ChatGPT Can Now Generate Images, Too

OpenAI has recently released an updated version of its DALL-E image generator to a select group of testers. Notably, this technology has been integrated into OpenAI's popular ChatGPT chatbot. What's particularly impressive about this new development is that ChatGPT can now generate highly detailed images.

By combining DALL-E 3 with ChatGPT, OpenAI is further establishing its chatbot as a central hub for generative artificial intelligence (AI). This powerful combination enables the chatbot to autonomously produce a wide range of media, including text, images, sounds, software, and more. Since gaining widespread attention last year, ChatGPT has spurred on a race among tech giants in Silicon Valley to push the boundaries of AI advancements.

In the same vein, Google recently unveiled a new version of its chatbot, Bard, which seamlessly integrates with several of the company's popular services like Gmail, YouTube, and Docs. Meanwhile, other image generators, such as Midjourney and Stable Diffusion, have also updated their models this summer.

While OpenAI has long provided ways to integrate its chatbot with various online services like Expedia, OpenTable, and Wikipedia, this marks the first time they have merged a chatbot with an image generator. DALL-E and ChatGPT were previously separate applications, but with the latest release, users can leverage ChatGPT to generate digital images simply by describing what they want to see. Alternatively, they can utilize descriptions generated by the chatbot itself, thereby automating and streamlining the process of creating graphics, art, and other media.

Demonstrating the capabilities of this integrated system, OpenAI researcher Gabriel Goh showcased how ChatGPT can now generate precise textual descriptions that translate into images. For instance, when provided with descriptions of a logo for a restaurant named Mountain Ramen, the chatbot rapidly generated several corresponding images.

According to Mr. Goh, the new version of DALL-E excels at producing images based on multi-paragraph descriptions, meticulously following detailed instructions. Nevertheless, similar to all image generators and AI systems, it is not without its potential mistakes.

OpenAI plans to refine DALL-E 3 further and will not make it publicly available until next month when it will be accessible through ChatGPT Plus, a subscription service priced at $20 per month.

Though image-generating technology holds immense potential, experts have cautioned about its misuse for spreading large-scale disinformation online. To address these concerns, OpenAI has implemented tools within DALL-E 3 to prevent problematic content, such as sexually explicit images and depictions of public figures. Additionally, efforts have been made to limit DALL-E's ability to mimic specific artistic styles.

In recent months, AI has been increasingly utilized as a source of visual misinformation. Instances like a synthetic and relatively unsophisticated impersonation of an explosion at the Pentagon, which briefly affected the stock market in May, have underscored the potential risks. There are also concerns among voting experts that this technology could be maliciously employed during major elections.

OpenAI researcher Sandhini Agarwal, who specializes in safety and policy, explained that DALL-E 3 tends to generate images with a stylized rather than photorealistic quality. However, she acknowledged that the model could create convincing scenes, including grainy images captured by security cameras.

Regarding potentially problematic content, OpenAI does not intend to block it outright. Ms. Agarwal considers a blanket approach to be too broad, as the nature and context of images determine whether they are innocuous or harmful.

As she aptly stated, "It really depends on where it's being used, how people are talking about it."