Making in AI
Posted On March 18, 2023
I’ve recently been ‘playing around’ with AI image generation. I think of them as ‘text based work’ as the images come from the text prompts. You can get interesting images out of a basic text prompt, but if you have an idea about what you actually want it to look like, you can end up putting hours into re-writing prompts in order to get what you want.
And sometimes you can spend ages and ages and not get any closer to what you want. I was trying to get a midshot photo of a grandmother ‘shooing’ a bee away from a tin can ful of raspberries held by a little girl in a forest at the Golden Hour shot on Afga Vista 400 film.
I tend to love images when they go wrong in some way – weird fingers, a strange facial expression- but you can get some atmospheric things if you work at it.
You can also get interesting things randomly – a friend on Facebook posted about how her daughter had ‘invented’ a ‘bunny mermaid’… so I popped that in.
There is, however, a lot of chance involved in using AI for image generation. Yes, you can use other images as prompts, add different perameters to the prompts and you can use other bots and apps to alter or manipulate the images created, but pure AI image generation isn’t something you have total control over. And I like that.
I like that I have control of the ‘writing’ – the text prompts – but then hand that over to the AI, which comes up with the image.
Using the internet meme Woman Laughing Alone With Salad I can show you a few examples of how you can get different images out of Midjourney. I will just do one pass at each prompt and post all four images I’m given.
Prompt: “A woman laughing alone with a salad. –ar 16:9 –v 5.1”
(the parameter –ar stands for ‘aspect ratio’ and –v determines the version of Midjourney you want to use)
This produces an image that looks like it’s been illustrated or painted.
Prompt: “A photograph of a woman laughing alone with a salad. –ar 16:9 –v 5.1”
This produces something closer to a photograph. Notice the watermark on the lower righthand image. This tells you that it’s been trained on stock photos.
Prompt: “A photograph of a woman laughing alone with a salad. –ar 16:9 –v 5.1 –style raw”
The addition of the perameter –style raw (which is available in MJ5) tones down the stylised look of the image.
Prompt: “A photograph of a woman laughing alone with a salad. –ar 16:9 –v 5.1 –style raw –c 25”
Adding the perameter –c followed by a number 1-100, which stands for ‘chaos’, produces more ‘unexpected’ compositions the higher the number you use.
Let’s try an adverb in the prompt: “A photograph of a woman wildly laughing alone with a salad. –ar 16:9 –v 5.1 –style raw –c 100”
Finally, let’s add some more action to it: “A photograph of a woman wildly laughing so hard that she has fallen on the floor alone with a salad. –ar 16:9 –v 5.1 –style raw –c 100”
I’ll generate another bunch with that prompt…
And then just for fun: “A photo of a laughing salad woman. –ar 16:9 –v 5.1 –style raw”