Lab

Experiments with Generative AI

July 29, 2022

The future looks bright, perhaps thanks to the three Suns?

These past few days, we’ve been tinkering with two AI engines.

DALL·E 2 is an artificial intelligence system that can create realistic images and art from a description in natural language. It is part of a larger OpenAI set of models.

Midjourney is a similar AI program that creates images from textual descriptions. The team and lab are independent and self-funded, focusing on design, human infrastructure, and artificial intelligence.

The experiments

We experimented to explore the capabilities, limitations, differences, bias and real-world potential uses for both these incredible tools.

At first, we started with common and well-understood general tasks to calibrate our own “machine speak”, then we moved closer to home with more local context challenges – places, things & topics more relevant to our part of the world.

A little fun to start us off…

A dung beetle on top of a woven yellow mat.

A pig doing cartwheels on the beach… (what else would they be doing there?)

Midjourney upscale of our favourite pig doing cartwheels on the beach.

A pig riding a bicycle on the rings of Saturn jousting against a lion.. because who uses words like “jousting”!?

Let’s start presenting challenges closer to East Africa

A fruit and vegetable produce market stand in an African city (yes, we went very generic and fluffy with the geography).

Midjourney upscale of a fruit and vegetable produce market stand in an African city.

A chapati on a blue plastic plate with floral patterns next to a steaming hot cup of tea, on a table with a red and white checkered tablecloth. Mind blown by OpenAI’s DALL·E 2! Chai na chapo anyone?

A highly detailed picture of a clay pot with an African pattern… (despite there being no such thing as an “African pattern”).

Midjourney upscale of the clay pot texture and lighting.

Safari rally car racing in a cloud of dust in Kenya.

Different version upscales of the cars. Something very wrong must have happened to Version 2 near that tricky corner at Elementaita.

People and faces

A Maasai holding a spear jumping high above earth in the stratosphere. Remember watching White Men Can’t Jump? Really? You’re old… that came out 30 years ago. Be strong.

A Maasai warrior enjoying a cold beer with a Viking in a rustic old pub.
(they obviously became best buds and exchanged clothing at some point).

Dialling up an artistic render component

An impressionist painting of Luanda Magere, oil on canvas.
DALL·E 2 must have picked Luanda (capital of Angola) but Midjourney was kind enough to paint a face there as well. Dope.

Bananas in Kampala, painted by Van Gogh.

Bananas in Kampala, painted by Pablo Picasso.

Exploring DALL·E 2’s ability to extend a version of Pablo’s initial masterpieces – because the client asked for changes once again.

And finally, some spaces and landscapes

Sunrise at the Maasai Mara, with lone acacia tree and hot air balloon.

Futuristic African city (somewhere near Timbuktu we suppose).
Did you think Timbuktu was an imaginary place? Google it then come back, we’ll wait 🙂

Snowing in Nairobi. This render was sponsored by today’s weather.

Tea plantation (now just imagine how much colder it is up there).

Midjourney upscaling the tea plantations.

Results

DALL·E 2

We soon realised that DALL·E 2 often had wider knowledge of unique datasets and “understood” natural language and the intent behind our queries better, presenting more expected and accurate objects, places and hyper-realistic faces and forms.

Midjourney

Midjourney on the other hand wouldn’t always understand certain terms and themes local to East Africa, but its results were more imaginative and its image processing power was phenomenal at quickly producing extremely polished spatial forms, landscapes and artistic lighting environments.

What’s next?

In Part II of this series, we’ll expand on the generative processes behind these engines and discuss potential use-cases in our design industry and the creative economy.

Lab