(Even if you aren’t artistic)
A few years ago, when my friend was very unwell, I went to buy her some mangos at a nearby shop. She said they were the best mangos in London.
I sketched the shop, and later in a competition for writers, I requested 30-odd words of “Alt Text” (the text used on the web to describe images to people who are visually impaired).
The winning entry was submitted by N.J. Here’s a grainy screen-shot, with N.J.‘s Alt Text floating on top:
Recently, on one of my monthly “Dreaming in Books” sessions, someone asked what might happen if N.J.‘s Alt Text were fed into ChatGPT. Would it generate something similar to my original?
I decided to try it. Here’s what N.J.‘s text yielded:
Very impressive! But if you compare the two you’ll see there are many differences.
Could I get even closer to my original sketch? I tried several times. Each time, I altered my written instruction. Can you guess, by looking at them in sequence, what kinds of changes I requested?
Can you guess what kinds of things kept frustrating me?
I’ll paste the series of instructions I gave below the pictures, so you can see for yourself what changed – and perhaps also tell me what I could have written differently.
My Prompts to ChatGPT4
Remember, these were attempts to get closer to my original image.
Version 1. Generate a picture that shows a London Street scene on a rainy autumn day. The street is a relatively small High Street and we are looking across the road at a grocery shop with colourful boxes of different coloured fruit stacked outside and woman is walking towards the shop, dragging a bag on wheels, the image should look like it has been created using gouache paint and loose brushwork, as well as a few black lines to draw attention to details
v2. Generate a picture that shows a London Street scene on an
rainyovercast autumn day. The street is a relatively small High Street and we are looking across the road at a grocery shop with colourful boxes of different coloured fruit stacked outside and woman is walking towards the shop, dragging a bag on wheels, the image should look like it has been created using gouache paint and loose brushwork, as well as a few black lines to draw attention to detailsv3. Generate a picture that shows a London Street scene on an overcast
autumnday. We are looking directly across the smallish high street at a grocery shop with boxes of brightly coloured fruit stacked outside. A woman is walking towards the shop on the pavement from the left. She’s dragging a bag on wheels, and looks a little frail. The image should look like it has been created hastily using gouache paint and loose brushwork, as well as afewsmall number of black lines to draw attention to detailsv4. Great, but our view is more constrained. We can only see the building directly opposite us, which means the grocery and a bit of the buildings on either side. Also, these pictures are too “finished”. Make the artwork look much hastier, sketchy
***