MensaApp icon indicating copy to clipboard operation
MensaApp copied to clipboard

Check image appropriateness using the free Gemini API

Open brodmo opened this issue 1 year ago • 1 comments

Images are often unrelated to the actual food.

The images along with the description of the food could be sent to the the free Gemini 2.0 Flash API to check whether they match. If they don't, reject the image. Perhaps the reason the LLM provides could be shown to the user.

From my testing, Gemini is reasonably good at this task. I used the following prompt, which could be further refined: Bildet das Bild folgendes ab: "<Beschreibung des Gerichts>"? Antworte mit "Ja" oder "Nein", gefolgt von einer kurzen Begründung.

The rate limit is 15 requests per minute and 1500 requests per day. I assume this should be plenty.

brodmo avatar Feb 26 '25 12:02 brodmo

Thank you for this issue, we will look into that.

worldofjoni avatar Feb 28 '25 10:02 worldofjoni