Generative AI has made remarkable advancements in recent years, particularly in the realm of image creation. Reflecting on my initial experience with an AI image generator on January 29, 2022, I attempted to generate an image of a sabertooth cat. The outcome was an abstract mishmash of colors that bore little resemblance to the intended animal. Fast forward to today, I experimented with DALL-E 3, and the results were stunning, though not without their quirks—like the oddly proportioned fangs and unintended tiger stripes. The progress made is truly impressive, especially compared to my ongoing struggles with other platforms like Midjourney.

The evolution of generative AI has been swift, transitioning from vague, psychedelic representations to polished images across various styles, including photorealism. However, the effectiveness of these AI systems is hindered by their reliance on datasets. They "learn" by analyzing vast numbers of images, which works well for popular creatures like Tyrannosaurus rex and mammoths, but poses challenges for lesser-known species that lack sufficient online representation.

For instance, AI-generated illustrations of iconic extinct animals like T. rex, woolly mammoths, and Stegosaurus are often recognizable. While my artistic instincts may pick apart minor flaws, the overall quality exceeds what I could achieve without references. The representation of these creatures is vivid and engaging, bordering on "pop art" quality.

As of late 2023, DALL-E 3 outperforms Midjourney in depicting dinosaurs and other extinct species, showcasing significant differences in output. The limitations of AI become apparent when attempting to create images of more obscure prehistoric life forms, as humanity's understanding of past epochs is still evolving.

Despite providing additional context, such as classifying Edmontosaurus as a hadrosaur, the AI struggled to produce accurate results. The same was true for other ancient creatures like Gorgonopsids, where the outputs were bizarre and far from the mark. However, there were moments of success, like DALL-E 3's depiction of Sarcosuchus, which recognized its crocodilian nature despite inaccuracies.

Creating paleoart involves extensive research and analysis of fossils. As a dedicated hobbyist, I study fossils, often through photographs, to grasp their structure. This process combines knowledge of living organisms and behavioral inferences from fossil records. My artistic journey involves sketching and developing a three-dimensional mental model of the animal.

In my watercolor sketch of a Barbourofelid mother and cubs, I incorporated recent findings about their saber teeth and the developmental stages of their young. I also drew color inspiration from extant big cats, noting the camouflage benefits of spotted coats in cubs.

When illustrating a prehistoric creature within its habitat, artists must also consider the flora and fauna that coexisted with it. The goal is to create a representation that not only highlights significant discoveries but also evokes a sense of the ancient world. The final product is a blend of scientific accuracy and artistic vision, bringing to life what once was.

AI image generators, while capable of producing impressive visuals, were not designed for the intricacies of paleoart. They struggle with specificity and accuracy, particularly with lesser-known extinct species. The field of paleoart requires human creativity, inquiry, and expertise to capture the richness of ancient life accurately.

In an era of rapid advancements in artificial intelligence, the nuanced world of paleoart remains a domain where human skill excels. The possibilities of depicting life from epochs past are vast, and for now, machines cannot replicate the depth and accuracy that human artists bring to the table.