Differences in results between Google Gemini Node and direct HTTP Request (Audio Analysis)

Levan · February 1, 2026, 4:06pm

Question

I am trying to replicate the behavior of the native Google Gemini “Analyze Audio” node using a manual HTTP Request node(https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-2.5-flash:generateContent) to hit Vertex AI, but I’m seeing significant differences in results and token usage. I have a few questions about how the native node is configured “under the hood”:

What are the default Temperature, Top-P, and Top-K values used by the native node?
Does the native node include any hidden System Instructions or “System Prompt” wrappers that aren’t visible in the UI?

Information on your n8n setup

n8n version: 1.123.18
Running n8n via (Docker, npm, n8n cloud, desktop app): cloud

achamm · February 1, 2026, 5:06pm

Hey @Levan, I checked your link, and it went to a invalid URL - “This aiplatform.googleapis.com page can’t be found

No webpage was found for the web address: https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-2.5-flash:generateContent

HTTP ERROR 404”

I also did some research and also checked the N8N documentation for the node, and I couldn’t find any details on temperature, Top-P, Top-K either. However, this is just a guest but the default might match the default through the actual correct URL, though I’m not sure.

I didnt see any system instructions but you can use the text input to help describe the audio, for a better response. You can also set the max tokens it returns for the analysis description.

Hope this Helps! If you have any more questions, feel free to ask!

mohamed3nan · February 1, 2026, 6:13pm

Hi @Levan, welcome!

The Gemini node does not use the Vertex endpoints (it uses https://generativelanguage.googleapis.com,

As per the source code:

github.com/n8n-io/n8n

packages/%40n8n/nodes-langchain/nodes/vendors/GoogleGemini/helpers/baseAnalyze.ts

89f8fb04c


      
          				role: 'user',
          				parts: filesData.map((fileData) => ({
          					fileData,
          				})),
          			},
          		];
          	}
          
          	contents[0].parts.push({ text });
          
          	const body = {
          		contents,
          		generationConfig,
          	};
          
          	const response = (await apiRequest.call(this, 'POST', `/v1beta/${model}:generateContent`, {
          		body,
          	})) as GenerateContentResponse;
          
          	if (simplify) {
          		return response.candidates.map((candidate) => ({

The only parameter sent is maxOutputTokens, which is already an option in the node UI,

also the Text Input is sent as part of the body content.

I believe all other settings (Top-P, Top-K, etc.) use the defaults for the /v1beta/${model}:generateContent endpoint..