[n8n] Improve Google Drive node to handle binary inputs directly

I built an automation that takes in a URL of a product collection or catalog page for any fashion brand or clothing store online and can bring each product to life by animating it with model demonstrating how the product looks and feels with Veo 3.1. This allows brands and e-commerce owners to easily demonstrate what their product looks like much better than static photos and does not require them to hire models, setup video shoots, and go through the tedious editing process. **Here’s a demo of the workflow and output:** https://www.youtube.com/watch?v=NMl1pIfBE7I ## Here's how the automation works ### 1. Input and Trigger The workflow starts with a simple form trigger that accepts a product collection URL. You can paste any fashion e-commerce page. In a real production environment, you'd likely connect this to a client's CMS, Shopify API, or other backend system rather than scraping public URLs. I set it up this way just as a quick way to get images quickly ingested into the system, but I do want to call out that no real-life production automation will take this approach. So make sure you're considering that if you're going to approach brands like this and selling to them. ### 2. Scrape product catalog with firecrawl After the URL is provided, I then use Firecrawl to go ahead and scrape that product catalog page. I'm using the built-in community node here and the extract feature of Firecrawl to go ahead and get back a list of product names and an image URL associated with each of those. In automation, I have a simple prompt set up here that makes it more reliable to go ahead and extract that exact source URL how it appears on the HTML. ### 3. Download and process images Once I finish scraping, I then split the array of product images I was able to grab into individual items, and then split it into a loop batch so I can process them sequentially. Veo 3.1 does require you to pass in base64-encoded images, so I do that first before converting back and uploading that image into Google Drive. The Google Drive node does require it to be a binary n8n input, and so if you guys have found a way that allows you to do this without converting back and forth, definitely let me know. ### 4. Generate the product video with Veo 3.1 Once the image is processed, make an API call into Veo 3.1 with a simple prompt here to go forward with animating the product image. In this case, I tuned this specifically for clothing and fashion brands, so I make mention of that in the prompt. But if you're trying to feature some other physical product, I suggest you change this to be a little bit different. Here is the prompt I use: ```markdown Generate a video that is going to be featured on a product page of an e-commerce store. This is going to be for a clothing or fashion brand. This video must feature this exact same person that is provided on the first and last frame reference images and the article of clothing in the first and last frame reference images.|In this video, the model should strike multiple poses to feature the article of clothing so that a person looking at this product on an ecommerce website has a great idea how this article of clothing will look and feel.Constraints:- No music or sound effects.- The final output video should NOT have any audio.- Muted audio.- Muted sound effects. ``` The other thing to mention here with the Veo 3.1 API is its ability to now specify a first frame and last frame reference image that we pass into the AI model. For a use case like this where I want to have the model strike a few poses or spin around and then return to its original position, we can specify the first frame and last frame as the exact same image. This creates a nice looping effect for us. If we're going to highlight this video as a preview on whatever website we're working with. Here's how I set that up in the request body calling into the Gemini API: ``` { "instances": [ { "prompt": {{ JSON.stringify($node['set_prompt'].json.prompt) }}, "image": { "mimeType": "image/png", "bytesBase64Encoded": "{{ $node["convert_to_base64"].json.data }}" }, "lastFrame": { "mimeType": "image/png", "bytesBase64Encoded": "{{ $node["convert_to_base64"].json.data }}" } } ], "parameters": { "durationSeconds": 8, "aspectRatio": "9:16", "personGeneration": "allow_adult" } } ``` There’s a few other options here that you can use for video output as well on the Gemini docs: https://ai.google.dev/gemini-api/docs/video?example=dialogue#veo-model-parameters ## Cost & Veo 3.1 pricing Right now, working with the Veo 3 API through Gemini is pretty expensive. So you want to pay close attention to what's like the duration parameter you're passing in for each video you generate and how you're batching up the number of videos. As it stands right now, Veo 3.1 costs 40 cents per second of video that you generate. And then the VO3.1 fast model only costs 15 cents per second, so you may honestly want to experiment here. Just take the final prompts and pass them into Google Gemini that gives you free generations per day while you're testing this out and tuning your prompt. ## Workflow Link + Other Resources - YouTube video that walks through this workflow step-by-step: https://www.youtube.com/watch?v=NMl1pIfBE7I - The full n8n workflow, which you can copy and paste directly into your instance, is on GitHub here: https://github.com/lucaswalter/n8n-ai-automations/blob/main/veo_3.1_product_photo_animator.json