The impossible statue: a work of art created by artificial intelligence

“The Impossible Statue” combines the styles of five sculptors who did not live at the same time.

Can you imagine a sculpture that was made by five artists who lived at different times in history, 500 years apart? Now, The Ai Framework, an AI company, and Sandvik, a high-tech engineering industrial group, have produced a sculpture that, although not created by the sculptors, does combine their signature styles.
“The impossible statue”, which is the name of the creation, is a sculpture 1.5 meters high and weighing 500 kilos, made entirely of stainless steel. The spectacular statue fuses the styles of five sculpture greats: Michelangelo, Auguste Rodin, Käthe Kollwitz, Takamura Kotaro and Augusta Savage.

The statue represents a bare-chested man holding a world globe with his right hand and with his left hand covering his legs with a cloth.
“The idea of creating a real-world statue designed by AI is quite novel. Instead of designing an AI system from scratch that went from concept to statue, we decided to use many AI systems so we could continually iterate and improve what came out. We know that AI can produce very visually attractive images, but that doesn't mean they can work in real life," says Robert Luciani, software engineer at The AI Framework, the company that has been in charge of developing the AI models, in a video that Sandvik has shared on its YouTube channel.

By creating the statue, the Swedish business group, Sandvik, wanted to demonstrate what it is capable of doing in CNC (computer numerical control) machining. This way of manufacturing parts is subtractive, that is, layers of material are removed from a raw part to generate a customized one. For this, computer controls and special machines are used.

“The Impossible Statue” was designed by training several artificial intelligence models and combining the most characteristic attributes of five sculptors: the dynamic poses of Michelangelo, the way of working the muscles of Rodin, the somber tones of Kollwitz, the Japanese influence of Takamura Kotaro and the defiant style of Augusta Savage.

Those responsible for the design used 2D tools. They used Stable Diffusion, they trained a model of that AI so that the different images that were generated had more coherence between them. They also used Midjourney and DALL-E, with the idea that they would record the styles and poses of the different artists. The result was fed into the Stable Diffusion model and the final 2D image was produced.
DALL-E is an AI whose strength is creating images from textual descriptions. It belongs to OpenAI, the owner of ChatGPT. Midjourney also specializes in generating images from text and is very good at interpreting artistic style and more subjective aspects such as emotions and atmosphere. Stable Difussion stands out for its ability to produce high-quality digital images from natural language.

The process was not completed in the blink of an eye since they had to do many tests generating images until they found the final one, which met the requirements and could be manufactured with Sandvik machines.

Another great difficulty was converting a two-dimensional model into a three-dimensional one, for which different AI tools were used, such as the Midas depth estimator, which, based on a flat image, estimates what it would be like if it had three dimensions.

At first, Sandvik planned to make the statue in a single piece, but the size of the sculpture made it impossible and, finally, the company decided to section the model into 17 components that had to be adapted to CNC machines.

Currently “the impossible statue” is on display at the Stockholm Museum of Science and Technology, in a permanent exhibition on artificial intelligence.

Source