OpenAI launched a public beta version of the API, making its image generation software DALL-E more widely available to enterprises. The API will make it easier for businesses to add DALL-E’s text-to-image capabilities to their products, providing developers with simplified tools to integrate and customize the software to their liking.
An early use case for the API is Microsoft’s Designer application, which uses DALL-E to generate images for Office users. Microsoft is one of the main investors in OpenAI and released the app last month.
Luke Miller, OpenAI’s product manager for APIs, said: “We already have some customers building on this in very interesting ways. Some are creative exploration, some are more business-oriented. A startup called Mixtiles is working on Use the API to generate posters and artwork for home decor, and another company called CALA is using it to help customers design their own clothing.”
Interest and adoption of text-to-image AI has exploded over the past year, and OpenAI, once a leader in the field, is now challenged by latecomers such as Midjourney and Stability AI, and they There are fewer restrictions on users, allowing users to build their own AI systems with little supervision. Meanwhile, other players in this space, such as Google and Meta, have taken a more cautious approach: developing systems with similar capabilities, but limiting their public use to very limited scenarios.
Despite the obvious creative benefits that text-to-image AI can provide, there are also multiple problems. The software can be used to generate misinformation and harmful images, and there are challenging ethical issues with the use of the data.
Text-to-image AI systems like DALL-E are trained on images scraped from the web, often including the copyrighted work of photographers, artists and designers. Many artists are outraged that the resulting technology can not only be used to emulate their personal style, but that they are not being compensated for using their work to generate revenue for a multi-billion-dollar company like OpenAI.
Some companies developing text-to-image applications have begun to provide compensation. Shutterstock, for example, which licenses their contributors’ data to OpenAI to create DALL-E and uses their API to generate custom stock images, recently announced that it is setting up a contributor fund to compensate people whose work is used to train AI. personal.
When asked if OpenAI plans to develop any similar program to compensate artists, Miller said there is nothing concrete to share right now and will continue to seek feedback from the community, saying it is a very complex issue that requires input from many different point of view.
For OpenAI, this question is really difficult to answer because the company has never shared the training data used to create DALL-E. In the United States, training AI models by scraping public images, even copyrighted ones, could be covered by fair use, legal experts say. But many artists point out that adequate legal protection does not equal moral approval.
Access to the DALL-E API will initially be rate-limited and will not be vetted on how customers use the technology, OpenAI said. Customers will be charged for the number of images generated and can choose between three resolution levels: $0.016 per 256 x 256 image; $0.018 per 512 x 512 image RMB); while a 1024 x 1024 image costs $0.02 (approximately RMB 0.146).
