Reflections from playing with DALL-E 2

"Colossal burger and fries in the early evening, in the style of Simon St?lenhag", courtesy of DALL-E 2, a new AI system developed by Open AI that creates realistic images from a natural language description.

It may be a minor achievement in the grand scheme, but AI is officially a better artist than me in all respects, saving perhaps intention. Is the same true of you? This was my first conclusion after playing with DALL-E 2 together with the Uncertainty tribe on Wechat (shout out to Ian and Vasily for doing the heavy lifting).

I first looked into DALL-E after a client asked if I knew of any AI story-boarding software. It looks like we're one executive decision (use of DALL-E 2 is tightly controlled today) and perhaps a year or two of AI training from the time when some entrepreneur embeds DALL-E into a program to solve this challenge for marketing teams. A long tail of similar design challenges will follow. The permission question is critical since the sheer capability of DALL-E's image generation engine poses a host of ethical and social challenges.

I'll leave those challenges for another day and focus instead on the joy of human-AI collaboration from the perspective of a non-artist. Here are my six lessons in learning to love your AI partner. All images were created by an AI in near real-time based on a text prompt.

1. DALL-E likes to reference artists. AI may lack its own inspiration but it is a ready student of the masters.

Prompt: "Tall men racing camels on Venus in the digital future in the style of Hieronymus Bosch" - Strong effort here... I wouldn't know where to start.

2. Respect the capabilities and limits of your AI. There are AIs built for specification design. DALL-E is not one. But I'd give it an "A" for creative exploration.

Prompt: "Bosch handheld saw" - Use at your own risk...

3. DALL-E is a master of magical realism. Some future version of Instagram will enable you to wow your friends by posing in a sentient spaghetti dress.

Prompt: "Marilyn Monroe wearing a sentient spaghetti dress"

4. The order of the grammar is critical... but not always intuitive.

Prompt A: "Woman drinking a coke in a porsche" - Brand fail... the coke appears to be drinking the woman... (apologies for the nightmares)

Prompt B: "Coca Cola drinking woman in a porsche" - Success!

5. Do you feel the tension? In five years, Hollywood producers will use AI + VR to visualize ideas for plot twists and alternative perspectives in real-time while sipping wine on the beach.

Prompt: "CCTV video of Julius Ceasar's assassination"

6. DALL-E can get dark. There is a long list of forbidden words in DALL-E's library. Nonetheless, the AI captures mood well. Perhaps ingesting enough data is sufficient to emulate emotion. It intuitively understands our fears, Christian spirituality, and Sailor Moon.

Prompt A: "a photo of the last minute of my life" - Beware water and quiet places...

Prompt B: "a photo of the first minute of my afterlife" - I see a flood of UFO sightings on the horizon...

Prompt C: "a photo of the first minute of my next life" - DALL-E blesses us with reincarnation as Sailor Moon...?

Prompt D: "November 5 2024" - Ending on a dark note... why has DALL-E cursed November 5th? I have no clue. Is it channeling the Astroworld Festival tragedy? This is but one example of the many edge cases where more research is needed before DALL-E is released into the wild. Racism and sexism are embedded in the data (i.e., in our history). Deep fakes will become commonplace. People will explore their darker instincts. And unforeseen risks lurk beyond the horizon.

We are only scratching the surface of AI as a partner in design inspiration, concept visualization, and storyboarding. It is already a joy and a puzzle. Soon it will be a powerful tool. If you haven't yet, take a visit to https://openai.com/dall-e-2 and explore what else DALL-E has to offer.

Since you made it all the way to the end, let me leave you with a sneak peek into your family photo album, circa 2024.