Andreea @ SaaSy wanted some witty comments on her LinkedIn post*:

So let's do this! But with AI.

Step 1: Use AI to interpret the image

We'll use Midjourney to create a description of what is in the image.

Fire up the Midjourney discord bot (you do need to be on a paid subscription for this), use the /describe command, and upload the image.

Hit enter. Results:

Midjourney bot /describe example

You'll see Midjourney produces four options. Copy your preferred description and head to step 2. For reference I went for:

a person posing with an octopus costume, in the style of serenity and calm, schizowave, dark navy and yellow, smilecore, internet academia, letterboxing, unicorncore

No, I didn't know what unicorncore was either.

Step 2. Feed the image description to ChatGPT

Before we throw this into ChatGPT, we need to give it some context on the task we're asking it to do. This is my first prompt in this chat session:

Setting context for the task

Lets paste in our description:

Pasting in the description

And heres the response:

5 options for a witty response

Quick aside: Here's the same prompt in Claude, Anthropic's ChatGPT alternative:

Witty response generator in Claude

I decided to go a little further, but with ChatGPT:

Give me 5 more, but make them 200% more crazy

Not too shabby. Not shabby at all.

Step 3: Select one of the captions

It wouldn't be a good test of AI if I didn't ask ChatGPT which caption to use.

But (quite impressively) ChatGPT pushed back saying it didn't really know how to choose because it has no idea what it was for:

So, lets just feed in the original LinkedIn post:

Feeding back in some context

And the winner is:

Complete with reasoning. Love it.

But I wanted to add a little extra sparkle. Maybe some emojis could help?

Suggest emojis

And yes this is a true story. Check out the post here and check out Andreea's company SaaSy - because she's amazing.


Thoughts on improvements

This whole process took minutes. It could definitely be improved. Some initial thoughts:

  • Image interpretation: Is Midjourney the best image description tool? The description it produces is great for generatine new Midjourney images. Could something else capture the scene for more effective use in ChatGPT? Would be good to explore. That said it did a pretty good job
  • Prompt engineering. It would be worth experimenting with the prompts I gave to see if you can get more nuanced results. Theres probably better and more ways to get results than ask for something that is 200% more crazy. But again, this all worked pretty well.

Footnote:

* to be clear, I don't think she wanted me to create the witty response with AI 😆