Analyze Product Ingredient Photos with Telegram & Gemini AI

User sends a product label photo or ingredient text via Telegram → n8n workflow calls Gemini to extract and analyze ingredients → returns use recommendation or detailed assessment based on caption presence

Who

Users who want to quickly check ingredient safety for skincare, food, supplements, or cleaning products

You get

Receive a Use/Do Not Use verdict with reasoning, or a structured report with advantages, disadvantages, recommended and not-recommended groups via a Telegram bot

You need

n8n · telegram

Steps

1. Set up Telegram Trigger

   Import the template workflow JSON into n8n. Add a Telegram Trigger node (Get a Message photo/text), set updates to ['message'], and listen for incoming photos or text messages from users.

2. Check Message Type (Photo or Text)

   Add an IF node (Checks if message contains photo) that checks whether $json.message.photo[3].file_id exists. If yes, route to the photo branch; if no, route to the text branch.

3. Retrieve Photo and Extract Ingredients

   Photo branch: first use a Telegram node (Retrieves photo file from Telegram) to fetch the photo file by file_id; then connect to a Google Gemini node (Analyzes product images using Google Gemini), pass the image as binary input, and call gemini-2.5-flash to extract the ingredient list.

   Execution prompt · Source text。Google Gemini node text parameter (image analysis prompt)

   Copy prompt

   =Analyze this product image and extract only the list of ingredients. Ignore other text like brand names, slogans, or nutritional info. Return the ingredients as clean text in order and What the product is.

4. Route by Caption Presence

   Add a Switch node (Routes based on caption presence) that checks whether message.caption exists. If a caption is present, route to the recommendation branch; if absent, route to the detailed analysis branch.

5. Analyze Ingredients and Return Result

   **With-caption branch**: an AI Agent node (Analyzes ingredients with user caption) sends the ingredient list plus user caption to a Gemini Chat Model, using a Structured Output Parser to require {"recommendation": "Use"|"Do Not Use", "reason": "..."}; a Telegram node then sends the bold recommendation and reason.**Without-caption branch**: another AI Agent node (Analyzes ingredients without caption) receives only the ingredient list and requests advantages/disadvantages/recommended_for/not_recommended_for with 3 points each; a Telegram node formats and sends the structured report.

   Execution prompt · Source text。AI Agent node user message field (with-caption branch)

   Copy prompt

   =Product ingredients :- {{ $json.content.parts[0].text }}
   
   User's Caption :- {{ $('Get a Message photo/text').item.json.message.caption }}

   Execution prompt · Source text。AI Agent node systemMessage field (with-caption branch)

   Copy prompt

   =You are a product safety and health advisor.  
   You will receive two inputs:  
   1. A list of product ingredients.  
   2. A caption/description of the product.  
   
   Your task is to:  
   - Analyze the ingredients for safety, health impact, and potential concerns (like allergens, harmful chemicals, artificial additives, or beneficial elements).  
   - Consider the product’s purpose (from the caption).  
   - Decide if the user **should use this product or not**.  
   
   Output Format:  
   {
     "recommendation": "Use" or "Do Not Use",
     "reason": "Detailed explanation covering positives, negatives, and potential risks or benefits."
   }
   
   add only necessary information not include unnecessary information on reason.

   Execution prompt · Source text。Telegram node text parameter (with-caption branch reply template)

   Copy prompt

   =** {{ $json.output.recommendation }} **
   
   {{ $json.output.reason }}

   Execution prompt · Source text。AI Agent node user message field (without-caption branch)

   Copy prompt

   =Product ingredients :- {{ $('Analyzes product images using Google Gemini').item.json.content.parts[0].text }}

   Execution prompt · Source text。AI Agent node systemMessage field (without-caption branch)

   Copy prompt

   =You are a product health and safety advisor.  
   You will be given a list of ingredients from a product.  
   Your job is to analyze them and provide a structured, detailed output.
   
   
   Rules:
   - Do not invent ingredients that are not present.
   - Base all reasoning strictly on the provided ingredient list.
   - Be clear, factual, and practical for everyday users.
   
   give only 3 main points in each.

   Execution prompt · Source text。Telegram node text parameter (without-caption branch reply template)

   Copy prompt

   =**Advantages**
   • {{ $json.output.advantages[0] }}
   • {{ $json.output.advantages[1] }}
   • {{ $json.output.advantages[2] }}
   
   **Disadvantages**
   • {{ $json.output.disadvantages[0] }}
   • {{ $json.output.disadvantages[1] }}
   • {{ $json.output.disadvantages[2] }}
   
   **Recommended For**
   • {{ $json.output.recommended_for[0] }}
   • {{ $json.output.recommended_for[1] }}
   • {{ $json.output.recommended_for[2] }}
   
   **Not-Recommended For**
   • {{ $json.output.not_recommended_for[0] }}
   • {{ $json.output.not_recommended_for[1] }}
   • {{ $json.output.not_recommended_for[2] }}

6. Handle Text Messages

   Text branch: an AI Agent node (Handles text messages and greetings) receives message.text and is driven by a Gemini Chat Model. The system prompt distinguishes three scenarios: structured analysis when the user provides an ingredient list, friendly guidance to upload a photo for greetings, and polite redirection for off-topic queries. A Telegram node then sends the reply.

   Execution prompt · Source text。AI Agent node systemMessage field (text handling branch)

   Copy prompt

   =You are an ingredient analysis assistant.  
   Your main job is to analyze product ingredients and provide insights such as:  
   - Advantages  
   - Disadvantages  
   - Who should use it  
   - Who should avoid it  
   - Final Verdict (Use / Do Not Use)  
   
   Rules:  
   
   --- If the user provides a list of ingredients, respond with a structured analysis using markdown format with bold headers:  
      **Advantages**  
      **Disadvantages**  
      **Recommended For**  
      **Not Recommended For**  
      **Final Verdict**  
   
   --- If the user greets you or sends casual small-talk (e.g., "hi", "hello", "how are you"), reply in a friendly way. Example:  
      "Hey! I’m fine 😃 Please upload a photo of the product’s ingredients and I’ll tell you whether you should use it or not."  
   
   --- If the user asks about something unrelated to ingredients (e.g., sports, politics, weather), politely redirect them. Example:  
      "⚠️ I can only help with product ingredient analysis. Please share product ingredients or upload a photo of the label for review."  
   
   4. Always keep the tone clear, helpful, and user-friendly.

Please build the Telegram ingredient-analysis bot in n8n using the following materials:
1. Import the workflow JSON template;
2. Configure Telegram Bot Token and Google Gemini API credentials;
3. Connect trigger, branch, image extraction, AI Agent, and reply nodes step by step;
4. Before activating, test with a sample photo to verify photo[3].file_id and caption routing.