According to artificial intelligence software, it looks like a “snail made of tomatoes.”

These polished images, a bit surreal or bizarre, are the brainchild of the American company OpenAI’s artificial intelligence. They are made from brief text instruction: “Garlic cat”, “Peach shaped armored chair”, “Drawing baby panda driving a car”

DALL.E, the name of the program (openai.com/blog/dall-e), is not primarily used to create credible synthetic images, but this function is performed by processing the text into images. The company is already known for its advances in strategy Tota2 Or automatic text creation, which has not yet provided much detail about its new record. The neural network it uses has more than 12 billion parameters, compared to 175 billion for GPT-3, according to its text-generating program DALL.E. These parameters are “learned” thanks to the pair pictures and texts. The second method allows you to categorize the 512 outputs that are automatically generated for each text request to keep the best text only 32