InstaGenie: Building Your Own AI Image Artist Telegram Bot with n8n

February 9, 2025
AI, Blog

Ever dreamed of having a personal AI artist at your beck and call? Imagine this: you message your Telegram bot, “Create a professional headshot of a person in a library setting,” and poof, a stunning, realistic image appears! Or, you spot a fantastic dress on a Pinterest board or an e-commerce site, share the image to your bot, and instantly see yourself (or a model) wearing a similar outfit. With InstaGenie, that dream becomes reality.

This guide will walk you through building “InstaGenie,” a Telegram bot powered by cutting-edge AI, that turns your text descriptions and image inspirations into custom-generated images. We’ll leverage the power of n8n, a visual workflow automation tool, making the process surprisingly accessible even if you’re not a coding wizard.

What Can InstaGenie Do?

InstaGenie is more than just a simple image generator. It’s designed to be your versatile AI art companion, capable of:

  • Text-to-Image Generation: Describe the image you want, and InstaGenie will create it. Examples:
    • “A majestic lion standing on a savanna at sunset.”
    • “A futuristic cityscape with flying cars and neon lights.”
    • “A cute, fluffy kitten playing with a ball of yarn.”
    • “Headshot of handsome young man, ,wearing (black leather and white tshirt), rectangular-framed glasses with dark rims, tousled black hair that’s slightly messy on top, smiling look on his face, grey background, soft studio lighting, portrait photography.”
  • Image-to-Image Transformation (Outfit Swapping): Find an outfit you love online? Share the image with InstaGenie, and it can generate an image of you (or a generic model) wearing a similar outfit. This is fantastic for:
    • Virtual Try-On: See how a particular style might look on you before buying.
    • Outfit Inspiration: Get ideas for new looks based on images you find.
    • Creative Content: Generate unique images for social media or personal projects.
  • Customization: Adjust image generation parameters like:
    • Style: Influence the artistic style (e.g., photorealistic, painterly, anime). This is primarily controlled by the LoRA you choose (more on that later).
    • Resolution: Control the image’s width and height.
    • Steps: Adjust the number of diffusion steps (more steps often lead to higher quality, but take longer).
    • Guidance Scale: Fine-tune how closely the generated image adheres to your prompt.
  • Voice Input: Just record and send a voice note. InstaGenie will transcribe it and create image.
  • Bot Configuration: Change the configuration of the bot by sending messages. Example, I want the image width to 512.

The Vision: A Telegram Bot That Creates Images

Our core goal is simple:

  1. User Interaction: Users message our Telegram bot with text descriptions or images.
  2. AI Magic: The bot uses an AI image generation model (Flux, enhanced with LoRA) to create the image.
  3. Delivery: The bot sends the generated image back to the user in Telegram.
  4. Configurability: Users can adjust parameters through bot commands.
  5. Voice and Image input: User can use voice or image as input.
  6. Error Handling: Bot will reply generic message when encounter issue.

The Tools: n8n, Telegram, Replicate, and Supabase

We’ll be using:

  • n8n (n8n.io): Our visual workflow automation platform.
  • Telegram API: For bot creation and interaction.
  • Replicate API: Provides access to the Flux image generation model and LoRA support.
  • Supabase: A backend-as-a-service (BaaS) for storing user configurations.

Prerequisites:

Before we start:

  1. A Telegram Account:
  2. n8n Instance: Self-hosted (recommended) or n8n Cloud.
  3. Replicate Account: Sign up and get your API key. Add credits.
  4. Supabase Account: Create a free account and a project.
  5. OpenAI Account: Create a free account.
  6. Basic understanding of JSON.
  7. (Optional) Custom LoRA: Or, find a pre-trained LoRA online.

The Workflow:

Step 1: Setting up Telegram Bot and Webhook

  1. Create a Telegram Bot:
    • In Telegram, search for “@BotFather”.
    • /newbot to create a bot. Follow the instructions.
    • Keep the bot token secret!
  2. Configure n8n Webhook:
    • In n8n, create a new workflow.
    • Add a “Webhook” node.
    • POST method, None authentication.
    • Give the “Path” a unique name (e.g., my-instagenie-bot).
    • Copy the “Test URL” and “Production URL”.
  3. Connect Telegram to Webhook URLS:
    • Add a Set node to your workflow
    • Add token and test_url, production_url in your Set node. Use your Bot Token and n8n webhook URL here.
    • Create 2 HTTP Request nodes. One to set the Test URL, one to set your Production URL.
      • URL: =https://api.telegram.org/{{ $json.token }}/setWebhook
      • Query Parameters:
        • Name: url
        • Value (test): ={{ $json.test_url }}
        • Value (production): ={{ $json.production_url }}
    • Create 1 HTTP Request nodes.
      • URL: =https://api.telegram.org/{{ $json.token }}/getWebhookInfo
      • This helps to debug if webhooks are properly configured.
  4. Test Telegram Connection
    • Create 2 Telegram nodes.
    • In the “Chat ID”, enter your own Telegram Chat ID. You can know it by sending a message to your created bot and see the Listen to Telegram Events node.
    • Connect to the output of the webhook configure nodes to test the connection.

Step 2: Setting up Supabase (Database)

  1. Create a Supabase Project:
    • Log in to Supabase and create a new project.
  2. Create a Table:
    • Go to the “Table Editor”.
    • Create a table named kanaklora (or your choice).
    • Columns: id (int8, primary key, auto-incrementing), userid (text), memkey (text), memvalue (jsonb).
  3. Get Supabase Credentials:
    • “Settings” > “API”.
    • You’ll need your “Project URL” and your “anon” (public) key. Keep the “service_role” key secret!

Step 3: Building the n8n Workflow

  1. Listen for Telegram Events and Validation
    • Add the Listen to Telegram Events node and connect with previous nodes.
    • Add a Set node named “Validation”
    • Add a If node named Check User & Chat ID
    • Add a Telegram node named “Error message”
    • Connect it as explained in Part 1 of previous section.
    • Add your own Telegram Chat ID in the “IF” node to validate incoming message.
  2. Supabase Configuration Retrieval
    • Add a Supabase Node named “Supabase Get Config”.
    • Add your “Supabase account” credential.
    • Select get operation.
    • Select table kanaklora
    • Add filters to get config for the incoming user.
  3. Bot Variables (Set Node):
    • Create a “Set” node named Bot Variables.
    • Add fields like enhance_img_prompt, cfg_scale, steps, width, height, and lora_scale.
    • Pull values from Supabase using expressions.
  4. Message Router (Switch Node):
    • Add a “Switch” node named Message Router.
    • Branches: “audio”, “text”, and “image”.
    • Configure as explained in the earlier section.
    • Set “Fallback Output” to “extra”. Create output named “extra”
    • Connect to the output of previous node.
  5. Audio Processing Branch:
    • Get Audio File (Telegram Node): Get the audio file.
    • Transcribe Recording (OpenAI Node): Use OpenAI “Transcribe Audio”.
  6. Text Processing Branch:
    • Edit Fields (Set Node): Extract the text property.
    • Text Classifier (Langchain Text Classifier Node): Use “Categorize Text”. Categories: botconfig, imagegen, other.
    • Connect the output to AI Assistant 👩🏻‍🏫 and Enhance prompt or not
  7. Image Processing Branch:
    • Image Schema (Set Node): Extract file_id and caption.
    • Get Image (Telegram Node): Download the image.
    • Move Binary Data (Move Binary Data Node): Convert data.
    • Analyze Image (OpenAI Node): Use OpenAI “Analyze Image”. Provide instructions.
  8. AI Assistant (for Bot Config and General Queries):
    • Window Buffer Memory (Langchain Buffer Window Memory Node): Conversational memory. Use Telegram ID as sessionKey.
    • OpenAI Chat Model (Langchain Chat OpenAI Node): Your main AI model.
    • Supabase2 (Supabase Tool Node): Update botconfig.
    • Supabase3 (Supabase Tool Node): Retrieve botconfig.
    • AI Assistant 👩🏻‍🏫 (Langchain Agent Node): The main agent. “Conversational Agent.” Define its persona with a system message. Connect to Supabase2 and Supabase3 tool nodes.
    • Telegram (Telegram Node): Send the AI’s response.
  9. Image Generation Pipeline:
    • Enhance prompt or not (IF Node): Check enhance_img_prompt.
    • AI Agent (Langchain Agent Node): If enhancing, refine the prompt.
    • Set Prompt and Style (Set Node): Prepare prompt and style.
    • Call Replicate API (HTTP Request Node):
      • Method: POST
      • URL: https://api.replicate.com/v1/predictions
      • Authentication: “Generic Credential Type” -> “HTTP Header Auth”. Replicate API key.
      • Headers: Content-Type: application/json, Prefer: wait
      • Body: See “Using Replicate, Flux, and LoRA” below.
    • HTTP Request (HTTP Request Node): Download the image.
    • Telegram (Telegram Node): Send the image back.

Using Replicate, Flux, and LoRA

  • What is Flux? A flexible, open-source image generation model on Replicate. A base model for fine-tuning.
  • What is a LoRA? LoRA (Low-Rank Adaptation) efficiently fine-tunes large models like Flux. Adds small, trainable “adapters” to customize style or teach specific subjects without modifying the core model.
  • Custom or Pre-trained LoRA:
  • The Replicate API Call (HTTP Request Node Body):
{
"version": "091495765fa5ef2725a175a57b276ec30dc9d39c22d30410f2ede68a3eab66b3", // Replace with the specific version of the Flux model you want to use.
"input": {
"prompt": "{{ $json.prompt }}", // The user's prompt (or the enhanced prompt).
"hf_lora": "kanakjr/nov_lora", // REPLACE with your LoRA URL. Or remove this line for no LoRA.
"lora_scale": {{ $('Bot Variables').item.json.lora_scale }}, // Strength of the LoRA (0.0 to 1.0).
"num_outputs": 1,
"aspect_ratio": "1:1",
"output_format": "webp",
"guidance_scale": {{ $('Bot Variables').item.json.cfg_scale }},
"output_quality": 80,
"prompt_strength": 0.8,
"num_inference_steps": {{ $('Bot Variables').item.json.steps }}
}
}

You’ve done it! Your very own AI image artist, ready to bring your imagination to life. Start experimenting with prompts, explore different LoRAs, and have fun creating amazing images with InstaGenie!

Leave a Reply