Stability AI and DeepFloyd just unveiled DeepFloyd IF, a cascaded pixel-level diffusion model released under a non-commercial research license. By packing the massive T5-XXL language model as its text encoder, it generates coherent, readable text directly on rendered images—a notorious blind spot for older models. A fully open-source release is planned down the road.
A leaked investor deck reveals that Stability AI may have exaggerated its role in creating Stable Diffusion and other top AI projects during its $101m Series A fundraise in October 2022. The presentation claimed "co-founding" status with several organizations, sparking scrutiny over who actually owns the IP. Why it matters: It casts doubt on the foundation of one of generative AI's earliest superstars.
Stability AI just launched the Alpha version of Stable LM, an open-source language model suite debuting at 3B and 7B parameter sizes. Trained on 1.5 trillion tokens, the lightweight models handle text and code, with base checkpoints licensed for commercial use under CC BY-SA-4.0. Why it matters: it gives developers a way to build local, private AI applications without being handcuffed to proprietary cloud APIs.
Stability AI just introduced Stable Diffusion XL (SDXL), an advanced text-to-image flagship packing 3.5 billion parameters. It focuses on photorealistic outputs and high-resolution generation across diverse visual styles. It gives creators and developers much sharper image quality and greater creative control straight out of the box.
Stability AI is in serious financial trouble, burning through cash at an alarming rate without a clear path to profitability, according to Semafor. After burning through most of its $100M raised last year, the startup's planned $4B valuation round is on shaky ground as investors hesitate. It turns out massive open-source popularity doesn't guarantee a sustainable business model.
Stability AI just launched Stable Diffusion Reimagine, a new Clipdrop tool that generates endless image variations without any text prompts. By replacing traditional text encoders with an image encoder, it cooks up fresh compositions without reusing a single pixel from the original image. Why it matters: Designers can iterate on visual concepts instantly with one click, and the open-source release is coming soon.
Stability AI just made its first-ever acquisition, buying Init ML, the team behind the popular image-editing suite Clipdrop. With over 15 million users already on board, Init ML will operate as an independent subsidiary, integrating Stability’s latest open-source generative models directly into Clipdrop’s everyday creative apps.
Stability AI is reportedly in talks to raise new capital at a valuation of around $4 billion, according to Bloomberg. The startup behind Stable Diffusion reached unicorn status just months ago after a $101 million round. While final terms could still change, the huge leap underscores soaring investor appetite for generative AI.
Getty Images just filed a major lawsuit against Stability AI in Delaware federal court, alleging the startup scraped over 12 million copyrighted photos to train Stable Diffusion. The visual media giant also claims the AI spits out mangled Getty watermarks, and is demanding the destruction of all infringing model versions.
A trio of independent artists just filed a class-action lawsuit against Stability AI, Midjourney, and DeviantArt for training image models on their work without consent or compensation. The plaintiffs argue anyone can now mimic their styles with a simple prompt, threatening their livelihoods. Why it matters: This marks the first major courtroom battle testing whether AI training qualifies as fair use.
Getty Images just filed a lawsuit against Stability AI in London, accusing the company of copying millions of copyrighted photos to train Stable Diffusion without a license. Getty emphasized that it is not against AI art and already licenses images to other AI firms, but claims Stability bypassed commercial licensing options to profit from its intellectual property.
Artists just filed a class-action copyright lawsuit against Stability AI, Midjourney, and DeviantArt for scraping billions of images without consent to train their generators. The plaintiffs allege copyright and DMCA violations, seeking damages and an injunction to block future harms. Why it matters: This lawsuit sets up a landmark test for whether AI training on web data qualifies as fair use.
The Guardian breaks down how image generators like Midjourney and Lensa are rapidly encroaching on commercial illustration, sparking intense pushback over copyright and livelihood. While initiatives like Spawning try to give creators opt-out controls, visual styles remain difficult to legally protect. Why it matters: The question is no longer whether AI can produce art, but how creators will survive alongside tools trained on their life's work.
Diffusion models just took over AI image generation, but their reach goes way beyond art. Inspired by thermodynamics, the technology reconstructs data by removing noise step-by-step, guided by OpenAI's CLIP. Why it matters: Researchers are already using diffusion to compose music, design custom proteins, and synthesize DNA sequences.
Kickstarter just shut down the crowdfunding campaign for Unstable Diffusion, a group aiming to build NSFW generative AI models, following intense backlash from working artists. The campaign had pulled in over $56,000 from 867 backers, all of which will now be refunded. Why it matters: Crowdfunding platforms are being forced to draw hard ethical lines around nonconsensual imagery and uncredited training data.
Stanford AIMI researchers just showed that fine-tuning off-the-shelf Stable Diffusion can generate clinically accurate chest X-rays with recognizable lung abnormalities. Why it matters: Building medical imaging models usually requires massive proprietary datasets and prohibitive compute, but repurposing open-source diffusion models slashes the barrier to training AI for rare diseases.
Stability AI just rolled out Stable Diffusion 2.0, headlined by a new depth2img feature that infers 3D depth maps via the MiDaS model. By locking in geometric structure rather than just relying on text prompts and pixels, artists can now graybox a rough 3D scene and let AI explore lighting and texturing with far fewer structural distortions.
