wouldit.work

AI tools

Interior AI: built in 5 days, later reported at ~$45K MRR with >99% margins

Pieter Levels built Interior AI in five days in 2022. In 2024 he reported about $45,000 MRR and around $250 in monthly costs. Here is what that teaches, and what the skeptics said.

Founder
Pieter Levels
Launched
2022-10
Model
Subscription

Published · 1 min read · 3 sources

Key takeaways

  • Levels wrote up how he built interiorai.com from scratch in 5 days (October 2022).
  • In March 2024 he posted it made about $45,000 MRR with ~$250/month in costs: a $200 GPU bill for 21,000 designs and ~$50 of a shared server.
  • Hacker News commenters questioned self-reported numbers and the no-refund policy.

What it is

Interior AI generates redesigned versions of a room from a photo. Levels documented the build in "How I built Interiorai.com from scratch in 5 days" on 4 October 2022 (levels.io).

Timeline

  • Sep 2022: Levels launched thishousedoesnotexist.org (AI-generated houses) and wrote about running Stable Diffusion locally (levels.io post list).
  • 4 Oct 2022: published the 5-day build story for Interior AI (levels.io).
  • Mar 2024: a post quoted on Hacker News said Interior AI "now has >99% profit margins": $200/month GPU for 21,000 designs (about 1¢ per render), ~$50 share of a $500/month VPS, and about $45,000 MRR (HN thread quoting the post).

Why it worked

  • A small experiment came first. thishousedoesnotexist.org came before Interior AI and tested AI architecture imagery (levels.io post list).
  • Cheap per-unit cost. No per-user model training (unlike Photo AI), so each render cost about a cent (HN thread).
  • Automation over staff. The post claims it is "100% ran by AI robots, no people"; commenters noted these "robots" are automated scripts, not autonomous agents (HN thread).

What to learn

  1. Before you build, estimate cost per output. Products where each use costs pennies but sells for dollars can run with almost no staff.
  2. Small public experiments (a "this X does not exist" page) are cheap demand tests.

Risks and what didn't work

  • Skeptics on Hacker News questioned self-reported revenue and pointed out the site's no-refund policy, which sat oddly next to >99% margins (HN thread).
  • Margins depend on GPU pricing and third-party models that can change.

FAQ

Short answers drawn from the sections above.

What is Interior AI?

Interior AI generates redesigned versions of a room from a photo. Levels documented the build in "How I built Interiorai.com from scratch in 5 days" on 4 October 2022 (levels.io).

Why did Interior AI work?

  • A small experiment came first. thishousedoesnotexist.org came before Interior AI and tested AI architecture imagery (levels.io post list).
  • Cheap per-unit cost. No per-user model training (unlike Photo AI), so each render cost about a cent (HN thread).
  • Automation over staff. The post claims it is "100% ran by AI robots, no people"; commenters noted these "robots" are automated scripts, not autonomous agents (HN thread).

What can indie builders learn from Interior AI?

  1. Before you build, estimate cost per output. Products where each use costs pennies but sells for dollars can run with almost no staff.
  2. Small public experiments (a "this X does not exist" page) are cheap demand tests.

What didn't work for Interior AI, and what are the risks?

  • Skeptics on Hacker News questioned self-reported revenue and pointed out the site's no-refund policy, which sat oddly next to >99% margins (HN thread).
  • Margins depend on GPU pricing and third-party models that can change.

Sources

Every link cited above, in order of first use.

  1. 1.levels.io (levels.io)
  2. 2.levels.io post list (levels.io)
  3. 3.HN thread quoting the post (news.ycombinator.com)

Related case studies

More in AI tools →