https://m.stacker.news/146596)

![](https://m.stacker.news/146597)

))