A Developer, a 1989 Floor Plan, and One Very Stubborn JPEG

Two months ago we signed for a 68 m² flat in a building old enough to have opinions. On handover day the previous owner handed us a plastic folder containing a paper floor plan, photocopied so many times that the walls looked like soft grey suggestions. Architect's stamp: 1989. Coffee ring: right where the living room used to be.

I am a developer. I've spent years telling people that everything is just data. So when my partner asked whether the kitchen wall could come down, I said the most dangerous sentence in my vocabulary:

"It's just lines. I'll trace it."

Reader, it was not just lines. That weekend taught me more about floor planner tools and image to DWG conversion than I ever wanted to know, and it started with me being very confidently wrong.

"It's just lines," said the man who had never met a JPEG

My first plan was a Python script: load the scan, threshold it, push it through potrace, get SVG, convert to DXF, done before lunch. On paper it's a five-line pipeline. In reality it produced 8,000 path segments for a drawing with maybe 60 walls.

Nobody tells you this: a raster image has no idea what a wall is. Every scanner shadow, every pencil smudge, every photocopier speck becomes a path. Hatching turned into confetti; the dimension text got vectorized into tiny spaghetti.

I spent a Saturday on it and got a file that was technically vector, technically openable, and useless. It looks great at 20% zoom and falls apart the second you work inside it.

Redrawing it from scratch was the second bad idea

So I switched strategies: forget the scan, draw everything manually. I installed a free CAD viewer, learned just enough to be dangerous, and started placing walls.

Three hours later I had a bathroom. One bathroom. Beautiful, and useless, because I still didn't know whether the layout was any good. I'd produced a perfect replica of a mediocre 1989 plan: maximum effort, zero decisions.

That was the moment I understood. I didn't need drafting, I needed iteration — five arrangements thrown at the wall to see which one didn't make me angry. What I needed was floor planner work: a hundred cheap decisions before one expensive one. So I stopped trying to be a CAD operator and went looking for a floor planner instead.

Where a floor planner actually earns its keep

I found Floor Planner while looking for a floor planner that would let me describe the flat in words instead of coordinates. I'd tried two other floor planners first and dropped both inside ten minutes, because they wanted me to build a grid before they'd draw anything. This one works the way you'd hope: pick a mode, describe what you want, it draws.

The parts that mattered:

• Room mode vs. house mode. One space with your own width and length, or a full multi-room layout where you set bedroom count, bathrooms, and whether the kitchen is open or enclosed.
• Three output flavours. 2D for measuring and arguing, 2.5D for a bit of depth, 3D isometric for convincing a human being rather than a spreadsheet.
• Styles you can swap without redrawing. Technical and blueprint read like construction documents; realistic and 3D isometric look like the version you show your partner at 11pm; hand-drawn is for the mood board you'll never build.
• Image plus prompt. Drop in a reference photo or rough sketch, describe what should change, and generate from that instead of an empty grid.

I typed "68 square meters, two bedrooms, one open kitchen, one bathroom, keep the hallway short" and the floor planner came back in roughly the time it takes to reheat coffee. Then I changed one thing and got another. And another. Most floor planners would have made me rebuild the grid each time; this one just re-drew.

Six variations in under twenty minutes, and it never once asked me to start over.

That's the real shift. Any online floor planner is only as useful as it makes your bad ideas cheap. It won't replace an architect or produce a permit-ready drawing. But I generated one with the kitchen facing the wrong way, looked at it for four seconds, and deleted it. In CAD that would have cost me an evening.

We picked a layout that night. The wall came down two weeks later.

Then the contractor asked for the drawing that already existed

Here's where the two halves of my weekend meet: a floor planner for what could be, and image to DWG for what already was.

Our contractor was lovely and extremely practical. He wanted the original plan as a CAD file so he could mark demolition lines, note the load-bearing bit, and hand it to the electrician. Except the only version of that plan was a 37-year-old photocopy, and the DWG didn't exist anywhere on earth.

I was not going to trace it by hand. I'd already paid for that lesson, and image to DWG was one browser tab away.

So I ran the scan through Image to DWG. It converts a JPG, JPEG, or PNG of a drawing into an actual DWG file. Not a picture dropped onto a CAD canvas — real entities: polylines, arcs, circles, closed hatch boundaries. You tell the image to DWG converter whether it's a general floor plan or an architectural plan, toggle text recognition, check the reconstruction in the browser, then download.

Three things about the image to DWG output genuinely surprised me.

Layer separation. It sorts elements by pixel density and colour into separate CAD layers with their own pen weights. Walls, hatching and annotations arrive as different things — the difference between "a file" and "a file you can work in." Every other image to DWG tool I'd tried gave me one flat layer of lines.

Straightening. Paper photocopied four times is not orthogonal — nothing in that scan was really at 90 degrees. The image to DWG conversion snapped endpoints and pulled skewed lines back to right angles, so the walls became walls instead of very committed diagonals.

Opening cleanly. The DWG opened in AutoCAD with no missing-entity warnings and no exploded blocks. For something running in a browser tab, that's the part I'd expected to fail.

The whole image to DWG pass took under a minute. I overlaid my own measurements, corrected two walls the 1989 stamp had gotten wrong, and sent it off. That drawing spent 37 years as a photocopy, then became editable geometry while I made tea.

The workflow I'd actually recommend

If you're about to do this yourself, here's the loop that worked: image to DWG at both ends, a floor planner in the middle.

1. Start with what exists. Run the old plan through image to DWG first, not last. You want a real editable base before you start dreaming.
2. Calibrate the scale. A scan has no reliable scale. Measure one real wall with a tape, compare it against the image to DWG output, and fix the ratio once. Everything downstream depends on that number.
3. Iterate in a floor planner, cheaply and often. This is where a floor planner beats CAD by a mile: dozens of rough variations, no layer management, no snapping regret. Judge them in 3D, not in your head — nobody can feel a room from a top-down plan.
4. Take the winner back to CAD. Once you've committed, go back to the image to DWG output and make it true. Dimensions from the real flat, not from the concept.

The mental model that finally clicked: image to DWG is for the building that exists. A floor planner is for the building that could exist. One is a converter, one is a sketchbook. Confusing them is how people waste weekends.

Mistakes I made so you can skip them

• Don't trust a concept drawing's measurements. AI floor planner output is good at proportion and bad at your actual walls. Treat it as an opinion, not a fact.
• Don't leave text recognition on if you only need geometry. Text just adds cleanup, and the image to DWG viewer shows you exactly where the OCR went feral.
• Don't expect image to DWG to fix a bad scan. A blurry 200px JPEG in, a blurry vector out.
• Keep the original raster forever. In six months you won't remember which final_v3 was the load-bearing wall.
• Show the 3D view to whoever has to live there. My partner vetoed two layouts I liked in about four seconds. That's not the floor planner failing. That's it working.

Where we're at now

The flat is half-renovated and smells like plaster. The wall I worried about is gone, the kitchen sits where the bedroom used to be, and the only drawing on site is a DWG that began life as a photocopy. I've since run two more of the contractor's scans through image to DWG, which tells you how the rest of this renovation is going.

I still think everything is data. I just have far more respect for how much work it takes to get data out of a piece of paper — and for how much of that work used to be a person moving a mouse along a line.

So: if you're staring down a renovation with a scan instead of a file, don't trace it by hand. Run it through an image to DWG converter, get something editable, then let a floor planner do the one thing it's genuinely great at — letting you change your mind before it costs money.