Learn the mammals with DALL-E3

Here are the mammals! Maybe some of your favorites are pictured.

Here's the prompt I gave ChatGPT4: "Please generate a set of mammals on a plain white background, each mammal species clearly labeled."

However, ChatGPT4 is a text-generating model, so it doesn't have the ability to generate its own images. Instead, it's designed to pass my prompt to another model called DALL-E3. But not before modifying my prompt for "better results". The prompt it actually sent to DALL-E3 was: "Photo of a collection of mammals neatly arranged on a plain white background. The mammals include a lion, an elephant, a kangaroo, a panda, a dolphin, and a bat. Each mammal has a clear, legible label beneath it indicating its species name."

Then DALL-E3 generated the image above, and now we know which model to blame for the labels. Note that ChatGPT4 did not specifically ask for a giraffe, but DALL-E3 did two of them anyways.

Here's another try, in which ChatGPT turned my prompt into "Illustration of a variety of mammals displayed on a pristine white backdrop. Featured are mammal species such as a Tiger, Giraffe, Koala, Whale, Hedgehog, and Chimpanzee. Directly below each mammal is a label stating its species name in readable font."

And then DALL-E3, ever the team player, generated this set of creatures.

I also asked for the mammals in Swedish.

My prompt: "Please generate three images, each containing a set of mammals on a plain white background, each type of mammal clearly labeled in Swedish."

What ChatGPT turned my prompt into: "Photo of a variety of common mammals neatly displayed on a plain white background, with each mammal clearly labeled in Swedish. The set includes 'Lejon' for lion, 'Elefant' for elephant, 'Giraff' for giraffe, 'Tiger' for tiger, 'Varg' for wolf, 'Björn' for bear, 'Flodhäst' for hippopotamus, and 'Zebra' for zebra. Each label is written in a clear, legible font placed directly below its corresponding mammal."

What DALL-E3 did:

Now, as far as I can tell, Chatgpt4's Swedish translations are correct, including the correct concentration of umlauts. It is DALL-E3 that appears to have taken "Swedish" as a cue to umlaut everything.

I tried an experiment where I asked ChatGPT4 "Please generate a set of common Swedish mammals on a plain white background, each mammal species clearly labeled in English." Sure enough, some of the generated images had umlauts.

Bönus: Möre sets öf mämmäls!