bear cub climbing on a log, surrounded by purple and yellow wildflowers

AI Generated Images

I am fascinated by the way AI works!

Adobe opened up the beta testing of their Firefly AI image generation application to the public. It's interesting to see where its strengths and weaknesses lie...

It did an excellent job of creating images in the style of different art materials, such as watercolor or oil painting.

It also handled instructions well that included directions on where the light was coming from, or what type of light to use (early morning sunlight, bright sun filtering through tree leaves, light coming in from a window on the right side).

It was good at interpreting organic subjects, including sand dunes, flowers, clouds, ferns.

Some of the most interesting results were when I asked it to create images of mechanical items... Without knowledge about how items work, it presented me with images that displayed the different components of a car engine, a hammer, a saw, or playground equipment, but they were assembled in a way that wouldn't function.

I found it to be a little more constrained by reality that I thought it would be. It struggled when I asked it to to put tulips next to cacti, fill a swimming pool with ice cream, make a globe out of blue and green jellybeans, or show a forest with snow covered aspen with yellow leaves. I described a scene with a fireplace, and it occasionally had flames on the mantel or on a sofa by the fireplace.

Images with animals were highly unpredictable. I was presented with a great photograph of a bear cub (well, his right leg is a little off, but the face and background are so pretty that it distracts me from the imperfections), but a wolf with legs that were a tangled mess. I had a lot of trouble with dogs and cats, but generated some images of birds that were beautiful.

I hope that Adobe keeps the beta period open for a long time... it has been so interesting to experiment and learn how the AI interprets things! I'm being pessimistic and assuming that once they take it live and come up with a pricing structure, I'll probably be priced out, so I'm enjoying the experimentation while I can.

See below for some examples. The captions include the prompt that I typed to generate the image.

playground equipment in a park

AI Generated Photo of a cartoon squirrel

cartoon style happy squirrel with hiking boots and a trekking pole

hammer, saw, and a pile of nails on a workbench

soft watercolor painting of a bluejay closeup

AI Generated Photo of a man on a bicycle