taruma/promptlab

PromptLab is a sketchpad for writing and testing AI prompts before you take them anywhere serious. Instead of a finished production tool, it is a place to draft your instructions to an AI, try them out against Google's Gemini models, and refine them until the wording feels right. You write a prompt template with placeholders like a name or a scene description, and PromptLab automatically turns each placeholder into a form field you can fill in without touching code. You can drag and drop reference images into the app to use alongside your prompt, and each one gets labeled and mapped so the AI knows which image goes with which part of your instructions. Images are compressed and stored in your browser rather than sent to any server, avoiding the storage limits that browsers normally place on saved data. You can save your own prompt setups as presets, export them as files, or share one with someone else through a link. A built-in comparison view shows what changed between your current setup and a saved preset, similar to how a code diff highlights additions and removals. The app includes a control panel where you pick which Gemini model to use, adjust settings like temperature and how much reasoning effort the model applies, and manage your own API keys. Every generation you run gets saved to a local history you can browse, mark as a favorite, or export later. There is also a sidebar library for uploading and reusing images across different prompt sessions, and three example presets covering things like screenplay writing and cinematic storyboarding to show what the tool can do. To run PromptLab yourself you need Node.js 18 or later and a Google Gemini API key, which you can either place in a local environment file or type directly into the app. The project's author states plainly that it was built through AI-assisted coding and should be used at your own risk. It is released under the MIT license.