Stable Diffusion 2.0 arrives and its new options to generate images with AI leave us more speechless than ever

  • 19

Come on, I say it. stable diffusion is (for me) the product of the year. This engine for generating images via artificial intelligence has become a true revolution that with its lights and shadows does not stop evolving. It has just done it roundly with the publication of its 2.0 version that goes even further than the first one. Which is to say.

Stable Diffusion 2.0. This “AI Imager Linux” has just announce their second stable versionand although the announcement details the improvements, the curious thing is that the company that manages everything, Stability.ai, doesn’t even have a proper landing page for Stable Diffusion. Its official website is neither more nor less than your github repository. Wonderful, as evidenced by the fact that it’s the GitHub project fastest growing of “stars” in all of history, far surpassing previous standouts such as Bitcoin, Ethereum or Apache Kafkaan event streaming platform.

Github

Can’t find Stable Diffusion? Normal, because it looks like the Y coordinate axis. It’s there, to the left of everything, almost like a vertical wall from the rest. Spectacular. Source: A16z.

It remains as an absolute reference. It wasn’t the first -DALL-E 2 or Midjourney are equally amazing- but the Open Source philosophy of Stable Diffusion it has been crucial to position itself as the great reference in this field. Can install it locally or use it as a plugin in other applications like Photoshop or Canva has shown almost limitless potential for creators, businesses, and mainstream users.

Sd2

Text-to-image conversion surpasses itself. In this version, a new text-to-image encoder called OpenCLIP is used, which according to those responsible for the project “greatly improves the quality of the generated images compared to the V1 versions”. The engine maintains the filters to remove adult content thanks to the nsfw filter from LAION-5B, the set of images used to train this model

Sd3

Upscaling and more resolution than ever. Although the engine natively generates images of up to 768×768 pixels, Stable Diffusion 2.0 includes a new upscaling system that improves the resolution of images by multiplying them by up to four. Thus, it is possible to generate images of 2,048×2,048 and even more, and to do so with fantastic definition.

Same base for different images. The diffusion model Depth-to-Image it goes beyond what was achieved with the picture-to-picture option of V1. That option allowed us to make a quick sketch of what we wanted to get and Stable Diffusion would generate the image based on that and the descriptive input text (prompt). The new model can use a base image, but it generates not one, but multiple images using both the text and “depth” information given by the starting image.

Come on, I say it. stable diffusion is (for me) the product of the year. This engine for generating images…

Come on, I say it. stable diffusion is (for me) the product of the year. This engine for generating images…

Leave a Reply

Your email address will not be published.