My First AI Design Concept Started With an Empty Room

 

My first AI design concept did not start with a style name. It started with an empty room photo and one simple question: what could make this space easier to discuss?

That question mattered more than I expected. It kept me from asking for a vague "better room" and helped me judge each image against the real space. I was not trying to create a final interior design. I wanted a useful visual starting point.

Why I started with an empty room

I am a city planner by training, and I have spent plenty of late nights working through drawings and visual options. That experience shaped how I approached ArchOne AI. I wanted to understand whether an AI design tool could help with early concept conversations without pretending to replace the design work that follows.

I am also building ArchOne AI as an independent developer. This experiment was both a design study and a way to test whether the product helps me make a clear decision instead of only producing attractive images.

An empty room is a good test because it leaves many decisions open. Furniture, materials, lighting, color, and layout can all change. At the same time, the room still has fixed features that matter: walls, openings, floor, ceiling height, and the camera angle.

That made the experiment concrete. I could ask the tool to explore a direction while checking whether the room still looked like the same room.

I started with a real room photo so the concept would stay connected to a real space.

The question I used

I kept the first question narrow:

How could this room feel warmer and more usable while keeping its main features recognizable?

This was more useful than asking for a style such as "modern" or "luxury." Those labels suggest a mood. They do not explain what should change or what should stay.

I also wrote down the features I wanted to preserve:

• the wall and window positions;
• the doorway and main circulation;
• the floor and ceiling proportions;
• the camera angle;
• the overall room shape.

Then I described the parts I wanted to explore: furniture arrangement, warmer materials, lighting mood, and a more comfortable setup for everyday use.

This distinction became the most important part of the test. "Make it better" describes a hope. "Keep the window, doorway, floor, and camera angle while testing warmer materials and a clearer furniture layout" describes a study.

What the first concepts showed me

The first useful result was not necessarily the most attractive image. It was the option that made a decision easier to talk about.

One concept suggested a warmer material direction and a more defined dining area. Because the source and concept showed the same space, I could compare the dining zone, the lighting mood, and the main circulation. I still had to check whether the arrangement would work in practice.

The concept became useful when it helped separate a material idea from a layout idea.

The image was not a measured plan. It did not tell me whether the furniture would fit, whether the lighting was possible, or whether the materials were suitable for the project. It helped me frame the next conversation.

That is also why I keep a person in the review loop. NIST's AI Risk Management Framework discusses the need to define human roles and responsibilities when people work with AI systems. The AIA AI Task Force similarly emphasizes professional judgment, accountability, and data governance in architecture.

Three lessons from the experiment

1. The photo matters more than the style label

If the source image hides the window, doorway, or room edges, it becomes harder to judge the result. A clean, useful photo does not need to look like a professional shoot. It needs to show the part of the space that affects the decision.

2. "What should stay" belongs in the prompt

I used to focus on describing the look I wanted. Now I also describe the features that should remain. This makes the result easier to check and reduces the chance that the image becomes a different room with a similar mood.

3. A few options are better than one "perfect" image

One image can look convincing and still leave the design question unanswered. A small group of focused options makes comparison easier. For example, I could keep the room layout stable and compare two material palettes, or keep the materials stable and compare two furniture arrangements.

What I would do differently next time

Next time, I would define the decision even more clearly before generating anything. I might choose one of these questions:

• Should the room feel warmer through materials or lighting?
• Does a smaller furniture arrangement improve circulation?
• Which color direction makes the room easier to connect with the adjacent space?

I would also record why one option worked better than another. Without that note, it is easy to collect images and forget what each one was meant to test.

For a repeatable process, I would use this short checklist:

1. Choose one design question.
2. Upload a photo that shows the relevant fixed features.
3. Write what must stay and what can change.
4. Generate a few focused options.
5. Compare them with the original question.
6. Write down the next decision or the missing information.

This is the same sequence I would follow in an AI interior design  workflow. ArchOne AI's guide to using AI for interior design covers the broader process. This diary focuses on the decisions I make before and after generating an image.

The boundary I want to keep clear

An AI concept image can help with early ideas, option comparisons, and conversations with a client or design team. It cannot confirm structure, code compliance, drainage, accessibility, electrical work, measurements, permitting, or final material choices.

That boundary is not a weakness of the experiment. It is what makes the experiment useful. The image helps me decide what deserves more attention. It does not make the later decisions for me.

My first concept taught me that the best prompt is not the one with the most style words. It is the one that connects a real question to a real space and gives me a clear way to review the result.

If you want to try the same kind of small study, you can start in the ArchOne AI  design workspace. Start with one room photo and one question. Then judge the image by whether it helps you decide what to do next.

What would you test first: layout, materials, lighting, or something else?