Illustrated history PDF book from a Telegram message

Illustrated history PDF book from a Telegram message

An n8n workflow that turns a history topic sent to a Telegram bot into an illustrated PDF book. A small OpenAI model
checks that the topic is history or an educational concept. A larger one researches it on the web and writes between 5
and 20 chapters, each with a date and an image prompt. Kie.ai draws one picture per chapter with Nano Banana Pro, and
APITemplate.io lays the chapters out as A4 pages. The PDF goes to your Google Drive, and a Google Sheet tracks the book
and every picture.

What is inside

• workflow.json: the workflow as n8n exports it, 37 nodes. Four are sticky notes that only name the stages. It holds
  no credential.
• README.md: these notes.

What we changed from the original copy

• The author's sheet and its tab id are gone. The five sheet nodes ask for YOUR_SHEET_ID and find the tab by the name
  Sheet1.
• The Kie.ai headers read Bearer <token>. They now read Bearer <YOUR_KIE_API_KEY>. The APITemplate.io header now
  reads <YOUR_APITEMPLATE_API_KEY>.
• The Telegram, OpenAI, Google Sheets and Google Drive credentials are gone, along with the author's account names on
  them, and so are the Telegram nodes' webhook ids.
• A repair in HTML to PDF: the export had dropped two backslashes from its body. It read .replace(/"/g, '\"') and
  .replace(/n/g, ''), which left every double quote in the HTML unescaped and deleted every letter n from the page.
  It now reads .replace(/"/g, '\\"') and .replace(/\n/g, ''), as the step needs to send valid JSON.
• The Cloudinary placeholders, <YOUR_CLOUD_NAME>, YOUR_PRESET_NAME and YOUR_FOLDER_NAME, were already in the file.

Before you start

• An n8n instance, on n8n Cloud or your own server. A Telegram trigger needs n8n on a public https address.
• A Telegram bot and its token, from BotFather.
• An OpenAI API key with access to gpt-5.1 and gpt-4.1-mini.
• A Kie.ai API key with credit.
• A Cloudinary account with an unsigned upload preset.
• An APITemplate.io API key.
• A Google account to sign in to Sheets and Drive with, and a Google Sheet with the tab below.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Make a Google Sheet whose first tab is named Sheet1, with the eight headers below in row 1. Copy its ID: the long
   part of its address between /d/ and /edit.
3. Add a Telegram credential with your bot's token and select it in Telegram Trigger, Send a text message and
   Send a text message1.
4. Add an OpenAI credential and select it in OpenAI Chat Model and OpenAI Chat Model1.
5. Add a Google Sheets OAuth2 credential. Select it in Create Job, Append row in sheet2, Update row in sheet,
   Update row in sheet1 and Update row in sheet2, and replace YOUR_SHEET_ID in each with the ID.
6. In nano-banana-pro and nano-banana-pro-result, replace <YOUR_KIE_API_KEY> after Bearer with your Kie.ai key.
   The brackets go too.
7. In Cloudinary, make an unsigned upload preset. In Upload Photo, replace <YOUR_CLOUD_NAME> in the address with
   your cloud name, brackets included, and put the preset's name and a folder name in place of YOUR_PRESET_NAME and
   YOUR_FOLDER_NAME.
8. In HTML to PDF, replace <YOUR_APITEMPLATE_API_KEY> in the X-API-KEY header with your key, brackets included.
   The address uses APITemplate.io's US region. Change rest-us to the region your account uses if it differs.
9. Add a Google Drive OAuth2 credential and select it in Upload PDF. The file goes to the root of My Drive.
10. Save and press Publish. Send your bot a short topic, such as the moon landing.

The sheet

Row 1 of Sheet1 holds these eight headers, spelled as here. A book takes one row and each of its pictures one more:

| Column | The book's row | A picture's row |
|---|---|---|
| ID | A new UUID | A new UUID |
| Parent ID | Empty | The book's ID |
| Concept | The topic as the classifier rewrote it | Empty |
| Type | Briefing Job | Asset Creation |
| Status | Generating..., then Completed | Queued, Generating..., then Completed |
| Prompt | Empty | The image prompt |
| Public Link | Empty | The picture's Cloudinary address |
| PDF Link | The PDF's Drive link | Empty |

How a run goes

1. Telegram Trigger takes every message sent to the bot. File Check answers a document with Files not supported.
2. AI Agent1, on gpt-4.1-mini, decides whether the text is a history or education topic. If it is not, the bot
   replies with a short refusal the model writes, and the run ends.
3. A valid topic gets a UUID from Crypto1, and Create Job adds the book's row with the rewritten topic.
4. AI Agent, on gpt-5.1 with OpenAI's web search, writes 5 to 20 chapters in date order. Each has a title, a date,
   a paragraph of three or four sentences and an image prompt that names the text to show in the picture.
5. Split Out and Crypto give each chapter its own UUID, and Append row in sheet2 queues a row for it.
6. Loop Over Items takes one chapter at a time. nano-banana-pro asks Kie.ai for a 16:9 PNG at 1K. Wait pauses 30
   seconds, nano-banana-pro-result asks for the task's state and Switch routes it: still working waits again, done
   moves on.
7. download-image fetches the picture, Upload Photo puts it on Cloudinary, and Update row in sheet1 marks the row
   Completed with its address.
8. When every chapter is done, Aggregate and HTML build one page per chapter: the title, a date badge, the picture,
   the paragraph and a footer with the chapter's ID.
9. HTML to PDF sends the page to APITemplate.io as an A4 PDF, Download PDF fetches it and Upload PDF saves it to
   Drive, named after the topic and the book's ID. Update row in sheet2 marks the book Completed with the Drive link.

Change the style

• Which topics pass: the system message in AI Agent1.
• The chapters' number, length and voice: the system message in AI Agent.
• The pictures: aspect_ratio, resolution and output_format in the body of nano-banana-pro.
• The page design: the HTML and CSS in HTML, including the fonts and the footer's text.
• Paper and margins: settings in the body of HTML to PDF.
• Where the PDF goes: Upload PDF's drive and folder.

Good to know

• The PDF never reaches Telegram. On success the run ends at the sheet, and the person who asked gets no message. Add
  a Telegram send node after Update row in sheet2 with PDF Link if they should get it.
• Open to anyone: the bot answers every message from anyone who finds it, and each book spends OpenAI, Kie.ai and
  APITemplate.io credit. Set Restrict to User IDs in Telegram Trigger to your own ID.
• Retired models: OpenAI shuts down gpt-4.1-mini on 14 January 2027, with gpt-6-luna as the replacement, and
  gpt-5.1 on 1 April 2027, with gpt-6-sol. Change OpenAI Chat Model1 and OpenAI Chat Model before those
  dates.
• A failed picture stalls the book: Switch sends a fail state nowhere, so the loop stops there. No PDF is made,
  the rows stay at Generating... and nobody is told.
• Other messages: only documents are turned away. A photo, a sticker or a voice note reaches AI Agent1 with no text,
  and the run fails there without a reply.
• No style picture: the prompt in AI Agent says an outside reference image sets the art style, but nano-banana-pro
  sends none. Each picture's style is the model's choice. Kie.ai's image_input takes up to eight reference images if
  you want one look.
• Public files: each picture sits at a public Cloudinary address, and APITemplate.io keeps the PDF on its own CDN,
  with no expiry unless you set one. The Drive link works only for people who can see the file in your Drive.
• Costs: Kie.ai bills every picture, one for each chapter. OpenAI bills both agents, plus the web searches.
  APITemplate.io counts each PDF against your plan.
• Time: each picture takes at least 30 seconds of waiting, so a long book runs for ten minutes or more.
• Your data: the topic and the chapters go to OpenAI, the image prompts to Kie.ai and the pages to APITemplate.io.
• Credentials: the JSON holds none, and nothing from its author's sheet.