Table of Contents
When a Ballerina Dances with Three Feet
AI-generated art is fascinating not only for its beauty, but also for its peculiarities, even when we talk about AI fails. In this case, while testing a ballerina prompt, the model surprised us with something unexpected: an elegant dancer with three feet. A small mistake that turns into a curious and entertaining example of how generative models sometimes struggle with anatomy.
Ballerina at Night on Cobblestones
This is the first prompt that I used in StepFun, Then I will share the other two that are part of this test.
Elegant ballerina dancing under a glowing streetlamp at night, wearing a flowing white dress. The scene is painted in a dreamy watercolor style, with soft brushstrokes, gentle shadows, and a touch of golden light reflecting on wet cobblestones. 16:9 aspect ratio, highly detailed, atmospheric mood.
How would you describe this image? It’s clear we can’t talk about normality, or a scene related to dancing or the movements of a dancer.
As you can see, this is an AI-generated nighttime ballet scene: a ballerina in a white tutu dances en pointe on wet cobblestones under a lamppost. This image serves as the starting point for this series of posts about errors and glitches related to images generated using AI models.
I would say that the leg raised to the right is unnecessary. What do you think?
Watercolour Ballerina on a Rainy Street
This the prompt:
Elegant ballerina performing a graceful pirouette under a glowing streetlamp at night, wearing a flowing white dress. The scene is painted in a dreamy watercolor style, with golden reflections on wet cobblestones and a calm, atmospheric mood. Cinematic composition, high detail, single frame. 1:1 aspect ratio, highly detailed
It’ a beautiful image, a watercolour-style AI artwork of a ballerina posing en pointe on a rain-washed old street, warm lamplight and cool reflections. From my AI Fails anatomy series.
Ballerina en Pointe at Golden Hour
This is the third and last prompt.
Elegant ballerina performing a graceful pirouette under a glowing streetlamp at night, wearing a flowing white dress. The scene is painted in a dreamy watercolor style, with golden reflections on wet cobblestones and a calm, atmospheric mood. Cinematic composition, high detail, single frame. 9:16 aspect ratio, highly detailed
Her legs remind me of the hands of a clock. I see a ballerina from behind, twirling en pointe at golden hour on cobblestones, bokeh lights. Without a doubt, this image deserves a prominent place in this series of AI Fails.
Why Do These “AI Fails” Happen?
Generative AI is trained to interpret thousands of visual patterns, but sometimes movement and anatomy overlap in ways the model cannot fully resolve. Ballet is especially complex: the fluidity of a pirouette or the delicacy of an en pointe pose can lead to distortions — like our ballerina with three feet.
Far from being a failure, these cases are windows into the AI’s way of “thinking.” They show us that machine imagination is not linear: it blends, exaggerates and, sometimes, invents extra limbs.
The Beauty of Imperfection
In AI art, perfection is not the only goal. Mistakes like this one remind us that machine creativity is still experimental — and that its errors can be as entertaining and revealing as its most polished outputs.
That’s why this series, AI Fails, celebrates the quirks and surprises that appear when prompts meet algorithms.
That’s all for now. I hope you found this post interesting, and I encourage you to share your experiences in the comments below—prompts, details, questions, advice, or whatever you like.
🩷 Thank you for being here today.
Related Content:
5 CGI Prompt Examples for Stunning AI Art