Stability AI发布最新图像生成模型Stable Diffusion XL 1.0 Stability AI has released its latest image generation model, Stable Diffusion XL 1.0.
今天,Stability AI 宣布推出 Stable Diffusion XL 1.0,这是该公司迄今为止描述为“最先进”的文本到图像模型。Stable Diffusion XL 1.0 可在 GitHub (https://github.com/Stability-AI/generative-models) 上以开源形式使用,并提供给 Stability 的 API (https://platform.stability.ai/docs/release-notes) 和消费者应用程序 ClipDrop (https://clipdrop.co/stable-diffusion) 和 DreamStudio (https://dreamstudio.ai/)。Stability 声称,与其前代产品相比,Stable Diffusion XL 1.0 提供了“更生动”和“更准确”的颜色,更好的对比度、阴影和照明效果。
在接受采访时,Stability AI 的应用机器学习负责人Joe Penna 指出,Stable Diffusion XL 1.0 包含 35亿个参数,可以“在几秒钟内”生成多种长宽比的 100万像素分辨率的完整图像。“参数”是模型从训练数据中学习到的部分,基本上定义了模型在解决问题上的技能水平。
上一代稳定扩散模型稳定扩散 XL 0.9 也可以生成更高分辨率的图像,但需要更多的计算能力
Today, Stability AI announced the release of Stable Diffusion XL 1.0, which is described as their "most advanced" text-to-image model to date. Stable Diffusion XL 1.0 is available on GitHub (https://github.com/Stability-AI/generative-models) as an open-source project and can be accessed through Stability's API (https://platform.stability.ai/docs/release-notes) as well as consumer applications like ClipDrop (https://clipdrop.co/stable-diffusion) and DreamStudio (https://dreamstudio.ai/). Stability claims that compared to its predecessor, Stable Diffusion XL 1.0 offers "more vivid" and "more accurate" colors, improved contrast, shadows, and lighting effects.
During an interview, Joe Penna, the Head of Applied Machine Learning at Stability AI mentioned that Stable Diffusion XL 1.0 consists of 3.5 billion parameters and can generate full images with resolutions up to one million pixels in various aspect ratios "within seconds." Parameters are essentially the learned components of the model from training data that define its skill level in problem-solving.
The previous generation model, Stable Diffusion XL 0.9 also had the capability to generate higher resolution images but required more computational power.
评论
发表评论