A browser-native image lab that hoards your history in IndexedDB

What it does

GPT Image Playground is a React interface for generating and editing images through OpenAI’s gpt-image-2 and compatible APIs. It covers text-to-image, reference images with mask editing, and multi-turn agent conversations, while keeping every task, image, and parameter in the browser’s IndexedDB. It also supports fal.ai and custom HTTP providers, with optional CORS proxying for self-hosted setups.

The interesting bit

The app treats the browser like a native client: it deduplicates stored images with SHA-256, supports desktop drag-to-select and mobile swipe multi-select, and exposes the exact API-side parameters—including the model’s rewritten prompt—so you can audit how your request was mutated. Its Agent mode, built on the Responses API, can branch conversations and batch-generate related images in parallel.

Key highlights

• Pure local persistence: all records and images live in IndexedDB, with one-click ZIP export for backups.
• Visual mask editor and up to 16 reference images per generation, with automatic resolution preprocessing.
• Streaming previews for both Images API and Responses API to dodge connection timeouts.
• Multi-provider support: OpenAI-compatible endpoints, fal.ai queues, and custom HTTP services via JSON import.
• Smart parameter diffing: highlights gaps between your requested settings and the API’s actual response, including rewritten prompts.

Caveats

• Enabling the built-in API proxy turns your server into an open relay; the README explicitly warns you to add access controls or keep it on a local network.
• Vercel’s .dev domain enforces HTTPS, so local or intranet API endpoints require GitHub Pages or self-hosted Docker instead.

Verdict

Useful if you generate images often and want a dashboard-quality UI without cloud-syncing your creative history. Overkill if you just need a quick curl one-liner or don’t trust browser storage with your work.