Image: GitHub
Why it mattersA designer can ask an agent to paint a picture and then edit each stroke. The agent does not make a flat image, so the person keeps control.
A designer who asks an agent for a picture has two paths today. The agent can call an image model. It gives back a flat pixel file. Or the agent writes code that draws shapes. Klad is a third path. The agent paints with brushes on the same canvas the person is working on. Every stroke stays editable.
Klad is a drawing and animation app from Indicode BV, a Dutch company. The repository has 141 stars. The author posted it as Show HN and the thread had 79 points. It is source-available under the PolyForm Internal Use License 1.0.0. Personal use is free. A company with under 500,000 euro in revenue can use it for free. The file format and the documentation are MIT.
What the agent and the person share
The person paints in the editor with a pen or a mouse. The agent paints by running the klad command line. Both use the same brushes, on the same drawing. Each side sees the other's strokes within about one second. A drawing is a folder of plain text files holding each stroke as a row. The agent reads and writes those rows the same way the editor does.
The agent's tools are small and named. klad stroke paints a brush stroke along a path. klad set changes a stroke's size, colour or position. klad strokes --json reads every stroke as structured data. klad fill applies a flat colour against a reference layer. The command list also covers notes, selection, in-between frames for animation, undo, and export to PNG, sprite sheets or Godot.
What this changes for someone building with coding agents
A coding agent that writes HTML or React code can only make pictures through an image model. An image model cannot be corrected one line at a time. Klad gives the agent brushes the person can edit later. The sample drawings in the repository include a creature animation with 4,344 strokes and 205 animation keys. Every stroke in that file is a row the agent could re-read and change.
The pitch on the repository is direct. "There is no image model in klad," the README says. "Your agent decides what to paint; klad gives it brushes, layers and a timeline."
The editor supports pen pressure and tilt on macOS. Linux starts with mouse input. Windows builds but has not been tested with drawing. About fifty open fonts and two dozen brushes come in the box.
No commercial price and no usage numbers
The README does not say how many people use Klad. The star count is the only traction number on the page. The pre-built app at klad.studio is a sign-up form today. The licence allows personal and small-company use for free. A team over the revenue cap has to buy a licence. The README does not price it.
Source
Primary source: klad-app/klad on GitHub. Discussion: Show HN thread.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
