When the AI ​​is wrong and creates an unlikely art gallery

The summer of 2022 brought to light, in no time, a host of new tools for generating images using artificial intelligence. By a simple text entry, everyone can express what they want visually, using often free tools and very advanced results.

The trend was such that image download platforms like Shutterstock had to work on the issue of licenses and usage rights. Today, Dall-E2, Stable Diffusion and Midjourney are the best known AI image generators, but Facebook, Google and Microsoft have also entered the market.

To generate an image, all you have to do is describe what you want to see. Then the software offers multiple tracks of images and once you choose one, the AI ​​can continue its work and polish the details. A work much faster than that of a human but who can, like him, make mistakes.

On a specialized Discord discussion, these improbable images are all listed and the results are very funny. Called “failed-diffusions”, it interested blogger Andy Baio (ex-Kickstarter) on Twitter, who shares several of the nuggets.

The images selected are selected for their grotesque errors, which for once, a human would have trouble doing without wanting to do it on purpose. Faces, human bodies, animals… the AI ​​can have trouble remembering norms on the question of anatomy.

Mistakes are also numerous when it comes to merging two topics that have nothing to do. artificial intelligence may have trouble integrating them together, and the results show that it is a synthetic, computer-generated image.

Google’s choice

While Dall-E2 offered particularly good renderings in terms of resolution, chromatic range and nuances in the lights, a small team at Google was developing DreamFusion on a completely different spectrum. The tool offered particularly basic images. But they were already in 3D, which means that it is possible to rotate the image and see a subject from all angles.

The tool also sought to avoid any errors, such as the one we were able to show you above. It was from there that Google sought to increase the quality. And the end result is called “Imagen”.

Of DreamFusion at Imagen, the difference is huge. At the beginning of October, we were able to discover the tool and its unfailingly realistic renderings, even with the most improbable image captions. For example, we have never seen such a realistic teddy bear swimming a 400 meter butterfly in the Olympics. Same thing with a dog house, essentially made up of lemurs.

© Image / Google

And you, what image would you like to generate from AI? At the house of lemon squeezer, we had recently tested the Midjourney tool and we had asked you to suggest images to be made on Twitter. Here are some of the results.

We wish to say thanks to the writer of this short article for this amazing content

When the AI ​​is wrong and creates an unlikely art gallery

Find here our social media profiles as well as the other related pageshttps://www.ai-magazine.com/related-pages/