See llms.txt for all machine-readable content.

Back to Templates

Analyze images with OpenAI Vision while preserving binary data for reuse

Created by

Created by: Robert Breen || rbreen
Robert Breen

Last update

Last update 4 months ago

Categories

Share

Use this template to upload an image, run a first-pass OpenAI Vision analysis, then re-attach the original file (binary/base64) to the next step using a Merge node. The pattern ensures your downstream AI Agent (or any node) can access both the original file (data) and the first analysis result (content) at the same time.

✅ What this template does

  • Collects an image file via Form Trigger (binary field labeled data)
  • Analyzes the image with OpenAI Vision (GPT-4o) using base64 input
  • Merges the original upload and the analysis result (combine by position) so the next node has both
  • Re-analyzes/uses the image alongside the first analysis in an AI Agent step

🧩 How it works (Node-by-node)

  1. Form Trigger
    • Presents a simple upload form and emits a binary/base64 field named data.
  2. Analyze image (OpenAI Vision)
    • Reads the same data field as base64 and runs image analysis with GPT-4o.
    • The node outputs a text content (first-pass analysis).
  3. Merge (combine by position)
    • Combines the two branches so the next node receives both the original upload (data) and the analysis (content) on the same item.
  4. AI Agent
    • Receives data + content together.
    • Prompt includes the original image (=data) and the first analysis ({{$json.content}}) to compare or refine results.
  5. OpenAI Chat Model
    • Provides the language model for the Agent (wired as ai_languageModel).

🛠️ Setup Instructions (from the JSON)

Keep it simple: mirror these settings and you’re good to go.

1) Form Trigger (n8n-nodes-base.formTrigger)

  • Path: d6f874ec-6cb3-46c7-8507-bd647c2484f0 (you can change this)
  • Form Title: Image Document Upload
  • Form Description: Upload a image document for AI analysis
  • Form Fields:
    • Label: data
    • Type: file
  • Output: emits a binary/base64 field named data.

2) Analyze image (@n8n/n8n-nodes-langchain.openAi)

  • Resource: image
  • Operation: analyze
  • Model: gpt-4o
  • Text: =data (use the uploaded file field)
  • Input Type: base64
  • Credentials: OpenAI (use your stored OpenAI API credential)

3) Merge (n8n-nodes-base.merge)

  • Mode: combine
  • Combine By: combineByPosition
    • Connect Form Trigger → Merge (input 2)
    • Connect Analyze image → Merge (input 1)
    • This ensures the original file (data) and the analysis (content) line up on the same item.

4) AI Agent (@n8n/n8n-nodes-langchain.agent)

  • Prompt Type: define
  • Text:
  • System Message: analyze the image again and see if you get the same result.
  • Receives: merged item containing data + content.

5) OpenAI Chat Model (@n8n/n8n-nodes-langchain.lmChatOpenAi)

  • Model: gpt-4.1-mini
  • Wiring: connect as ai_languageModel to the AI Agent
  • Credentials: same OpenAI credential as above

Security Note: Store API keys in Credentials (do not hardcode keys in nodes).

🧠 Why “Combine by Position” fixes the binary issue

  • Some downstream nodes lose access to the original binary once a branch processes it.
  • By merging the original branch (with data) and the analysis branch (with content) by position, you restore a single item with both fields—so the next step can use the image again while referencing earlier analysis.

🧪 Test Tips

  • Upload a JPG/PNG and execute the workflow from the Form Trigger preview.
  • Confirm Merge output contains both data (binary/base64) and content (text).
  • In the AI Agent, log or return both fields to verify availability.

🔧 Customize

  • Swap GPT-4o for another Vision-capable model if needed.
  • Extend the AI Agent to extract structured fields (e.g., objects detected, text, brand cues).
  • Add a Router after Merge to branch into storage (S3, GDrive) or notifications (Slack, Email).

📝 Requirements

  • n8n (cloud or self-hosted) with web UI access
  • OpenAI credential configured (Vision support)

🩹 Troubleshooting

  • Binary missing downstream? Ensure Merge receives both branches and is set to combineByPosition.
  • Wrong field name? The Form Trigger upload field must be labeled data to match node expressions.
  • Model errors? Verify your OpenAI credential and that the chosen model supports image analysis.

💬 Sticky Note (included in the workflow)

“Use Binary Field after next step” — This workflow demonstrates how to preserve and reuse an uploaded file (binary/base64) after a downstream step by using a Merge node (combineByPosition). A user uploads an image via Form Trigger → the image is analyzed with OpenAI Vision → results are merged back with the original upload so the next AI Agent step can access both the original file (data) and the first analysis (content) at the same time.

📬 Contact

Need help customizing this (e.g., filtering by campaign, sending reports by email, or formatting your PDF)?