• 2 Posts
  • 3 Comments
Joined 1 year ago
cake
Cake day: July 2nd, 2023

help-circle
  • What’s New in v3.0.0 Quite a lot has changed, both internally and externally

    Web User Interface:

    A ControlNet interface that gives you fine control over such things as the posture of figures in generated images by providing an image that illustrates the end result you wish to achieve.

    A Dynamic Prompts interface that lets you generate combinations of prompt elements.

    A redesigned user interface which makes it easier to access frequently-used elements, such as the random seed generator.

    The ability to create multiple image galleries, allowing you to organize your generated images topically or chronologically.

    A graphical node editor that lets you design and execute complex image generation operations using a point-and-click interface (see below for more about nodes)

    Macintosh users can now load models at half-precision (float16) in order to reduce the amount of RAM used by each model by half.

    Advanced users can choose earlier CLIP layers during generation to produce a larger variety of images. Lots of new samplers/schedulers!





  • We present SDXL, a latent diffusion model for text-to-image synthesis. Compared to previous versions of Stable Diffusion, SDXL leverages a three times larger UNet backbone: The increase of model parameters is mainly due to more attention blocks and a larger cross-attention context as SDXL uses a second text encoder. We design multiple novel conditioning schemes and train SDXL on multiple aspect ratios. We also introduce a refinement model which is used to improve the visual fidelity of samples generated by SDXL using a post-hoc image-to-image technique. We demonstrate that SDXL shows drastically improved performance compared the previous versions of Stable Diffusion and achieves results competitive with those of black-box state-of-the-art image generators. In the spirit of promoting open research and fostering transparency in large model training and evaluation, we provide access to code and model weights.