As you already know, DALLE is an artificial intelligence program that uses descriptions in natural language to produce art and realistic visuals. It uses text and image pairing and is based on GPT-3, which has 175 billion parameters.
Recently, OpenAI published its beta version, which has significantly enhanced accuracy, realism, and resolution. Over the next several weeks, 1 million people will be invited off their waitlist. Invited users will receive 50 free credits in the first month, which can be used to complete one original DALL-E prompt and produce four photographs or to obtain an edited change tip, which produces three images. After that, invited users will continue to receive 15 free credits each month.
More than 3,000 artists from more than 118 countries around the world have taken part in testing the system since the preview version was released three months ago. These artists range from illustrators, AR designers, writers, landscape designers, tattoo artists, costume designers, and directors to illustrators, writers, and directors.
To inspire and hasten their creative processes, artists and creative professionals use DALLE, which enables users to produce swiftly and easily. DALLE has already been used to produce magazine covers, music videos for young cancer sufferers, and the realization of original ideas.
The new version also offers the following features:
- Users can utilize Edit to realistically and contextually alter photos they create using DALLE or upload by providing a natural language description.
- Variations can take a DALLE-generated or user-uploaded image and produce new versions inspired by the original.
- Users can preserve generations directly on the DALLE platform by using My Collection.
OpenAI team collaborated with researchers, artists, developers, and other users to learn about the dangers before releasing DALLE in beta. Based on their discussions, the team has taken the following actions to enhance the safety systems:
- Preventing abuse: Their system disallows image uploads with realistic faces and attempts to imitate famous people and influential politicians to reduce the possibility that DALLE will be exploited to create misleading information. They also employed cutting-edge methods to avoid the photorealistic generation of genuine people’s faces.
- Preventing Harmful Photos: To better prevent photographs that violate their content policy, they have improved the accuracy of content filters while maintaining the freedom of speech. This forbids users from creating violent, adult, or political content, among other categories. The most graphic material was taken from DALLE’s training data to limit its exposure to these ideas.
- Reducing Bias: To make DALLE’s generated images of people more representative of the diversity of the world’s population, they have applied a novel technique. When provided with a prompt that does not specify race or gender, such as “CEO,” this method is used at the system level.
- Monitoring: To prevent abuse, we will keep using automated and human monitoring techniques.
Please Don't Forget To Join Our ML Subreddit
Tanushree Shenwai is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Bhubaneswar. She is a Data Science enthusiast and has a keen interest in the scope of application of artificial intelligence in various fields. She is passionate about exploring the new advancements in technologies and their real-life application.