The Rise of Generic AI Art: Why Older Models Still Spark the Imagination
As AI image generation evolves, a surprising trend emerges: newer models, while technically superior, are producing art that lacks the soul and evocative power of their predecessors.
The initial arrival of tools like DALL-E sparked a sense of wonder. “It took me a couple of weeks to pick my jaw up from the floor,” one early adopter recalled, describing the experience as “magical, miraculous,” akin to discovering a new universe. The ability to conjure images from text prompts fueled a creative explosion, with users eagerly anticipating each new generation of output.
However, a recent comparison between older and state-of-the-art models reveals a concerning shift. A user testing the latest Nano Banana Pro (NBP) against Midjourney v2 found that while the newer model excels at technical precision, it falls short in artistic expression. The core issue? Newer models are “infinitely worse at making art.”
The Loss of Imperfection
The comparison highlights this divergence through a series of side-by-side examples. When prompted to create “Electron contours in the style of Italian futurism, oil on canvas, 1922, trending on ArtStation,” Midjourney v2 produced an image that, while not a perfect representation of futurism, possessed a compelling vibrancy. NBP’s output, in contrast, was described as “studiously in the style of Italian futurism, but the colours are so muted and dull.”
This pattern repeated across various prompts. A request for a “Painting of an alley in the Kowloon Walled City, Eugène Boudin, 1895” yielded a beautiful, if inaccurate, impressionistic scene from Midjourney v2. The NBP version, however, appeared overly desaturated and lifeless. Even attempts to force more evocative qualities – specifying “coarse, impressionistic, vague, evocative, contradictory, brimming with mystery” – failed to rescue the image from its drabness.
Literal Interpretation vs. Artistic License
The problem extends beyond color palettes. NBP consistently interprets prompts literally, stripping away the intended nuance and ambiguity. When asked to depict “Attar and Ferdowsi in a dream garden, Persian miniature, circa 1300, from the British Museum,” Midjourney v2 offered an evocative, dreamlike image. NBP, however, rendered what was described as “a photograph of a Persian miniature in the British Museum,” missing the point of the prompt’s evocative intent.
Similarly, a prompt for “The Burning of Merv by John William Waterhouse” resulted in a visually stunning, albeit semantically debatable, image from Midjourney v2, brimming with detail and atmosphere. NBP’s rendition lacked the same artistic flair and historical accuracy. Even a request for a portrait of Ada Lovelace by Dante Gabriel Rossetti saw NBP produce a photograph of a painting at an auction, rather than an image in the style of Rossetti. “Pure philistinism,” one observer quipped.
The Power of Vagueness
The key difference, it seems, lies in the ability of older models to embrace imperfection. These models, with their inherent limitations, produced images that were “vague, mistakes, and contradictions,” creating spaces for the viewer’s imagination to fill in the gaps. The resulting images were not fixed entities but rather invitations to explore infinite possibilities.
Newer, high-resolution models, focused on precision, struggle with this ambiguity. They excel at creating “specific, concrete things” but fail to capture the abstract, multi-faceted nature of true art. As one user poignantly stated, “What are we doing here?”
The call to action is clear: “We need to make AI art weird again.” The future of AI-generated art may depend on embracing the imperfections that once made it so captivating.
