Artificial intelligence research lab OpenAI released DALL-E 2 in April, the successor to DALL-E due in 2021. Both AI systems can generate stunning images from natural language text descriptions; they can make images that look like photos, illustrations, paintings, animations, and basically any artistic style you can express in words. DALL-E 2 has many improvements : better resolution, faster processing and an editor function that allows users to modify the resulting image using only text commands, such as “replace vase with plants” or “let The dog’s nose gets bigger”. Users can also upload their own images and then tell the AI system how to adjust them. The world’s initial reaction to DALL-E 2 was amazement and delight. Any object and creature can be combined in seconds; any art style can be imitated; any location can be depicted; and any lighting condition can be depicted. Who wouldn’t be impressed by an image of a Picasso-style parrot flipping a pancake, for example? But concerns arise when people ponder which industries are likely to be disrupted by this technology.
OpenAI has yet to release the technology to the public, commercial entities, or even the entire AI community. OpenAI researcher Mark Chen told IEEE Spectrum: “We share the same concerns as people about abuse, and that’s something we take very seriously.” The company invited some people to try DALL-E 2, allowing them to share their results with the world . The policy of limited public beta contrasts sharply with that of Google, which just released its own text-to-image generator Imagen . When releasing the system, Google announced that it would not release code or public demos because of the risk of misuse and the generation of harmful images. Google released some pretty impressive images, but didn’t show the world anything that was in question as it implied.
This article is reprinted from: https://www.solidot.org/story?sid=72139
This site is for inclusion only, and the copyright belongs to the original author.