Tried it on my own floor plan, 8 min: how house-plan-designer handles the step floorplan-3d doesn't

house-plan-designer is a Skill for agents like Claude Code and Codex: feed it a floor-plan image, say one sentence, and it returns a renovation-design web page you can open with a double-click. It hit 31 stars in 359 days, a tenth of its predecessor floorplan-3d, but its fork rate is high, growth is steady, and every rise has a traceable cause. What's worth learning here is the playbook; the floor-plan topic itself is already crowded.

Oct 23, a three-bedroom two-bath

On Oct 23, Sspai published "Has home renovation gotten a 'less heartburn' version? I tested this open-source AI project on my own floor plan." The author, "熊猫不是猫" (Panda Is Not a Cat), normally writes a NAS column. This time he gave Codex the floor plan of his three-bedroom, two-bath home and said only two things: first "install this skill," then "turn this floor plan into a renovation-design web page."

Codex set up the dependencies and checked the browser on its own. 8 After a few minutes, he had an HTML file of about 1.4MB and opened it in Chrome: 9 rooms, 60 pieces of furniture, a net interior area of 120.47㎡, with a switch to 3D bird's-eye view and first-person walkthrough, the plumbing and electrical points on their own layer, and a quote section that produces a bill of quantities from the plan. His verdict: "kind of like Beike's VR home tours."

The same article ran on Smzdm the day before, and the same week it appeared on Tencent Cloud Developer Community and Zhihu columns. The repo's star counts those days: 10-22 was 80, 10-24 was 147.

What it actually does

The repo was created Oct 6. The author ZWD11 has only this one public repo and 17 followers. The README's first line: "One floor-plan image (or one CAD drawing), one sentence, and the agent restores it into a renovation-design web page you can directly design in 2D / 3D."

The work is split up: the model reads the drawing, scripts measure walls, add doors and windows, and place furniture, then the result is overlaid on the original image to check it, and finally packaged into an offline HTML file. It does more than good-looking 3D:

• Outlets, lighting, network cables, hot and cold water, and drainage 5 types of piping are laid out automatically, claimed to follow national standards, with switches placed on the handle side of the door and not behind it, per construction convention;
• It can export national-standard DXF construction drawings and calculate quantities and quotes, with editable unit prices;
• Before delivery, it automatically runs three cross-checks: walls, dimension chains, and furniture placement;
• Later it added office floors, recognition of black-background CAD screenshots, suspended-ceiling light strips, and ray-traced rendering.

One caveat: the generated designer is compiled, and the source (TypeScript + Vite + three.js) isn't in the repo. Someone has already asked about it in the issues.

Why it took off: someone had already educated the demand

In Sep 29, wy51ai/floorplan-3d launched: a single-file 2D/3D floor-plan designer that looks good and supports walkthroughs. On Oct 1, GitHubDaily pushed it on X, getting 38.1k views and 592 bookmarks, and then a wave of Chinese sites reposted it, bringing it to 3,582 stars by 11-06.

But its floor plan is a sample hardcoded in the code. The README itself says "change a few sets of room and wall data to swap in your own," which for most casual viewers amounts to "you can't swap it." So the first reaction after watching was almost always: how do I put in my own home?

5 days later house-plan-designer created its repo, doing precisely that step, and handing it to an agent. The author never said it was inspired by the earlier project, so this is an inference, but the timeline and positioning line up. Layer on this year's Claude Code / Codex Skill craze in Chinese circles, and "install one Skill, drop in one image" is inherently easy to share.

How the stars grew

Numbers come from GitHub's daily star records (since 10-17). 11-06 has 359 stars and 67 forks; the fork rate is close to 19%, while similar showcase repos usually sit at 5–10%. That suggests many people are actually running it on their own floor plans. Requests in the issues like "my home needs central AC" and "I want to swap the tile textures" back this up.

It hasn't made GitHub trending (the repo was under 60 days old), and only reached #24 on Trendshift's HTML language chart at 10-23. I found no big X account reposting it; GitHubDaily and Ruan Yifeng's weekly haven't covered it; I found no videos on Bilibili, Xiaohongshu or Douyin (not finding them doesn't mean they don't exist). Whether Panda's post was organic or sponsored is unverified.

The pattern is clear: one hands-on review plus fresh screenshots and a changelog with each major update, and each one brings a wave of stars.

Who it's competing with

"Floor plan → 3D" is a crowded topic on October, all riding the floorplan-3d wave. house-plan-designer is the only one that turns the full workflow—read the image, measure walls, self-check, lay out plumbing and wiring, output construction drawings—into a Skill. I haven't seen anyone copying its approach.

It's this deep, so why does it have only a tenth of the stars?

floorplan-3d gets people to star it with a single animated GIF. house-plan-designer's benefits—accurate wall measurements, code-compliant wiring, DXF files a construction crew can use—only become obvious after you run it yourself. Most people who land on the repo never do.

So its depth has earned high fork rates and good reviews from testers, but not star counts. If you want stars, impact works better than workflow depth. If you want proof that people are actually using it, depth is what matters. That's the most important takeaway from this project.

If it were you

Verdict: Copy its approach, but don't build another floor plan tool. "Floor plan → 3D" is already packed: floorplan-3d took the bulk, and house-plan-designer made it engineering-grade with 105 acceptance checks yet only got 359 stars. The Nth entrant is unlikely to beat it.

Who it suits: Indie devs who can write Agent Skills, who know the industry in real detail (like the national-standard wiring and switch placement here), and who can ship a solid update every week.

Who it doesn't suit: Anyone hoping for thousands of stars overnight from a single demo, or anyone unwilling to write a lot of validation scripts and who just wants a prompt. That layer already exists for free.

How to do it: Watch for the next "pretty but only shows sample output" demo that a big account like GitHubDaily boosts, and build the step it's missing: "swap in your own." Hand it to an Agent: the model reads the input, scripts handle measuring, validation and packaging, and the output is a file that opens on double-click. You could change the domain (circuit diagrams, garden layouts, booth setups, PCB, family trees), or change the audience: build a version that reads English floor plans, uses feet, and outputs US-style formats. I haven't verified whether anyone has done the English-speaking version, so search first before you start.

How to get users: Find a reviewer with their own column and have them run it on their own real data. Publish one article on sspai, 什么值得买 and Zhihu at the same time. Pair each major update with a fresh set of screenshots and a changelog. Answer "is there a good xx skill?" threads on developer communities like linux.do.

When to move: Build the repo within 1–2 weeks after the original demo goes viral. house-plan-designer came only 5 days later, right when people were asking "how do I swap in my own house?"

Signs you've misread it: ① house-plan-designer organically gains over 2,000 stars within a month, which would mean depth can also earn stars and the floor plan niche still has room. ② Someone ships an English floor plan Skill that passes a thousand stars in a month, which would mean the audience-swap route is open and worth jumping on right away. ③ The author takes down the designer source and turns it into a paid product, and the forks stop updating, which would mean this kind of Skill's heat is just a passing wave.

Data and sources: GitHub repo page, README and issues, daily star records (10-17 to 11-06); trendshift.io/repositories/315706; GitHubDaily X post (10-01); wy51ai/floorplan-3d repo page; sspai.com/post/117175 (10-23), 什么值得买 (10-22), Tencent Cloud Developer Community, Zhihu column; linux.do/t/topic/3098556 (10-30); kujiale.com/skills; egmalt/house-planner repo page.