What is i-cant-draw?

i-cant-draw is a plugin for Claude Code that helps people who struggle with visual creativity write better prompts for AI image generators. Instead of generating the image itself, the skill produces a polished, ready-to-paste prompt with the correct settings for either ChatGPT's GPT Image or Google Gemini's Nano Banana. You describe what you want in plain words, and the tool formats that idea into a structured prompt that the chosen image engine understands well. The workflow is straightforward. You call the skill, pick your model, and describe the scene you have in mind. The tool loads one engine-specific guide, follows its rules, and returns a finished prompt along with that model's configuration details. The two engines never interact with each other, so there is no confusion between their different requirements. Each guide is self-contained and tailored to its platform. The skill is built around a set of universal prompting principles drawn from official OpenAI and Google documentation. It writes in full prose rather than comma-separated keywords, orders descriptions from background through to constraints, uses concrete terms for lighting and color, and handles in-image text by quoting exact strings and naming font styles. It also notes that neither engine supports negative prompts, so the skill keeps exclusions brief. Installation is designed for Claude Code. You clone the repository, add it as a plugin, and install it. After that, you type a command, select a model, and describe your image. The skill files are plain markdown, so you can also copy them into a Custom GPT's instructions, a Gemini Gem, or any assistant's system prompt if you prefer a different setup. The README offers guidance on which engine to choose for different tasks. Gemini is recommended for character consistency across a series and conversational editing, while ChatGPT is suggested for transparent backgrounds and generating multiple variants in a single call. Both engines add a watermark to every output by default. The project is licensed under MIT.