Experimental Offline Standalone Mac Installer for Stable Diffusion Web UI (unofficial) 20230804

[brkirch]

This is an offline standalone installer for a modified version of AUTOMATIC1111's Stable Diffusion web UI that includes experimental changes and uses the nightly PyTorch build from 8/4. Note that this is an test build so crashes and bugs should be expected. All dependencies needed to start using Stable Diffusion (other than the model) are included and this does not require Xcode command line tools to be installed.

This release currently only has the installer for Apple silicon Macs, an installer for Intel Macs may be added later. macOS 13.4 or later is highly recommended, older versions of macOS are untested.

IMPORTANT

AUTOMATIC1111#12181 and AUTOMATIC1111#12227 are included. Also included is a new experimental option Hires noisy latent.

Hires noisy latent (currently works for k-diffusion samplers only) does an incomplete (noisy) image generation and then sends the result to the highres model which then completes the image generation. One significant difference from the usual process of img2img that is done for highres is that no additional noise is applied to the image when starting the highres pass. This new option is intended to more closely match the official SD XL base model + refiner workflow.

On Macs with 32 GB+ of RAM, for using SD XL with the refiner it is highly recommended that the Stable Diffusion > Maximum number of checkpoints loaded at the same time setting is set to 2 and the Stable Diffusion > Only keep one model on device (will keep models other than the currently used one in RAM rather than VRAM) option is unchecked. For Macs with less than 32 GB of RAM it may still make sense to keep 2 checkpoints loaded, but it might work better to enable the option to only keep one model on device (I can't currently test this, so please give feedback in the discussion for this release on what works best for your setup).

Also, there is currently a bug that sometimes the SD XL base model is not reloaded after using the refiner - if that happens then use the model dropdown to set the model to the refiner and then set it back to the base model.

---

This discussion was created from the release Experimental Offline Standalone Mac Installer for Stable Diffusion Web UI (unofficial) 20230804.