The Surreal Prompt: How Text-to-Image Artists Craft Visual Poetry

Why the most meaningful AI art starts with words that dance between description and dream

I’m writing this on a windy afternoon, the kind where the sky can’t decide between sunshine and drizzle. It feels fitting, because text-to-image art thrives in those in-between spaces — half certain, half dreaming.

Artists who use these tools know the odd thrill that comes when a phrase, typed almost absent-mindedly, blooms into something visually stunning. I still remember the first time it happened to me. I had typed a prompt about “a lantern carrying its own sorrow” after a long day, and the system responded with a scene so gentle and eerie that I had to stop and breathe for a moment. I wasn’t sure whether to be delighted or slightly alarmed. Maybe both.

At its best, this craft turns language into pigments and probabilities into poetry. And somewhere in that process, the prompt becomes more than an instruction — it becomes a creative act in its own right.

Beyond Description: The Art of the Poetic Prompt

Most beginners start with plain prompts. “A forest with soft light,” for instance. It’s serviceable, yes, but it rarely surprises. Experienced artists shift toward something more evocative — suggestion instead of description. A phrase like “a city remembering its future visitors” does not describe a literal place. Yet these systems answer with architecture tinged with longing, as if the city itself were waiting.

What makes metaphor so potent here? Models lean on patterns they’ve seen, and metaphors — those curious blends of imagination and association — allow them to traverse unusual combinations. Prompting becomes a way of whispering to the model rather than commanding it.

One artist told me they once entered “the lullaby of broken satellites” with no clear expectation. The outcome, a cradle of faint blue light wrapped in scattered cosmic debris, remains one of the most haunting images in their collection. They said it made them rethink what “control” even means in digital art.

It raises a quiet question in the background: If symbolism can bend machinery into making something emotionally resonant, what exactly is happening between those words and the pixels?

The Dramatic Evolution: From Clumsy to Sublime

Looking back at the early versions of text-to-image tools feels a bit like flipping through childhood sketches. Endearing, but full of strange limbs, mismatched eyes, and unintentional surrealism.

Now, the leap is obvious. Between 2022 and 2024, model training improved so quickly that artists could practically feel the ground shifting under them. Larger datasets expanded visual comprehension; diffusion methods sharpened edges and deepened color palettes. Even the vocabulary models could understand broadened. Terms like “kinetic geometry,” “neon melancholy,” or “Baroque ornamentation” began producing images far closer to what artists imagined.

Still, there’s a trace of unpredictability that lingers — and perhaps it should. People sometimes complain when a model doesn’t obey their prompt perfectly. Yet perfect obedience isn’t always the goal. Art gets less interesting when every line behaves.

Anyway… I sometimes miss those early happy accidents. They taught us to embrace a bit of chaos.

Crafting Prompts Like Literature: Technique and Strategy

Some artists now treat prompt-writing the way poets treat line breaks. Rhythm matters. Repetition matters. Even a slightly awkward phrase — left intentionally unpolished — can nudge a model in strange and rewarding directions.

A prompt like “silver on stone, silver on smoke, silver on silence” tends to create visual echoes that feel musical, almost as if the model hears the rhythm. Writers might call it anaphora; prompt artists simply notice that it works.

Balancing precision with room for interpretation is one of the real arts here. Too many details, and the model stiffens; too few, and it wanders. I like to think of it as tuning a lens: wide enough for mystery, narrow enough for shape.

Juxtaposition can do wonders. “Delicate machinery” or “a fragile monolith” often generate compositions charged with tension. Paradox deepens that effect. And synesthesia — asking for “the taste of cold brass” or “the sound of red glass cracking” — invites models to search for cross-sensory associations hidden in their training memories.

Should there be rules for prompting? I’m not convinced. Some days the work feels like literature; other days it’s more like trying to talk a stubborn cat into coming inside.

When Surreal Becomes Deeply Meaningful

Surreal prompts tap into shared symbolic reservoirs. Even when we’re not trying to reference mythology or psychology, the model’s vast exposure to imagery means it often picks up on archetypes. A phrase like “a doorway dreaming of its visitors” tends to yield scenes infused with transition and yearning — interpretations viewers bring their own meaning to.

I once walked through an exhibition where every piece originated from ambiguous, introspective prompts. People lingered longer than usual. One viewer stood in front of an image of a translucent staircase dissolving into mist and said quietly, “This feels like how last year went for me.” The artist later admitted they had typed the prompt in a moment of confusion and self-doubt.

There’s something quietly democratic about this. You don’t need a traditional art education to create work that resonates. You don’t need perfect skill — only curiosity, maybe a little emotional honesty, and the willingness to ask the model to meet you halfway.

The world could use more shared metaphors, even the strange ones.

The Prompt as Artwork: Preserving Creative Intent

As prompts become more literary, artists have begun asking whether they should be displayed beside the images. After all, the text is part of the creative lineage. Without it, viewers miss the origin story — the spark that gave the image shape.

Intellectual property debates inevitably enter the room. If a prompt carries creative weight, who owns what? And how do we honor the fact that the model’s training data shapes the final output just as much as the user’s intention? I don’t have tidy answers. Actually, no — that’s not entirely true. I suspect we’re still inventing the norms for this medium, slowly and collectively.

Many artists now keep a “prompt history” as part of their portfolios. It reads like a diary sometimes, full of uncertain lines, playful experiments, half-formed ideas. Looking back at mine, I see prompts written in late nights, prompts written between emails, and prompts written while waiting for a kettle to boil.

Oddly enough, those snippets of text tell as much of my artistic journey as the images do.

I wish I had a perfect conclusion — something wise about how language and math have found each other in this strange century. But the truth is, I’m still figuring out what prompts can do — and what they reveal about the imagination behind them.

Jean Marie Bonthous (publishing as JM Bonthous) is the author of two books on digital art and one on AI-created music

How to Thrive in the Digital Art Market: the Artist’s Guide to Online Platforms, Collectors, and Trends

The Digital Artist’s Guide to Success: How to Create, Promote, and Profit from Your Art

Mastering SUNO: The Ultimate Guide to AI Music Creation

(All available on Amazon)

He has also authored seven other books on the human side of artificial intelligence and six on nonfiction filmmaking. His writing also explores the intersections of AI, digital art, and nonfiction filmmaking. See his latest books: www.jmbonthous.com

He also blogs on MEDIUM, in a separate blog, about the human side of AI: https://medium.com/@jm_26203

Connect with him on MEDIUM and follow https://medium.com/@jmbonthous to stay in the loop with the latest stories about digital art and with @jmbonthous1 to stay in the loop with the latest stories about the human side of AI.