Facing increasing competition and ethical challenges, AI startup Stability AI continues to improve its generative AI model.
Today, Stability AI announced the release of Stable Diffusion XL 1.0, a text-to-image model that the company claims is the "most advanced" version to date. In addition to Stability's API and consumer applications, ClipDrop and DreamStudio, Stable Diffusion XL 1.0 is also available on GitHub as an open-source project. Compared to its predecessor, Stability claims that Stable Diffusion XL 1.0 offers "more vibrant" and "accurate" colors, as well as better contrast, shadows, and lighting effects.
During a media interview, Stability AI's Head of Machine Learning, Joe Penna, pointed out that Stable Diffusion XL 1.0 contains 3.5 billion parameters and can generate images with resolutions of up to one million pixels in multiple aspect ratios "within seconds." Parameters are the part of the model that the model learns from training data and essentially define the model's skill level in solving the image generation problem.
The previous generation Stable Diffusion model, Stable Diffusion XL 0.9, was also capable of generating high-resolution images but required more computational power.
"Stable Diffusion XL 1.0 is customizable and suitable for fine-tuning concepts and styles," said Penna. "It is also easier to use and can achieve complex designs with basic natural language processing prompts."
Stable Diffusion XL 1.0 has also made improvements in text generation. Penna stated that while many of the best text-to-image models struggle to generate images with clear signs, let alone calligraphy or fonts, Stable Diffusion XL 1.0 can achieve "advanced" text generation and readability.
Furthermore, Stable Diffusion XL 1.0 supports repair (reconstructing missing parts of an image), expansion (extending existing images), and "image-to-image" prompts, which means users can input an image and add some text prompts to create more detailed image variations. Additionally, the model can understand complex, multipart instructions with brief prompts, whereas the previous Stable Diffusion model required longer text prompts.
"We hope that by releasing this more powerful open-source model, not only will the resolution of images be significantly improved, but also more progress can be made for the benefit of all users," he added.
However, like previous versions of Stable Diffusion, this model has also raised thorny ethical issues.
In theory, the open-source version of Stable Diffusion XL 1.0 could be used by malicious actors to generate harmful content, such as unauthorized deepfake videos. This, to some extent, reflects the training data used, which consists of millions of images from the internet.
Numerous tutorials demonstrate how to use Stability AI's own tools, including DreamStudio, an open-source frontend for Stable Diffusion, to create deepfake videos. There are also countless tutorials showing how to fine-tune the basic Stable Diffusion model to generate explicit content.
Penna did not deny the possibility of misuse and acknowledged that the model also contains biases. However, he added that Stability AI has taken "additional steps" to reduce the generation of harmful content, such as filtering "unsafe" images from the training data, issuing new warnings when problematic prompts arise, and blocking individual problematic vocabulary as much as possible in the tools.
The training set of Stable Diffusion XL 1.0 also includes works by some artists who have protested against companies, including Stability AI, using their works as training data for generative AI models. Stability AI claims to be protected by the fair use principle under US law. However, this has not stopped some artists and stock image company Getty Images from filing lawsuits to prevent this practice.
Stability AI has partnered with startup Spawning to respect the artists' "opt-out" requests, stating that it has not removed all tagged artworks from its training dataset but continues to consider the artists' demands.
"We are continuously improving the safety features of Stable Diffusion and are committed to iterating on these measures," said Penna. "Furthermore, we respect the artists' requests to remove their works from the training dataset."
To coincide with the release of Stable Diffusion XL 1.0, Stability AI has also introduced a beta version of fine-tuning functionality in its API, allowing users to "specifically" generate images related to specific individuals, products, etc., using at least five images. The company has also brought Stable Diffusion XL 1.0 to Bedrock, Amazon's cloud platform for hosting generative AI models, expanding on its previous collaboration with AWS.
While seeking partners and new features, Stability AI has also faced commercial challenges, facing fierce competition from companies like OpenAI and Midjourney. According to Semafor, Stability AI has raised over $100 million in venture capital in April but has been burning through cash, leading to the closure of $25 million convertible notes in June and the search for executives to help boost sales.
"The latest SDXL model represents Stability AI's innovation capabilities, providing the AI community with the most advanced, openly accessible model," said Emad Mostaque, CEO of Stability AI, in a press release. "The release of 1.0 on Amazon Bedrock demonstrates our close collaboration with AWS and our steadfast commitment to providing the best solutions for developers and customers."
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.