How can I use n8n to extract images and text from a .docx and update it with AI-generated alt texts?

Below is one approach to break down the problem into smaller parts. Although there isn’t a single out‑of‑the‑box solution that handles every step, you can combine several n8n nodes plus a little custom code to get the job done. Here’s one way to approach it:

---

1. Receiving and Reading the DOCX File

• Webhook Node:
  Use a Webhook node to receive the uploaded file from your Custom GPT. This lets n8n start a workflow with the incoming .docx file.

• Binary/File Handling:
  Once you have the file, you’ll need to extract its contents. n8n doesn’t include a dedicated DOCX parser node, so you can use a Code node (or Function node) that calls a Node.js library (such as mammoth or docx) to:

  • Extract images from a specific document section.
  • Pull additional text from another section for context.

---

2. Sending Data for Image Analysis

• Preparing Payloads:
  With the image(s) and context text extracted, you’ll generate a payload for your image analysis. You mentioned using ChatGPT Vision for image analysis. Although n8n doesn’t have a dedicated ChatGPT Vision node, you can use one of these options:

  • HTTP Request Node:
    If ChatGPT Vision (or any endpoint that offers similar functionality) is available via API, you can use an HTTP Request node (or even the Custom n8n Workflow Tool node if you’ve built a wrapper workflow) to send the image data along with the additional text context.

  • AI Agent Node:
    Alternatively, if you have a broader AI workflow set up (for example, with an AI Agent node that coordinates with a language model and tools), you can include the image analysis as one “tool” within that workflow.

• Model Choice:
  You mentioned “Claude 3.5 Sonnet” as the preferred model for generating alt texts. To call that model, you might need to set up a custom integration (perhaps via an HTTP Request node configured with the proper authentication) or adjust your AI Agent’s settings if you have support for that model available.

---

3. Receiving, Processing, and Generating Alt Texts

• Processing the Response:
  Once the visual analysis is complete, gather the AI’s responses (alt text suggestions) from each image. This might come back via your HTTP Request or as a result of your custom AI workflow.

• Custom Code Node (Post-Processing):
  Use another Code node to take the alt texts and pair them with the corresponding image data.

---

4. Reinserting Alt Texts into the DOCX

• DOCX Update:
  Similar to extraction, updating the .docx file is not provided natively in n8n. You can:

  • Use a Code node that leverages a Node.js library (for example, the docx library) to open the file and insert the alt texts at the desired location.

  • Tune your custom code to write into a specific section or perhaps use bookmark markers in your DOCX to know where the alt texts should be inserted.

• Output or Save Updated File:
  Once updated, you can either store the new file (for example, using an FTP or cloud storage node) or send it back to your desired destination (using an HTTP Request node, email, etc.).

---

Summary of Key Nodes/Approaches

1. Webhook Node: To start the workflow with the uploaded DOCX.
2. Code/Function Nodes:
   • To extract images and context text from the DOCX using an external library.
   • To update/insert the generated alt texts back into the DOCX file.
3. HTTP Request Node or Custom n8n Workflow Tool:
   • To interface with your image analysis service (ChatGPT Vision) and/or to call out to the Claude 3.5 Sonnet model.
4. (Optional) AI Agent Node:
   • To manage more advanced AI workflows if you’re integrating multiple tools within one central workflow.

---

Final Considerations

• Custom Integrations:
  Since DOCX manipulation isn’t part of core n8n nodes, you’ll be writing custom code to interact with DOCX libraries. Make sure the environment where n8n is running has access to any required Node.js modules.

• API Credentials & Endpoints:
  For both the image analysis endpoint and the Claude 3.5 Sonnet integration, be sure to set up the appropriate credentials in n8n so that your HTTP Requests or custom tool nodes function correctly.

• Testing in Steps:
  Divide your workflow into sections and test each part separately—start with file upload and extraction, then move on to image analysis, and finish with file update—to simplify troubleshooting.

---

This modular approach should give you a solid foundation for implementing your requirements in n8n. While it’s not a ready-made solution, combining these nodes with a bit of custom JavaScript should help you build the workflow you described.

Below is one potential approach using n8n’s capabilities. While there isn’t a ready-made workflow in the n8n docs that exactly matches your scenario, you can combine several nodes and custom code to put together the workflow. Here’s a high-level breakdown:

1. Receiving the .docx File

• Webhook Node
 Use the Webhook node as the entry point. Configure it to accept file uploads so that your Custom GPT can send the briefing file to n8n.

2. Parsing the .docx File

• Custom Code / Function Node
 n8n doesn’t include a built-in docx parser. You can use a Function or Code node (or even call an external API/library) to:
 – Read and parse the .docx file.
 – Extract the specific image data from the defined section.
 – Extract the additional text from another specific section for context.

(For example, you might use a Node.js library such as “mammoth” or “docx” from within a Code node if you’re comfortable writing custom code.)

3. Analyzing Images with ChatGPT Vision

• Agent or Custom AI Integration Node
 Since you want to use ChatGPT Vision (or even Claude 3.5 Sonnet, if you have custom API access), you’ll need to send the extracted images (and additional context text) to the AI model. Options include:
 – Using an AI Agent node that is configured with your AI model credentials.
 – Or, calling out to the external API using an HTTP Request node from n8n.
 The AI model would then be prompted to generate alt texts for each image.

4. Inserting the Results Back into the .docx

• Custom Code / Function Node
 Once you receive the generated alt texts, you’ll need to insert them into the appropriate location in the original briefing document.
 – This can again be handled via a Function or Code node which uses a library for editing .docx files.
 – Once modified the docx file, you can either save it for download or send it onward via another webhook or email node.

5. Orchestrating the Workflow

In summary, your overall n8n workflow might look like this:

1. Webhook Node – receives the .docx file.
2. Function/Code Node – parses the docx & extracts images and additional text.
3. HTTP Request/Agent Node – sends the content to your chosen AI (ChatGPT Vision or Claude 3.5 Sonnet) for processing and alt text generation.
4. Function/Code Node – incorporates the returned alt texts back into the original docx file.
5. (Optional) File Node / Email Node – distributes the updated file.

Points to Consider

• Custom Integration: Since working with .docx files (both for extracting and re-inserting content) isn’t available as a pre-built node, you’ll most likely need to write some custom code or use an external API.

• AI Model Integration: If you have specific requirements (for example, using Claude 3.5 Sonnet), ensure you have the API credentials and documentation. You might call these APIs directly via the HTTP Request node if they aren’t available as pre-built integrations.

• Modularity with Workflow Tools: If you want to keep your workflow modular, consider using the Custom n8n Workflow Tool node to offload any of these steps to separate workflows. This mirrors what’s done in some of the AI examples in the n8n docs.

While there isn’t an out-of-the-box solution for this exact use case, combining these nodes and techniques should provide a solid starting point. Every step might need some custom implementation, especially around parsing and editing the .docx file.

Hope this helps steer you in the right direction!