Generate Recipes from Fridge Photos via Telegram

Input: fridge photo or text ingredient list → Process: n8n workflow identifies ingredients and calls AI to generate recipes → Output: formatted emoji-rich recipe sent back on Telegram

Who

Operators or developers with an n8n instance who want to build an AI recipe assistant on Telegram

You get

Users receive 3 complete recipes with cooking time, difficulty, and step-by-step instructions after sending a photo or text of ingredients

You need

n8n · telegram

Steps

1. Configure trigger and input routing

   Add a Telegram Trigger node, enable message updates, and turn on download to receive images. Then add a Check Input Type (IF) node that checks whether $json.message.text exists: if not, route to the image recognition branch; if yes, route to the text input branch.

2. Image branch: identify ingredients

   On the image branch, add an AI Vision Agent node connected to an OpenAI Vision Model (model: openai/gpt-4-vision-preview, temperature 0.3). The agent receives the image and outputs a structured ingredient list, which is forced into JSON by an Ingredient Parser (Structured Output Parser), then converted to a comma-separated string by a Format Ingredients (Set) node.

   Execution prompt · Source text。AI Vision Agent node text parameter (user prompt)

   Copy prompt

   =Please analyze this food image and identify all visible ingredients. List them clearly.
   
   Image URL: {{ $json.body.events[0].message.id }}

   Execution prompt · Source text。AI Vision Agent node options.systemMessage parameter (system prompt)

   Copy prompt

   =You are a food ingredient identification expert. Analyze food images and identify all visible ingredients accurately. Return a structured list of ingredients.

3. Text branch: extract ingredients

   On the text branch, add a Get Text Ingredients (Set) node that assigns $json.message.text directly to the ingredients_text field, with no AI processing needed.

4. Generate recipes

   After both branches converge, feed into a Recipe Generator (AI Agent) node connected to an OpenAI Recipe Model (model: openai/gpt-4o-mini, temperature 0.7). The agent generates 3 recipes from the ingredient list, then a Recipe Parser (Structured Output Parser) forces the output into a JSON structure with name, cuisine, difficulty, cookingTime, servings, ingredients, and instructions.

   Execution prompt · Source text。Recipe Generator node text parameter (user prompt)

   Copy prompt

   =Based on these ingredients: {{ $json.ingredients_text }}
   
   Please suggest 3 delicious recipes that can be made with these ingredients. Consider both Japanese and international cuisine options. Text is Japanese.

   Execution prompt · Source text。Recipe Generator node options.systemMessage parameter (system prompt)

   Copy prompt

   You are a professional chef and recipe expert. Create detailed, easy-to-follow recipes based on available ingredients. Include both traditional Japanese dishes and international cuisine. Always provide cooking time, difficulty level, and step-by-step instructions.

5. Format and send response

   Add a Format Response (Set) node that uses a JavaScript expression to convert the Recipe Parser's JSON output into an emoji-rich, readable message. Finally, add a Send a text message (Telegram) node to deliver the formatted message back to the user's chat.

6. Configure credentials and activate

   In n8n credentials, add telegramApi (Bot Token) and openRouterApi (API Key), then bind them to the Telegram Trigger/Send a text message nodes and the OpenAI Vision Model/OpenAI Recipe Model nodes respectively. Save and activate the workflow. [First-test safety] Before the first test, disconnect or disable the Send a text message node, or temporarily set chatId to a test account; reconnect real sends only after Format Response looks correct.

You are an n8n workflow setup assistant. The user wants to replicate a Telegram AI recipe bot. First confirm they have: 1) a running n8n instance; 2) a Telegram Bot Token; 3) an OpenRouter API Key (or OpenAI API Key). Then guide them to import workflow.json or build the nodes manually, with special attention to AI Agent prompt variables, Structured Output Parser JSON Schema, and Set-node JavaScript expressions referencing upstream output fields.