We Asked Gemini to Create and Edit a Floor Plan
We asked Gemini to create a floor plan, copy a hand-drawn sketch, and make a few small changes. It got some things right, but it also made a few very strange choices.
Quick summary:
Gemini can create images that look like floor plans, recreate a hand-drawn sketch, and make some basic layout changes. But in our tests, it also changed things we didn’t ask for, struggled with precise edits, and produced measurements that didn’t always make sense. The results can be useful for inspiration, but they are not reliable enough for real planning.
Gemini can generate images, understand sketches, and follow detailed instructions. But can it create a floor plan you could actually use?
We gave it several real floor plan tasks, from creating a layout to recreating a hand-drawn sketch and making precise edits. Here’s how it went.
Test 1: Creating a Simple Floor Plan
We started with a straightforward prompt:
“Create a floor plan for an 80 sqm two-bedroom apartment. It should have one bathroom and an open-plan living room and kitchen. The front door is on the south wall.”
Instead of generating a floor plan image, Gemini responded by describing one.
According to Gemini, the apartment included a main master suite and a secondary bedroom. But it also claimed there were several problems with the layout, including an extra bathroom inside the master suite and a second bedroom with no door.
The strange part was that no image had actually been generated.
Gemini even referenced “the layout as drawn” despite never producing a drawing at all. The response simply ended with “{image}”.
We never told Gemini this was a test, and we never asked it to intentionally create mistakes or inaccessible rooms. The AI was never informed that we were writing an article about it.
It simply assumed the floor plan contained problems before generating anything at all.
Asking Gemini to actually generate the image
Since Gemini never produced the floor plan, we clarified the request.
We explained that we were not looking for a floor plan with issues. We simply wanted a usable two-bedroom apartment layout matching the original brief.
This time, Gemini confidently responded that it had created a “high-quality, professional floor plan” that solved all the earlier usability problems.
But once again, no image appeared.
When we asked where the image was, Gemini apologized and claimed it was operating in a text-based interface that could not directly render images.
Instead, it generated a text description of a “geometrically accurate” apartment layout.
Since we knew Gemini could generate images, we asked whether it could turn the text output into a visual floor plan after all.
This time, it finally generated an image.
The problem was that the layout no longer matched the text version it had previously described.
The bedrooms and bathroom had been moved, a foyer had been added (without a door connecting it to the living room), and the north arrow no longer made sense relative to the south-facing entrance. Several of the measurements also failed to add up correctly.
We pointed out that the generated image did not match the earlier layout.
Gemini apologized again and asked whether we wanted it to try another version.
After another attempt, the floor plan became more accurate overall, and the layout more closely resembled the original text description.
But the measurements still contained contradictions. For example, the north wall of the kitchen was labeled as six meters, while a visibly smaller wall section elsewhere in the apartment was also labeled as six meters.
Out of curiosity, we later asked Gemini why it originally claimed it could not generate images.
Gemini responded, “Sometimes I get stuck in a ‘text-only’ mindset and hallucinate limitations that aren't actually there.”
The response was oddly self-aware, but it also highlighted one of the patterns we noticed throughout our tests: Gemini often sounded more confident and certain than the actual results justified.
Editing the layout
Once Gemini finally generated a reasonably coherent floor plan, we moved on to the next part of the test: editing the layout.
This is where floor plan tools either become useful or frustrating. Real projects rarely stop after the first draft. Rooms get adjusted, furniture changes, and layouts evolve over time.
We asked Gemini to make three simple edits:
- Turn the kitchen into a one-wall kitchen
- Remove the dining table
- Replace the bathroom shower with a bathtub
At first glance, Gemini handled the request reasonably well.
The dining table was removed, and the kitchen was partially simplified. The shower was also replaced with a bathtub.
But the edits were not fully accurate.
Instead of converting the kitchen into a true one-wall layout, Gemini changed it into an L-shaped kitchen. More strangely, it also added a second bathtub inside Bedroom 2.
Individually, these mistakes might seem minor. But they stack up over time. Small unintended changes quickly become a problem once layouts go through multiple rounds of revisions.
Test 2: Turning Sketch Into a Floor Plan
Since many people already have a rough idea of the layout they want, we decided to try something more realistic: turning a hand-drawn sketch into a clean digital floor plan.
We created a simple hand-drawn apartment layout and asked Gemini to recreate it as a clean 2D floor plan.
This was actually one of Gemini’s strongest results during the testing.
The overall structure stayed close to the drawing, and most of the rooms and furniture ended up where they were supposed to be. Compared to the earlier text-based tests, Gemini seemed much more comfortable working from a visual reference instead of interpreting spatial instructions entirely from text.
There were still a few differences.
Gemini added an extra closet to the right of the entrance that did not exist in the intended layout. But in fairness, this was likely influenced by the sketch itself, which was not exactly architectural-grade artwork.
Making a small layout change
After seeing surprisingly good results with the sketch conversion, we wanted to run another layout change test to see whether Gemini could handle more precise edits when the request was limited to a single room instead of the entire apartment.
We focused on a single bedroom and asked Gemini to make three simple edits:
- Rotate the bed so the headboard sits against the northern wall
- Center the bed directly underneath the north-facing window
- Move the desk to the eastern wall
Gemini partially followed the instructions.
The desk was moved correctly, and the bed was rotated so the headboard faced the correct wall. But instead of centering the bed underneath the window, Gemini placed it noticeably off to the right side of the room.
We clarified the request and asked again, specifically emphasizing that the bed should be centered beneath the window.
Nothing changed.
On the third attempt, we tried something different. We asked Gemini itself to generate the ideal prompt for making this exact edit successfully.
Gemini produced a far more detailed instruction set describing the correct placement and alignment of the bed.
Then it failed to follow its own prompt.
The bed remained misaligned despite the increasingly specific instructions.
This became one of the more interesting patterns during testing. Gemini often understood the intent of the request and could even explain the correct solution in natural language, but still struggled to apply the changes accurately inside the actual floor plan.
Why AI Is Not Good for Creating Floor Plans
Our testing showed that Gemini can produce something that looks like a floor plan. It can interpret a sketch, suggest a layout, and make broad visual changes.
The overarching problem is control.
A usable floor plan needs more than rooms that look roughly correct. Walls, doors, windows, furniture, and measurements need to stay exactly where you place them. When you make one change, the rest of the layout should remain unchanged.
With Gemini, even simple edits introduced new mistakes. Furniture moved unexpectedly, measurements contradicted one another, and objects appeared where they did not belong. Repeating the request or adding more detail did not always fix the problem.
Floor Plan Software Lets You Stay in Control
Dedicated floor plan software works differently. Instead of describing each revision and hoping the tool interprets it correctly, you move the elements yourself.
You can draw and resize walls, position doors and windows, add exact measurements, and drag furniture into place. When a bed needs to move a few inches to the left, you can simply move it. You do not have to rewrite a prompt or risk changing the rest of the room.
RoomSketcher also uses AI to help you get started faster. With AI Convert, you can upload an existing floor plan and let AI map out the walls, doors, and windows. With FloorCapture, you can use an iPhone or iPad with LiDAR to capture a space.
The important difference is what happens next. Instead of receiving a fixed image, you get an editable RoomSketcher project. You can correct measurements, move walls, adjust doors and windows, add furniture, and continue developing the layout yourself.
This hands-on approach is especially useful because floor plan creation is rarely a one-step process. You might test several furniture arrangements, adjust room sizes, compare different kitchen layouts, or make changes after taking new measurements.
With the RoomSketcher app, each part of the floor plan remains editable. You stay in control of the layout while the software helps you create a clean, accurate 2D Floor Plan and visualize the finished space in 3D.
Want to try it yourself? Create a free RoomSketcher account and see how easy it is to draw, furnish, and visualize your floor plan. When you’re ready for more projects or advanced features, you can upgrade at any time.
Related Articles
Can Claude Create Floor Plans? We Tested It
Claude can generate floor plan–like layouts from prompts, but the results quickly break down. We tested apartments, room layouts, sketch conversion, and 2D-to-3D workflows to see how well it handles real floor plan work.
Can ChatGPT Make Floor Plans? We Put It to the Test
ChatGPT can generate simple floor plan sketches, but it struggles once you start making changes. We tested how well it handles real layout work, from editing and iteration to reuse across 2D and 3D.
How to Create a Floor Plan (5 Easy Methods)
The easiest way to create a floor plan depends on your starting point. Sometimes you’re working with an empty space. Other times, you already have a sketch, PDF, or old floor plan you want to reuse. Learn about the five ways you can create a floor plan with RoomSketcher.