Going back to the storyboard
These are the last iteration of the images for the storyboard. I used royalty free images from unsplash.com and lightly edited them in Photoshop to create a mood and to help me work out where the drama was going to come from.
Feeling stressed, I wondered if ChatGPT might be the best way for me to generate the images for the prototype which maybe I could later then draw myself.
Using ChatGPT help me generate the images…
So the first thing I attempted was to ask ChatGPT to use the script it had developed to generate a series of images for me in the style of Edvard Munch and in a green tone. It started out by generating this image of the anxious young man at a train station.
I liked the fact that the image looked so polished but wanted it to look more modern. I asked it to update the setting for 2025:
I thought that this was a good starting point and then decided to ask it to make the rest of the images specifying the dimension I needed for my game but in the style of Picasso’s blue period. This is what it did next:
It then gave me this image:
This scene is was not anything to do with the script and I started to get a bit frustrated.
When I tried to get it back on task it made this image:
A bit strange but aesthetically pretty and in keeping with the style I requested. The train looks uncanny but I thought I would push on with the image generation. I then came up against an image limit and it stated I would have to wait till tomorrow or buy some credits. Still feeling stressed, I decided to pay for a month’s subscription to GPT. It thanked me and then said it would take a couple of hours to make all the images.
I waited a few hours, prompted it again and then it said it would take some more hours.
The next day, realising that something had gone wrong, I tried to explain this to ChatGPT. It apologised and said it would get back to it. It produced a single image each time and needed additional prompting to continue. When I asked it to please make all of the images it generated this:
Again, although there are some aesthetically pretty images, it is not what I asked for. The images as a whole are inconsistent, have strange artifacts in them, the text has spelling mistakes and the dimensions of the images make them unusable.
Feeling extra stressed because of the time I had lost, I decided to cancel the subscription and try and get a refund. I then got stuck in a loop with the chatbot where it kept telling me to login to process my refund, but every time I logged in, it prompted me to do the same thing. I managed to get an email sent to a human agent who processed my request the following day. No mention was made of the poor service that I had got from ChatGPT or the fact that the chatbot did not work. I decided to stop wasting my time and to allocate a day to get the drawing done by hand.
Keeping things simple and going back to digital drawing
I decided to dedicate a whole day to the drawing process and started out with a pencil drawn storyboard. I used the script to generate the scenes:
I used reference images that I collected on a recent night train journey to help me construct the train scenes:
The finished images
I started with an idea to work with paper and ink but after doing a few test images, decided that this would need much more time. These are some of the images that I planned to use for the ‘desire’ mini-game:
I worked through each image using a monochrome palette, trying to be quick and expressive.
I took inspiration from the art of the magazine https://www.adbusters.org/ and used a scanned magazine cover to add texture to the digital images.
The texture:
The finished images:
Next steps
Now I have the images and the text, time to finish constructing the game in Godot and to work out the mechanics and music.
