Inspirobot: generating motivational images with AI

Inspirobot generated motivational image. Deep and meaningful or meaningless rubbish?

Motivational posters, whether in their original form or the short images shared on social media, can instil multiple emotions.  They can be positive or downright cringe-worthy, inspiring or bad advice… all superimposed on an image that may or may not correspond to the text.

The latest “fun” AI to go around is Inspirobot.  This AI has been trained on the form and tone of motivational images, and at the touch of a button will generate one for you.  There is a limited stock of images (I have had the same image more than once, but am still also getting new ones, so I’d estimate this is in the hundreds), but the text itself appears to be generated each time1.

Seems legit?
Possibly my favourite…

The outputs make about as much sense as the average motivational image – you can read into them a lot of different meanings.  On the face of it they might seem inspired but then you look deeper and some of them are meaningless rubbish.  You could say the same about these AI generated images.  If you are looking for deepness then you can find it, if you’re looking for something funny then you’ll find that too.

So, it’s a cool new web toy and will give everyone a giggle and even has its own subreddit, but why do it? Obviously they’ll make some money from the mugs that are linked from each generated image2.

Look a little deeper.  The website is fairly slick and it has the capacity to cope with the surges from this going viral.  It’s also a very interesting piece of technology.  The AI is generating sentences in a style, and retaining context across clauses.  It’s also choosing a placement and font size to avoid putting the text on top of the significant parts of the image3.  There’s some interesting tech here and I think with different training data it could be turned to other tasks.  An obvious idea is automatic generating of marketing copy, especially if combined with an image choice… auto-generation of social media for maximum engagement?

We’re moving into the era of artificial creativity and I’ll be watching InspiroCorp with interest.
or the mistakes of chemistry…
  1. I have tried a *lot* of examples 😉
  2.   They would have made a lot more if they could link to t-shirts, bags etc
  3.   They could be doing this manually for a small image set, but it’s far more likely that this is image segmentation .

Published by

janet

Dr Janet is a Molecular Biochemistry graduate from Oxford University with a doctorate in Computational Neuroscience from Sussex. I’m currently studying for a third degree in Mathematics with Open University. During the day, and sometimes out of hours, I work as a Chief Science Officer. You can read all about that on my LinkedIn page.