Comparing Apples to Apples

To kick off the Daily A.I. section of this site, I wanted to start with something simple like how well the different services can render an apple.

In the future, we’ll be doing other simple tests like this one, exploring different use cases for GPT-3 and/or AI Image generation, and work through prompt-crafting. DALL-E 2 recently went into beta and I’m keen to get an invite and begin including that tool in these tests.

Services Used

As you’ll see, Craiyon (formerly DALL·E mini) is quite good at generating simple images of common objects. This makes sense and is expected compared to the more ‘artsy’ AI generators like Wombo or Starry AI that seem to take away a lot of your control through the prompt, especially for very simple prompts.

It was surprising to me how poorly MidJourney performed without being given more information. Several generations gave me oddly colored and drawn apples. Outside of simple everyday objects, MidJourney has been most impressive to me in its ability to create complex believable scenes.

Performed Best

CraiyonMidJourney
“Apple”image-20220722181125094Image
“A billion Apples”image-20220722180653850Image
“Granny Smith Apple. Photorealistic. Ultra detailed."image-20220722181446473Image

Performed Worst

Wombo.artStarry A.I.
“Apple”image-20220722180915548image-20220722182002562
“A billion Apples”image-20220722181333333image-20220722182035302
“Granny Smith Apple. Photorealistic. Ultra detailed."image-20220722181531874image-20220722182103579

Results

Clearly, in this test, Craiyon wins hands down against all other tools. Its poor image resolution and lack of upscaling make even these great results difficult to use outside of playing around. There are some image upscalers that use some kind of AI and have better results than you will in GIMP or Photoshop, but honestly they’re still not incredibly impressive. Cutout.pro is the free, best one I’m currently aware of.