The Era of Limitless Creativity is Upon Us. So What?

I had the luxury of playing with a few AI tools over the last few days. And the capabilities and outputs are nothing short of amazing. While we are only scratching the surface of what is possible using AI, we are also only scratching the surface of the harms that are possible too. This article is meant to be a joyous celebration of the limitless creativity that is now possible, because these tools of production are finally available to the masses. I will write a separate harms and dangers article another time. But I will share a few observations.

The following was created in Midjourney v5 with a simple prompt: "photorealistic still life of spectacularly colorful fruits, flowers and leaves arranged very neatly in a vertical clear glass vase on an antique wooden table, 35mm film, natural lighting from the left side."

You can use chatGPT to generate a more verbose prompt, by asking it to embellish the simple prompt above "photorealistic still life of spectacularly colorful fruits, flowers and leaves arranged very neatly in a vertical clear glass vase on an antique wooden table, 35mm film, natural lighting from the left side"

ChatGPT: "a stunning photorealistic still life that captures the beauty of a colorful array of fruits, flowers, and leaves. The arrangement is placed neatly in a clear glass vase that stands tall on top of an antique wooden table. The fruits are vividly portrayed in their natural shades, from the bright yellow of lemons to the deep red of apples. The flowers boast a range of colors, from the soft pink of roses to the vibrant orange of marigolds, and their delicate petals are expertly captured. The leaves provide an elegant backdrop to the arrangement, with their varying shades of green and intricate details. The scene is captured on 35mm film, giving it a vintage aesthetic. The natural lighting from the left side creates a beautiful interplay of light and shadow, highlighting the textures and depth of the composition. Overall, this still life is a true masterpiece that showcases the beauty of nature through the art of photography."

Movie poster concepts made in seconds

Mock runway fashion shows

Oil paintings in the style of 17th century Dutch masters

Convincing photos of non-existent French Castles surrounded by vineyards

Billionaire's penthouse, midtown Manhattan, looking south

Irises in the style of a botanical book

Ferrari super e-car

Having gawked at the incredible examples above, you may have noticed the misplaced hand on the first image of the set of four with the blond woman in dark navy blouse. Some errors and lack of accuracy start to become obvious. For example, the prompt for the set of four uniform shirts was "all white shirt design with small rectangular German flag on the front, very modern style for sports team uniform." Where's the "small rectangular German flag on the front?" From this example, I realized that AI generators like Midjourney are just remixing photos and images that were in its training data. Regardless of how vast that training data set is, it's still not executing exactly what you told it to do (like you would do yourself in Photoshop or Illustrator). It is remixing things to come as close as possible to what you described in the prompt -- like the following hilarious example: "salmon swimming in a river." The AI returned exactly what the prompt asked for.

Even the image above, generated a minute ago in Midjourney v5 still can't get the fingers right, right? It takes a human 1 second to recognize the error.

To me, this means that anything that requires accuracy, precision, and "the truth" should not be done or generated by AI. For example, code. While AI is helpful in building base code, for more elaborate code, the algorithms can accidentally mix in bad or malicious code, and the human using AI to generate the code won't realize that had happened. The AI is basically remixing code that it had access to. And bad guys can even "poison" code by making libraries of malicious code publicly available so it could be accidentally ingested into training data.

One other observation is that AI is great for "divergent" thinking -- i.e. creativity. It is very helpful at suggesting ideas and making a variety of concepts for consideration. But it is very poor at "convergent" thinking or doing -- i.e. arriving at a single, correct answer. This is best illustrated with the 2 examples below. The prompt on the left asked it to generate a silver metallic pin with the letters "FA" on it. The example on the right asked for text "FLP" on the logo. The AI simply could not do it, and do it reproducibly. This could have been done in Photoshop easily. Again, the point here is to know the limitations and boundary conditions of AI, as magical as it may seem at first glance.

But lest I digress too far into the negative territory, let me get back to the amazing creativity that is now possible and in the hands of anyone.

In the following examples, I used the following: 1) photo generated in Midjourney 5, 2) voice with British accent synthesized by elevenlabs.io, and 3) combined into animation using d-id.com. What I learned is that the AI animation works better on faces that are facing front, towards the camera. Faces that are facing off to the side, or faces of real humans (like myself) uploaded, do not work well. They were so scary bad, I am not even going to show them here. Just the good examples below.

Let me re-iterate, AI should be used with caution for anything that requires accuracy, precision, and "the truth." Know that bad actors will be taking advantage of the available tools in creative and nefarious ways. Humans can also use these tools to thrive in the era of limitless creativity which is truly upon us already. Humans' role is evolving, fast. Humans need to be curators of what is right, accurate, and truthful. Humans can also use these tools of production to curate and create content that is verified and valid. Otherwise, there will simply be too much content of all formats (text, images, video) for any human to know what is real or not. See the example below -- Midjourney in 2022 versus 2023. The photo on the right side didn't happen. It was made by Midjourney.

What is clear to me is that I will never have to use a stock photo again for my LinkedIn article mastheads. I can just generate what I want with Midjourney in 60 seconds. It may not be exactly what I asked for, but if it's 90% of the way there, it's still way better than premade stock photos. One final note is that you cannot reproducibly make the same images with the same prompt. I used the exact same prompt a day after and got entirely different results. Even the same prompt used a few minutes apart resulted in entirely different images. The prompt for both set of images below was: "beach landscape oil painting, dynamic lighting, oil painting."