The idea is:
Please let us pass external image URLs to the AI Agent instead of requiring them to be converted to binary data.
My use case:
My use case is simply any kind of chatting with AI.
I think it would be beneficial to add this because:
Base64 eats up thousands of tokens, large portions of any available context window, and are a drastic waste. Most modern multi-modal model providers allow the image url to be passed and it cost, on average, around 765 tokens for an image that has been optimized… Base64, for the same image costs 22,000+ tokens. Requiring Base64 is sort of archaic at this point. I know local models require Base64, so this would need to be dependent of the provider, but Vision capabilities are dependent on the provider/model anyway, so it should be an option for those that are compatible. The AI Agent node stringifies all JSON it passes, making it impossible to pass the standard JSON format needed to pass an external image URL, and it just seems a bit silly that there is no way to pass such a small bit of JSON to the model… So in other words, not only is there no option to pass the image url, but there is no way to workaround it either via passing our own JSON snippet. Please know I mean this all 100% respectfully… I think n8n is still the most amazing tool in the AI market today and I have nothing but love for what you all do. <3