EN
Back to the archive

The archive · Product Ideas · Product decision · 2021

Max Fischer builds a home AI art installation that makes new pieces on the fly

Max Fischer's open-source art kiosk runs a generative model on an edge GPU, swaps the framed piece on a button press and blanks the screen when nobody is there.

DIY-ai-art (maxvfischer)

The ideaMake the wall a gallery that never repeats: an edge GPU tops up the piece pool, a button swaps the framed artwork, and motion sensing blanks the screen when alone.incremental

What it had to solve

Max Fischer wanted an AI art installation at home that behaved like a living gallery rather than a screen: each piece shown once, a new one on demand, and a dark frame when nobody is there to look. Static generations on a display also risked burn-in, and a TV that showed a 'no signal' logo between images would break the illusion.

How it works

Most AI-generated art lives as a file or a scrolled feed: static, repeatable, and lit by a computer screen. Max Fischer's idea was to give it the behavior of a real gallery piece — an artwork that hangs on the wall, changes when the viewer asks, and goes dark when no one is in the room.

His installation runs on an Nvidia Jetson Xavier NX, an edge computer chosen because the generative model has to run locally, and drives a Samsung The Frame 32" (2020) hung on the wall. A button beside the frame swaps the displayed piece for a random one from an image pool; when the pool runs low, the system spawns the generative code to create a new batch, so the art is produced on the fly rather than curated in advance.

The details are what make it feel like art instead of a slideshow. A passive-infrared sensor watches the room: when no motion is registered for a while, a screensaver blanks the screen — protecting The Frame from burn-in while giving the installation a gallery's dark, waiting state. All electronics, power adapters and cables hide in a hand-built MDF control box that was painted to match the wall, and the button box routes cables invisibly between screen and control unit.

Fischer published the entire project as an open-source guide — OS setup, four-process Python kiosk architecture, woodworking, high-voltage wiring, and the hooks where a builder plugs in their own generative model. The Show HN drew 399 points and 153 comments on 2021-08-18, and the repository's from-scratch build guide had gathered about 566 stars by 2026.

Why it lands

  • Scarcity makes generated art precious: each piece is shown, then gone, so the viewer watches the moment instead of scrolling past a file.
  • Running the model on an edge GPU keeps the installation self-contained and private, a deliberate counter to the cloud-dependent AI art tools of the time.
  • The motion sensor treats absence as part of the design — a blank screen when no one watches is what a real gallery does.
  • Shipping it as a full open-source guide let anyone recreate the piece, which turned one person's wall into a replicable format.

What it did

The Show HN on 2021-08-18 drew 399 points and 153 comments, among the day's top posts. The repository, still presented as a from-scratch build guide, showed about 566 stars at the 2026 snapshot.

Their siteDIY-ai-art repo: build guide and code

What you can take

Treat generated images as perishable gallery pieces: display each until the viewer asks for the next, generate replacements in the background, and let an empty room switch the piece off.

Since then

The repository remained a maintained how-to with 163 commits and about 566 stars by the 2026 snapshot, described as 'how to build your own AI art installation from scratch'. The kiosk's architecture stayed modular on purpose: the event handler that tops up the image pool is the one place a builder must add their own generative code, so the guide supports whatever model comes next. Fischer kept the project open to contact and questions, and the hardware choices — an edge GPU, a frame-style TV, a physical button and a presence sensor — became the recipe others could follow or remix.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe the brief you are staring at, and see who has been given the same problem.

Free account · 3 free questions · no card

Related cases