Google Labs has various experiments you can go to. The Experiment “Say What You See” is especially useful for anyone who wants to get better at image prompts. I’ve generated thousands of images with other systems, mostly Deep AI and SuperMachine, and it was fun and different to try this Say What You See Google Labs Image Prompts version.
So here’s how I did:
“Can you decode AI art? Test your image-reading skills and become a prompt master by guessing the prompts behind these Google-AI generated creations.”
Here’s my attempt at matching the prompt at Level 1, image 1.
I’m surprised at the level of detail in the prompt that is “tolerated” I suppose you could call it. I do a lot of image generation in the AI image generator “SuperMachine” and it gets pretty confused if there’s 3 or more elements in the prompt or if it’s too wordy.
next try at prompt: realistic photo bowl of noodles
with shredded and sliced vegetables on top
Ok so I got a 55% match – definitely not used to this prompt system. Don’t know a bowl of ramen when I see it, I guess. On the other hand, look at the “egg” – it’s green around the edge – weird.
Last one in Level One: my prompt is: soft focus photo of a pile of blue and yellow berries
Ok 52% match – I really don’t know photography lingo – what’s a “macro shot”?
Level 2: realistic photo of an ancient black stone sculpture of an Indian elephant, head only – 63% match – probably should have mentioned it was glossy
image 2: crochet image of 8 red roses with long stems, one green leaf per stem, going into tall oval vase; vase is outline-only- 64% match. Notice how AI didn’t get the number of roses correct (that happens a lot in my experience) and I said “crochet” when it was embroidery but still close
Image 3: tall rectangular impressionistic watercolor painting of 6 green spatulate fern fronds, some partially off-frame, pink background – 60% match says try again. Need to say “ferns” (what I meant to say) and impressionistic
ha! – I entered those small changes and it says “it’s the same prompt and try again” – see, AI is not really smart, but you have to learn to deal with it’s sometimes random fussiness.
new prompt: rectangular tall impressionistic watercolor painting, 6 green fern fronds,
some partially off-frame, pink background
Say What You See Google Labs Image Prompts – conclusion
Ai Image Prompts take a good bit of work to get what you are looking for. Occasionally you get lucky and hit it right off the bat, and sometimes you get something even better than what you were aiming at. I didn’t go all the way through Level 3 of Say What You See for this article – want to leave some surprises for you if you try it yourself.
Here’s something of my own using SuperMachine, and the prompt: garden gnome playing guitar with beer
Definitely get into generating your own images – it can be a lot of fun!