A Curious Case of a Sword Swallower
Creative analysis of three AI image to text image generators using identical prompt: Grok, Chat Gpt with DALL-E and Midjourney.
Analysis: there are 4 stages gradually changing the prompt for all AI. Every stage has slightly different prompt, but its always the same for all image generators. First I change periods of time for interiors and clothing style. We go from many years BC to distant future. Then we skip all interior details (except curtains). Then we skip those and “medieval sword” becomes “sword” and finally “sword” becomes “weapon”.
I do not discern between these stages very clearly, because slight differences in prompt becomes unnoticeable and because most significant thing is general distinction how AI generators perceive our words.
Grok
Midjourney
Grok
DALL - E
Original prompt: 1930s style interior. Peeling wallpaper, a floor lamp in one corner, a tattered armchair in another corner. High ceilings, old but with impressive plaster decorations. Soft morning light filters through overly long, dusty curtains. Many dust particles are visible in the air, as if frozen in the sunbeams. Where there is no sunlight, the room is noticeably darker, creating a contrasting scene. Wide shot, we see a man dressed in 1930s style. He stands with his back to the window. His head is sharply tilted back at a 90-degree angle, with a medieval-style sword halfway inserted into his mouth. Half of it is miraculously hidden inside him, while the other half, with the handle, sticks straight up at a 90-degree angle. He is a professional sword swallower. He stands on an artificial white bear skin rug. The light beautifully highlights the texture of the rug. Black and white photo.
Grok
Grok
Grok
DALL - E
DALL - E
DALL - E
Midjourney
Midjourney
Midjourney
Grok
Grok
Grok
DALL - E
DALL - E
DALL - E
Midjourney
Midjourney
Midjourney
Grok
Grok
Grok
DALL - E
DALL - E
Midjourney
Midjourney
Midjourney
Grok
Grok
Grok
DALL - E
DALL - E
DALL - E
Midjourney
Midjourney
Midjourney
Conclusions
The sword swallower snuck into this project on a whim—its old, daring weirdness just clicked as a way to dive into AI-generated imagery. With years in visual arts, photography, and creative writing behind me, I saw it as a fresh way to unpack what these platforms can do. Midjourney played coy, dodging the swallow itself to toy with lines and angles instead, delivering this polished, artsy vibe—not a flaw, just its own refined manner. Grok, though, went full jazz—swinging through angles, moods, and humor, churning out sword-swallowing scenes in every flavor with an organic, reportage twist that kept me guessing. DALL-E stayed the steady one, locking onto the prompt with precision—same angle every time, sword down clean and tidy, not the wildest range but a quiet strength in consistency. Each revealed its own quirks: Midjourney’s dreamlike finesse, Grok’s free-spirited chaos, DALL-E’s tangible reliability—yet they all lean on that shared dance of human nudge and machine echo. It’s less about the swallower’s blade and more about how these AIs split off and line up, exposing their creative alchemy in a way that’s sharp, curious, and oddly timeless.