Generate and edit images with Gemini's multimodal AI

Image Generator uses Google's Gemini Nano Banana Pro model (gemini-3-pro-image-preview) to generate and edit images from natural-language prompts. It handles text-to-image generation (photorealistic, stylized, or artistic outputs, including accurate text rendering for logos, infographics, and diagrams), image editing (adding or removing elements, semantic inpainting, style transfer), and multi-image composition (combining elements from up to 14 reference images, such as putting a dress from one photo onto a person in another). Advanced options include output resolution up to 4K, ten supported aspect ratios (1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9), Google Search grounding for generating images based on real-time data, and multi-turn editing that iteratively refines an image across a conversation. Every generated image carries a SynthID watermark, and the model runs a "thinking" process on complex prompts; text-heavy images render more reliably if the exact text is generated first and then requested inside the image.

Use it whenever a user asks to generate images, create visuals, edit photos, create logos, generate product mockups, or otherwise perform an image generation or editing task. It requires a GEMINI_API_KEY environment variable, set from a free Google AI Studio key, before any call is made - the skill checks for it as a pre-flight step and stops with an explicit error if it's missing, rather than attempting the call. Do not expect it to store generated images anywhere - the API does not persist outputs, so results must be saved locally as soon as they're produced. A request may also come back with no image if a prompt trips the API's own safety filters, in which case the fix is to rephrase rather than retry identically.

Input is a text prompt, optionally alongside one or more reference images (up to 14) for editing or composition; images are read from disk and base64-encoded before being sent. The skill deliberately writes the request body to a temporary JSON file rather than passing it as a command-line argument, since a base64-encoded image is too large for a shell argument list and triggers an "argument list too long" error. Output is returned as inline base64 image data (plus optional text) in the API response and is decoded and written to a local file - the response is never a URL or an asset stored on Google's side. Common failure modes documented for this flow: no image returned (check that response_modalities includes IMAGE), a blocked prompt (safety filters), and rate limiting (handle with exponential backoff).

Calls the Gemini API directly at https://generativelanguage.googleapis.com/v1beta/models/gemini-3-pro-image-preview:generateContent, authenticated via the x-goog-api-key header populated from GEMINI_API_KEY. Supported client paths are the Python SDK, the JavaScript/TypeScript SDK, and raw REST via curl. Installing the Python SDK:

pip install google-genai pillow

The JavaScript SDK is @google/genai. It ships as an MIT-licensed skill from the dair-ai/dair-academy-plugins repository.

Anyone who needs images generated or edited from natural language without leaving their coding agent - logo creation, product photography mockups, style transfer, infographics, or iterative multi-turn image editing - and is comfortable managing a Gemini API key. It documents concrete prompting guidance for each of these use cases: being descriptive rather than keyword-based, specifying photographic or artistic style and mood, being explicit about text to render and its placement, stating what to preserve during an edit, and naming lighting, background, and camera-angle details for product shots.