Image caption written by Gemini and drawn onto the picture

Image caption written by Gemini and drawn onto the picture

An n8n workflow you start by hand. It downloads a sample photo and sends a smaller copy to Google Gemini, which writes a
punny title and a caption. The Edit Image node then draws both in white on a dark band along the bottom of the picture.

What is inside

• workflow.json: the workflow as n8n exports it, sixteen nodes. Eleven do the work and five are sticky notes that
  explain the steps. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The chat model named a Gemini 1.5 model. Google's models and deprecations pages, read on 12 November 2026, no longer
  list any 1.5 model, so Mindket's copy uses models/gemini-3.6-flash, which the models page lists as available with
  image input.
• The model node was linked to a credential in its author's n8n. The link is gone, so select your own.
• The file had no settings block, so Mindket's copy adds executionOrder v1, the setting n8n gives a new workflow.
• Nothing else. The prompt, the nodes, the sticky notes and the sample photo's address are the file's own.

Before you start

• An n8n instance, on n8n Cloud or your own server. The Edit Image node runs on GraphicsMagick, which n8n's own Docker
  image includes.
• A Google Gemini API key, from Google AI Studio, for a Google Gemini credential.
• A font file the Edit Image node can read. The steps below say how to pick one.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Google Gemini credential with your API key and select it in Google Gemini Chat Model.
3. Check the model in that node. The file uses models/gemini-3.6-flash. Pick another from the list if Google has
   retired it.
4. Open Apply Caption to Image and look at the Font field of its text step. The file points to an Arial file that
   n8n's own Docker image holds. On a machine without that file, pick a font from the node's list.
5. Press Execute workflow. Open the last node's output and look at the picture on its Binary tab.
6. To caption your own pictures, replace Get Image with a node that returns a file in a field named data.

How a run goes

1. The manual trigger starts the run when you press Execute workflow.
2. Get Image (HTTP Request 4.2) downloads a 600 pixel wide sample photo from a stock photo site.
3. The picture goes down two branches. Resize For AI makes a smaller copy for the model, and Get Info reads the
   original's width and height.
4. Image Captioning Agent (Basic LLM Chain 1.4) sends the smaller copy to Google Gemini Chat Model as an Image
   (Binary) human message. Its system message asks for a caption built from who, when, where and context, and a punny
   title. Structured Output Parser makes the answer two fields, caption_title and caption_text.
5. Merge Image & Caption joins the size and the caption by position.
6. Calculate Positioning (Code, run once for each item) works out the font size, the line length and where the band
   and the text go. Merge Caption & Positions brings those numbers together with the caption.
7. Apply Caption to Image (Edit Image, multi step) draws a dark rectangle and then the caption in white Arial. The
   finished picture is in the field data of its output.

Change the look

• Picture: the address in Get Image, or a node of yours that returns a file in data.
• Caption style: the system message in Image Captioning Agent.
• Fields Gemini returns: the example JSON in Structured Output Parser. Calculate Positioning and the text step read
  caption_title and caption_text, so rename them there too.
• Model: modelName in Google Gemini Chat Model.
• Band and text: the colour, font and size in Apply Caption to Image, and the numbers in Calculate Positioning.
• Where the picture goes: add a node after Apply Caption to Image, such as Google Drive, Slack or Gmail.

Good to know

• The band's height comes from a rough count of text lines, so an extra long caption can run past it.
• The two Merge nodes join by position, so send one picture through at a time.
• The font path in the file exists in n8n's own Docker image. Elsewhere the node may find no file there, so pick a font
  from its list.
• The sample photo comes from a stock photo site, and one sticky note shows an example picture loaded from an outside
  address.
• Google bills the Gemini call to your key under your plan.
• A smaller copy of each picture goes to Google.
• The file holds no credential and no key.