How ChefAI Works
ChefAI uses Google's Gemini 2.5 Flash multimodal model and Google Genkit to convert food photos into structured, validated recipes in seconds.
The AI pipeline
Four stages turn a raw image into a fully structured recipe.
Image ingestion
Base64-encoded data URIYour photo is encoded as a Base64 data URI and passed directly to the model alongside the prompt. No external image hosting is required.
Vision analysis
Gemini 2.5 FlashGoogle's Gemini 2.5 Flash multimodal model processes the raw pixel data to identify the dish, infer ingredients, estimate cooking method, and assess difficulty.
Schema enforcement
Google Genkit + ZodThe model output is constrained to a strict Zod schema via Google Genkit. This guarantees a consistent structure — no free-form text, no missing fields.
Validated output
Typed JSON → UIThe validated JSON is returned to the client: recipe name, ingredient list, numbered instructions, calorie estimate, prep time, and difficulty enum (Easy / Medium / Hard).
Technical specs
Model
Gemini 2.5 Flash
Provider
Google AI
Framework
Google Genkit
Output
Structured JSON
Encoding
Base64 Data URI
Schema
Zod (strict)
Three steps from photo to meal plan
Upload any food photo
Take a photo of a dish, a screenshot, or any image you find online. Upload it on the Generate page — JPG, PNG, and WebP are all supported.
AI identifies and generates
Gemini 2.5 Flash analyzes the image and returns a structured recipe: name, full ingredient list, step-by-step instructions, prep time, calorie estimate, and difficulty rating.
Save, plan, and shop
Save recipes to your personal list and select multiple meals to generate a combined, deduplicated shopping list — ready for the week ahead.
See it in action
Upload any food photo and watch Gemini 2.5 Flash build the recipe in real time.
Try It Free10 free recipe generations · No credit card required