AI Image generation is quite expensive, comparitively

Please see exhibit A. My AI cost per day using LLMVision to send notifications about activity from my doorbell camera. I wanted to experiment with AI image generation as detailed in this video from J-Lo.

  • In September, only text was generated. it cost maybe half a cent per day. I typically ended up with monthly bills of 25 to 30 cents.
  • October 1st - 4th, I was experimenting, doing manual image generation, writing some test scripts, etc.
  • October 5th and 6th, every LLMVision event was being redrawn as a cartoon. It was very amusing. My family got a kick out of all the different interpretations of themselves. But wow, it got expensive quickly. After October 6th my monthly bill was already around $7!

So, I quickly put a stop to that fun little experiment.

Edit: FYI, I was using Gemini 2.5 Flash

Interesting, thank you for sharing. I still don’t personally see the use case for image generation but I can see the fun factor.

Back when I was trying Gemini (lots of errors despite being on a paid plan), I recall it being hard/impossible to set a monthly limit; is that still the case? OpenAI allows you to set a budget not to exceed and alerts. I recall when I started working with LLM Vision I was doing a lot of testing and had a few days in the multiple dollars, but now it’s settled into about $0.30/day on average (analysis of static images only).

Yeah, that was the only reason that I knew this was so bad. I got notified that I had used half of my monthly budget of $10. I had set that randomly when I started using the service because I had no idea how much it would cost.

1 Like

Dynamic UI on demand…

Imagine a 16*9 format photo with your AI agent presenting the same options they’re presenting you in voice as a fast touch screen interface ready to be plugged into a picture card so it can have touch zones. Which equate to the Pic zones. Totally possible now if not a tich slow

1 Like