Stable Diffusion AI has been updated to produce more realistic images

The startup Stability AI is from the latest version of the artificial intelligence model called Stable Diffusion SDXL v0.9 unveiled This version is able to create more realistic images and does not have problems such as strange depictions of hands.

According to Decrypt, the new version of the Stable Diffusion artificial intelligence model performs much better than the previous versions and only by receiving a few words It can make fantastic pictures. This version is designed to do most of the work for you, and even when talking to the artificial intelligence, you can speak to it in a more natural language.

Stability AI says SDXL 0.9 is capable of enhancing image detail and Applications of image production increase with the help of artificial intelligence. The company says version 0.9 has much better text-to-image performance than the beta version released earlier in April.

To see this progress, you can compare the images from a common input in these two versions. For example, the new version for the description “Wolf in Yosemite National Park with a nature documentary theme” has created a more realistic image that has much more detail than the previous version.

The number of Stable Diffusion parameters has increased a lot

Part of the reason for this noticeable improvement is related to the significant increase in the number of SDXL 0.9 parameters. now 3.5 billion parameters The foundation allows the new Stability AI model to learn more deeply and produce better images according to its learnings.

Version 0.9 is also more inclined towards simple inputs and can produce better results based on them. For example, if you give the AI ​​the description “two hands pointing at each other”, you can get a very realistic representation of the model. The progress of this version compared to previous versions is quite evident.

The new version of Stable Diffusion, which first became famous in August 2022 and is considered one of the open source models of image production, can now be a more serious competitor to Dall-E and Midjourney with this update. This model is considered to be the most popular open source artificial intelligence for producing images.

The latest version of Stable Diffusion comes through Init ML’s image generation and editing tool. This model cannot be used or trained locally at this time, but when it goes public, it will require at least 16GB of RAM and a graphics card with 8GB of VRAM to run. The public release date of this version is likely Mid July will be.