Telegram marketing agent for images, videos and posts

Telegram marketing agent for images, videos and posts

An n8n workflow for a Telegram bot that takes marketing requests by text, voice note or captioned photo. A Gemini agent
reads each request and hands it to one of five sub-workflows of your own: an image, a video, or a post for LinkedIn,
Instagram or Facebook. The sub-workflows do the making. This file is the front desk.

What is inside

• workflow.json: the workflow as n8n exports it, 24 nodes. Six are sticky notes that name its parts: the trigger, the
  three kinds of message, the agent, its model, the tools and the reply. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The five tool nodes pointed at sub-workflows in the author's own n8n: Create Image, Create Video, LinkedIn Post,
  Insta post and Facebook post. Each now arrives empty, and you pick your own.
• A voice note never reached the agent. Transcribe Audio returns Gemini's answer as content.parts[0].text, while
  Marketing Team Agent read only $json.text, so it got an empty prompt in place of what was said. Its prompt now reads
  {{ $json.text || $json.content.parts[0].text }}, which leaves text and photo messages as they were.

Before you start

• An n8n instance, on n8n Cloud or your own server, reachable from the internet so Telegram can call it.
• A Telegram bot and its token, from BotFather.
• A Gemini API key from Google AI Studio.
• A Cloudinary account with an unsigned upload preset, for photos.
• Five sub-workflows of your own, one for each tool (below), and whatever keys they need. The video one gets a Kie.ai
  key from this workflow.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Telegram credential with your bot's token. Select it in Telegram Trigger, Download Voice File, Get a file
   and Telegram.
3. Add a Google Gemini (PaLM) API credential with your key. Select it in Transcribe Audio and 2.5 Flash.
4. Open API Values and fill the five fields: Cloud name, folder and Upload Preset from Cloudinary, then your
   Gemini key in Gemini API and your Kie.ai key in Kie ai API.
5. Build or import the five sub-workflows. Each starts with an Execute Workflow Trigger that accepts the fields listed
   below.
6. Open Create Image, Video, LinkedIn, Instagram and Facebook, and pick the matching sub-workflow in each.
7. In Telegram Trigger, add your Telegram user ID under Restrict to User IDs, so strangers cannot use the bot.
8. Save and press Publish. n8n registers the bot's webhook with Telegram when the workflow goes live.

What each sub-workflow receives

All five get chatID, the Telegram chat the request came from, so each can send its own result to the user. The rest
of the fields come from the agent, which fills them from the message, or from API Values.

Create Image   imageTitle (four words at most), imagePrompt, chatID
Video          Image link, User request, chatID, Cloud name, folder, Upload Preset, Kie.ai API
LinkedIn       postTopic, targetAudience, chatID, Gemini API
Instagram      postTopic, targetAudience, chatID, Gemini API
Facebook       postTopic, targetAudience, chatID, Gemini API

How a run goes

1. Telegram Trigger receives each new message, and API Values adds your five values to it.
2. Switch sorts the message, and the first rule that matches wins. A voice note goes to Download Voice File and
   Transcribe Audio, which has gemini-2.5-flash write out what was said. Plain text goes to Set 'Text'. A photo with
   a caption goes to Get a file, then Image Upload puts it on Cloudinary, and Set writes the request as the image's
   address plus the caption.
3. Marketing Team Agent reads the request with the model in 2.5 Flash and calls one tool. Its system message tells it
   never to ask questions or write content itself. When it is unsure it uses Think, and when it still cannot decide it
   makes an image.
4. The chosen sub-workflow runs and does the work.
5. Telegram sends the agent's last words to the chat, or Done. when it has none. The prompt asks for one fixed line on
   success and another on failure.

A photo sent with no caption matches no rule, so the bot does not answer it.

Change the look

• What the agent may do: its system message in Marketing Team Agent. Add a tool there and in the canvas to give it a
  sixth job.
• The model: 2.5 Flash runs models/gemini-2.5-pro, despite its name. Pick a Flash model there for lower cost.
• Which messages it accepts: the three rules of Switch.
• Where photos go: folder in API Values.
• The reply: the text of Telegram, and the two fixed lines in the agent's prompt.

Good to know

• Who can use it: without a user restriction on Telegram Trigger, anyone who finds the bot can spend your Gemini,
  Cloudinary and Kie.ai credit.
• Photo size: Get a file takes the second of the sizes Telegram sends with each photo, a medium copy, and that is what the
  video sub-workflow gets. Use {{ $json.message.photo.last().file_id }} in its file ID for the largest.
• A failed upload: Image Upload carries on after an error, so the agent then gets an empty image address.
• No memory: the system message mentions a memory, but the agent has none, so each message stands alone.
• One call or several: the prompt says to make exactly one tool call, and later to call every tool a request names.
  Expect one result when you ask for posts on two networks at once.
• Models: Google's deprecations page, read on 2 November 2026, gives no shutdown date for gemini-2.5-pro or
  gemini-2.5-flash.
• Keys in a plain node: the Gemini and Kie.ai keys in API Values travel to the sub-workflows as fields, and they show in
  every run's execution data. Take them out before you share the workflow.
• Your data: voice notes and requests go to Google, photos become public on Cloudinary through the unsigned preset, and
  the sub-workflows get whatever the agent passes on.
• Costs: each message costs a gemini-2.5-pro agent run, a voice note one gemini-2.5-flash call more, and each
  sub-workflow bills on its own.