ideoboard

Ideoboard is a browser based front end for prompting Ideogram's v4 text to image generation model. Rather than typing everything into one plain prompt box, it gives users a structured, spatial editor: a scene is built from a high level description, style fields, reusable text fragments called tags, and boxes drawn directly on a canvas that mark Text or Object regions with their own descriptions and stacking order. From that structured scene, the app generates one or more images.

Every prompt used to generate images is tracked as its own node inside a visual graph view, with the images it produced attached to that node. Small edits to a prompt create new nodes chained onto the sequence, and any node can be branched off to start a separate generation path, letting a user explore multiple directions from the same starting point.

The app is entirely client side, built with React, TypeScript, Vite, and Tailwind, and it stores all scenes and generated images locally in the browser's IndexedDB rather than on a server. The only outside network calls go to the chosen image generation provider. It can talk to Ideogram through either Fal.ai, called directly from the browser, or Ideogram's own API, which is relayed through a small same-origin proxy because Ideogram does not allow direct browser requests. Each provider uses its own API key, entered in the app's settings and stored only in the browser's local storage, never on a server. If no key is set, the app falls back to a locally rendered placeholder image so the graph and branching flow can still be tried offline.

Ideoboard is designed to run without a backend server, typically deployed to Vercel, and is released under the MIT license.