GPT-4 Photo Analysis Capabilities

Hello everyone! I had a question about GPT-4 capabilities. I have been using my custom web app connected to GPT-3.5-Turbo for a little while now, which is working fine for my needs. Although I really feel as though the image analysis aspect will be a great addition to what I offer. I am curious if anyone has heard any updates on this? I have recently been given access to the GPT-4 playground (I already pay for ChatGPT premium though for my own use/tasks that are not client-facing).

I am curious when GPT-4 will have the ability to use uploaded photos as a prompt? Essentially I want it to analyze the photo and produce an end result/summary in a text response. If it can do that then I would be extremely happy. Especially if I can control the amount of detail it would produce, for example… if it was a man holding an apple, you could choose the degree of complexity that you receive back:

Temperature 1: Man holding an apple.
Temperature 3: An elderly man holding a red apple with his right hand.
Temperature 7: An elderly man in a field standing and holding a small red apple in his right hand smiling on a sunny day. He is holing a cane in his left hand and he is wearing glasses. There are mountains behind him.

I don’t know… that is just an unrelated example to get my point across, but that would be really cool and useful for me to receive a text response like that, that is usable in an API sense.

Hopefully someone has some insight on that. I remember that in the demo but have not heard anything since then and just went over everything available and I still just see the info on “image generation/edit” but not photo prompt analysis.

IF that is an extremely long time away, I have briefly checked Amazon’s similar service Rekognition, since I use their S3 buckets for a couple things and saw that while poking around… but have not dove
any deeper into that. I saw that you need to upload a bunch of photos for that and train it yourself. That could be very interesting and I would be willing to put time into something like that, although if GPT-4 is just going to do it for me anyway then maybe that wouldn’t be necessary?

Thanks for any help! Have a great day.

4 Likes