What If Machine
Make boring camera roll not boring.
During the sun-filled August of 2024 I've realized and subsequently confirmed that the most photos we have always taken are of the sunsets, sceneries, selfies, friends, family and pets. While beautiful, quite often they are also the most boring, especially if they are from the vacation.
“No one wants to hear about your vacation.” - Matthew Dicks, Storyworthy
Can we have the AI do the hard work and turn the boring photo into an interesting photo?
I've setup a basic form on tally.so to collect a photo and send it to Flux 1.0 over at replit.
This was the first(-ish) result. Not terrible, certainly pretty, but not all that exciting. What was missing? A story. The real question was "Can we have the AI do the hard work and turn the boring photo into an interesting story?"
I've let the user select a story scenario and updated the workflow to write a story using the photo.
Which added good amount of artificial drama. We are on the right track!
But! The one panel felt limiting. I've turned it into a 3-panel story, and allowed the user to provide photo for each. This allowed users to play around with the story, without doing too much thinking.
I guess we are going to space with the precious Omega watch while munching on corn.
In rare cases story captions carried over into the comic.
Most of the time though it was gibberish. Just like the story itself. I quickly discovered writing dramatic or funny punchline for the micro story was not any of the model's strengths.
Understanding the human condition is still very much something human. Just like knowing its audience and writing for it - case in point:
When I revisited the project with the release of Dall-E 3 the text following vastly improved. The punchiness of some of the comics as well.
Eventually though I saw that both Google and OpenAI are running these projects in their labs, while using "Children's Book" and "Personalized Story" often as demos. I had my fun and decided to end the exploration.
