Stability AI launches Stable Diffusion XL 1.0, its latest image generator; Know all about it
Stability AI has released its new stable version of its text-to-image generation model called Stable Diffusion XL 1.0. Check here to know more.
With each day passing, we are seeing more generative AI platforms being introduced. As these tools are made easily accessible the competition within the market has grown with competitors like Midjourney and Adobe. Keeping the ever-growing user demand in mind, Stability AI has launched its improved version of the image generation model, Stable Diffusion XL 1.0. The company claims that the current model is its most advanced offering in the segment. However, there are some ethical issues still persistent in the model.
According to a Tech Crunch report, Stability AI launched a text-to-image model called Stable Diffusion XL 1.0, which it claimed to be the “most advanced” tool released now. The platform is available in open source in GitHub alongside Stability's API and consumer apps, ClipDrop and DreamStudio.
About Stable Diffusion XL 1.0
The latest stable diffusion model has implemented a number of upgrades to enhance its functionality. Let's take a look at what's new in the latest version. As per the company, the latest AI image generator can deliver “more vibrant” and “accurate” colours and better contrast, shadows and lighting as compared to the previous version.
Joe Penna, Stability AI's head of applied machine learning told TechCrunch that the new AI platform has 3.5 billion parameters. For the unaware, parameters are factors added in a model to contextualize the training data. With a high number of parameters, Stable Diffusion XL 1.0 can create full 1 MP resolution images in seconds across various aspect ratios. Penna also added that the model is being made easy to use and users can design complex images with simple text prompts as it supports natural language processing.
According to SiliconAngle and VentureBeat reports, Stable Diffusion XL 1.0 also enable users to reconstruct missing parts of the images and extend the existing image. Additionally, it allows users to submit an image and incorporate text prompts for making more sophisticated variants of the picture. The Stable Diffusion XL 1.0 also does not require long text prompts as it understands short inputs.
With an upcoming beta release of its API, Stability AI will allow users to "specialize" generation with as few as five images by targeting specific individuals or products. Furthermore, with its collaboration with Amazon Web Services, Stable Diffusion XL 1.0 will also be available on Bedrock, Amazon's cloud platform.
The ethical problem
There are still some ethical questions around the AI tool. As the platform is offered open source, issues like nonconsensual deepfakes, misinformation images, and images showcasing harmful content cannot be ruled out. While the spokesperson told TechCrunch that the company filtered the model to block "unsafe imagery" and issuing warnings for harmful content, it is unlikely to be a full-proof solution given similar issues have existed in the company's other models, specifically DreamStudio.
Apart from that, there are also questions on the source of the training data for the model. Stability AI claims that it used images around the web, which includes artwork from those artists who have protested against tech companies for using their work to train AI. Even as the company maintains that it is protected from legal action due to fair usage doctrine, major players such as Getty Images have filed lawsuits to put a stop to it.
Stability AI, to adhere to ethics, has given artists an 'opt-out' option, so if an artist wants, they can take away their images from training the model. However, the company has not removed all flagged artwork from its training model.
“We are constantly improving the safety functionality of Stable Diffusion and are serious about continuing to iterate on these measures. Moreover, we are committed to respecting artists' requests to be removed from training data sets," Penna told TechCrunch.