[Showcasing some of the pictures that my daughter and I created with the open source generative model Dalle-Mini. No deeper insights, just an appreciation how well it works in some cases.]

I have been aware of a small open source Dall-e version for some time now. My initial check left me with the impression that it was way worse than Dall-e or Dall-e 2 and not really worth engaging with.

However, recently I started to implore Dall-e Mini together with my little daughter who is less critical of machines that can create any picture she proposes. Originally we ended every prompt with " - a children's book illustration".

We quickly got better at prompting and realized that there were some things Dall-e Mini couldn't do very well, like faces and animals and complex scenes, but also others where it wasn't as far behind of the SOTA-models as you'd reasonably expect from the model size. (Above a ship sailing through the clouds with a full moon, a castle in the clouds and a magical door in an enchanted forest.)

We switched to " - a beautiful fantasy illustration" to often get just that. These prompts all start with a version of "A beautiful magical enchanted golden ... ".

We created swords ...

And magical artefacts ...

Birds ...

And beautiful magical enchanted golden eyes ...

Flowers ... yes, magical, beautiful, enchanted ones.

Magical, beautiful, enchanted, golden bananas ...

Crowns ...

Magical, beautiful, enchanted, golden glasses ...

And other stuff ...

I might also implore the generation of icons for the ubiquitous PPT. 

Angels taking walks with wild boars ...

These are all cherry-picked of course. But the point remains that even this small open source model can create quite amazing pictures.