Local Ollama Vision Model Comparison and Google Docs Documentation - n8n Workflow

AI Developers & Enthusiasts: Those testing and benchmarking different local LLMs using Ollama.
Data Analysts: Users needing exhaustive, structured descriptions extracted from visual data.
Automation Specialists: Anyone seeking reliable n8n templates for advanced local AI integration.
Researchers: Individuals requiring detailed, documented image analysis for reporting.

Determining the best Vision Language Model (VLM) for a specific task often requires direct comparison. This sophisticated n8n workflow automates that process by fetching an image from Google Drive, processing it against a user-defined list of local Ollama models (e.g., granite3.2-vision, llama3.2-vision), and documenting the output.

This automation solves the challenge of manually running tests across multiple endpoints. It leverages a robust prompt structure within an n8n node to demand comprehensive, structured markdown output, ensuring quality results. The entire operation is managed within a single, efficient n8n workflow, allowing users to rapidly iterate and find the best performing local VLM for their needs. This provides an excellent example of utilizing an n8n trigger to initiate complex local AI tasks.

The entire process is initiated by the manual n8n trigger, When clicking ‘Test workflow’.

1. Image Retrieval: The workflow first uses the Google Doc Image Id n8n node to define the target file ID, followed by the Download Image File from Google Drive n8n node to retrieve the image binary.


2. Base64 Conversion: The Get Base64 String n8n node converts the downloaded image into a Base64 string, which is necessary for the Ollama Vision API call.


3. Model Setup: The List of Vision Models n8n node defines the array of local Ollama models to test. The Split List of Vision Models for Looping n8n node then prepares this array for iteration.


4. Iteration and Prompting: The Loop Over Ollama Models n8n node iterates through each selected model.

1. AI Analysis: The Ollama LLM Request n8n node, configured as an HTTP Request, sends the payload to the local Ollama endpoint (http://127.0.0.1:11434/api/chat).


2. Documentation: The Create Result Objects n8n node structures the model's response, and the Save Image Descriptions to Google Docs n8n node appends the comparative analysis, clearly labeled by model, into the target Google Docs file. This ensures every test run is properly documented by this powerful n8n workflow.

To set up this powerful n8n workflow, follow these steps:

1. Import: Copy the provided JSON into your n8n instance using the 'New' menu and 'Import from JSON'.


2. Ollama Setup: Ensure you have a local Ollama instance running and have pulled the required vision models (e.g., granite3.2-vision).


3. Credentials: Configure your Google Drive and Google Docs credentials within n8n. These are required for the Download Image File from Google Drive n8n node and the Save Image Descriptions to Google Docs n8n node.


4. Input Configuration:

1. Execution: Click the 'Test workflow' n8n trigger to run the comparison.

When clicking ‘Test workflow’ (n8n trigger): This manual n8n node initiates the entire comparison process.
Download Image File from Google Drive (Google Drive n8n node): Function is to retrieve the target image file using the provided file ID.
Get Base64 String (Extract From File n8n node): Converts the binary image data into a Base64 encoded string, preparing it for the Ollama API request.
List of Vision Models (Set n8n node): Defines an array of local Ollama models (e.g., granite3.2-vision) that will be tested sequentially.
Split List of Vision Models for Looping (Split Out n8n node): Takes the array of models and creates individual items, enabling the downstream loop.
Loop Over Ollama Models (Split In Batches n8n node): Controls the iteration, ensuring each model runs through the image analysis pipeline.
General Image Prompt (Set n8n node): Sets a highly structured, exhaustive user_prompt requiring four sections of analysis (Inventory, Contextual, Spatial, Textual). This prompt configuration is key to the effectiveness of this n8n workflow.
Create Request Body (Set n8n node): Dynamically builds the final JSON payload for the Ollama API, merging the current model name, the detailed prompt, and the Base64 image data.
Ollama LLM Request (HTTP Request n8n node): Executes the vision request against the local Ollama server (http://127.0.0.1:11434/api/chat).
Save Image Descriptions to Google Docs (Google Docs n8n node): Appends the resulting analysis (including the model name) to a specified Google Docs file, providing clean documentation for the n8n workflow output.