Prism Image Generator is a complete text-to-image app front end: a prompt box with example chips and a Surprise me button, eight style presets, five aspect ratios, a seed with randomize and lock, and batches of four results with a shimmer-and-progress animation. Every image has actions for variations, favorite, PNG download and a large lightbox view with the prompt, seed and a copy / reuse-settings option. History lives in a local gallery with a favorites filter, delete and clear.

No AI backend is required. A deterministic procedural art engine on Canvas2D draws every picture from a hash of prompt + style + seed - same inputs, same image - and reads the prompt for meaning: colour words choose the palette, mood words change energy, darkness and density. The eight looks (Aurora, Ink Wash, Vaporwave, Blueprint, Mosaic, Nebula, Topo, Risograph) use flow fields, layered noise, Voronoi tiles, marching-squares contours, halftone screens and film grain, so results look finished rather than placeholder.

History stores only small thumbnails and parameters; large views and downloads are re-rendered on demand.

The engine sits behind one small provider interface in js/ai.js, documented step by step. Switch to an OpenAI-compatible images endpoint from Settings, or point it at your own proxy so no secret key ever reaches the browser. Plain HTML, CSS and JavaScript, no dependencies, works by double-clicking index.html.