If the saying "a picture is worth a thousand words" is not true, this new artificial intelligence (AI) program from OpenAI will make the statement true.
OpenAI recently created a new version of its DALL-E text-to-image generation program that features higher-resolution and lower-latency images depicting descriptions written by users, called DALL-E 2, per The Verge.
OpenAI hopes that people will use DALL-E 2 to express themselves creatively. The non-profit organization also added that the AI program helps them understand how AI systems see and understand the world, which is "critical to [OpenAI's] mission of creating AI that benefits humanity."
Interested people will have to get on OpenAI's DALL-E 2 waitlist to access the AI program.
DALL-E 2 Details
According to The Verge's report, DALL-E 2 also features new capabilities like making realistic edits to an existing image from a natural language caption, aside from generating higher-resolution and lower-latency ones. Meanwhile, OpenAI said that DALL-E 2 could add and remove elements from an image while taking its shadows, reflections, and textures into consideration.
The Verge gives a deeper explanation. DALL-E 2 users start with an existing picture, select an area, and tell DALL-E 2 to edit it
OpenAI's DALL-E 2 website also shows that the AI program can take an image, such as a piece of historical artwork, and create different variations using the original photo as inspiration.
Read More : Google Docs To Get Emoji Reactions Support
OpenAI also added a process called "diffusion," which DALL-E 2 uses to generate an image from text. This process starts with a pattern of random dots that DALL-E 2 gradually alters until it creates an image when it recognizes specific aspects of one.
Because of the diffusion process, DALL-E 2 is preferred over its predecessor, DALL-E, for its caption matching and photorealism when evaluators were asked to compare a thousand image generations from each AI program. 71.7% of evaluators preferred the newer version for caption matching, while 88.8% chose DALL-E 2 for photorealism.
DALL-E 2 Limitations and Safety Mitigations
OpenAI recognizes that the DALL-E 2 AI program could be misused by irresponsible or abusive people. As such, the non-profit organization placed limitations on DALL-E 2. These limitations include DALL-E 2's inability to generate violent, hateful, or adult images through the removal of the most explicit content from the AI program's training data. OpenAI also used "advanced techniques" to prevent photorealistic generations of faces involving real-life people or public figures.
OpenAI also placed human monitoring systems to guard DALL-E 2 against misuse, further rendering the AI program to generate the excluded content.
The AI Program will also evaluate different approaches to handle potential copyright and trademark issues to filter out specific types of content and work directly with copyright/trademark owners on these issues, per OpenAI's DALL-E 2 preview page on GitHub.
Finally, OpenAI is also working with a limited number of external experts and trusted users to preview DALL-E 2 to help the non-profit organization learn more about the technology's capabilities and limitations.
