Analyse client photographs in a Google Workspace Drive folder. Classify according to poses used. Looking later to search for correlation between client poses and how many pictures the client purchased.
What is the problem & what have you tried?
I have done proof-of-concept for feeding one file to the Google Cloud Vision seek labels module, that works. The problem is the next step, what to use to analyse the results (and obviously feed the Cloud Vision module larger amounts of files, but I think I can handle that). A Google sales person suggested using Cloud Vision and then Gemini Vector search, but can that be done here with Make?
Or should I look to stay within Google’s own tools, create some agent?
I remember it’d return also x/y coordinates, but at least here there’s nothing like that. Hence I wonder if this can be used for pose analysis. Still, at least Gemini was able to quite correctly analyse the pose on a single photo I uploaded to it.
the labels module only does object detection (“Thigh”, “Hand”), not pose classification.
Easier path: feed the image straight to a vision LLM (Claude, Gemini, GPT-4o) and let it classify. Define your pose categories upfront and have it sort each image directly into them, ideally as JSON. No vector search needed.
Works well in Make via the Anthropic or Gemini module (image as base64), iterating over the Drive folder. Gemini Flash is pretty cheap for high volumes.
The vector search tip only helps if you later want fuzzy “similar pose” search. For fixed categories it’s overkill.
Two things: enforce a strict taxonomy and JSON output, otherwise it won’t aggregate cleanly across thousands of images.
What is your goal?
What is the problem & what have you tried?