Announcing the deprecation of some legacy models on November 14

Announcing the deprecation of some legacy models on November 14

Note: While some models are being deprecated, they're being replaced with better models at the same price. See "What's coming instead?" below.

Which models are being deprecated?

On the 14th of November we'll be retiring some older Stable Diffusion models:
  1. Stable Diffusion 2.0
  2. Stable Diffusion 2.0 (768)
  3. Stable Diffusion 2.1
  4. Stable Diffusion 2.1 (768)
  5. Stable Inpainting 1.0
  6. Stable Inpainting 2.0
This is part of a move from our legacy image generation back-end to a newer one that is faster and allows us to add more features and models.
Despite the inpainting models being retired, inpainting will continue to work.

Models that will move to the new back-end

Other models that are running on the legacy image generation back-end that are NOT being retired are:
  1. Stable Diffusion 1.4
  2. Stable Diffusion 1.5
These will continue to be available, but they will run on our new back-end. This means that while the models are the same and will generally generate the same things as before, the same generation settings and seed will NOT generate the exact same image as the old back-end did.

What's coming instead?

The new back-end is faster and more flexible than the old back-end. When these changes land, there will also be a range of new models that are significantly better than all the models mentioned above, available at the same price. The new back-end will also (soon) support fine-tuning on Stable Diffusion 1.5, and other features that the old back-end does not support.

What else will change?

The new back-end supports pretty much everything the old back-end supports, with a few exceptions:
  1. CLIP Guidance is not supported
  2. Start images with non-standard aspect ratios will no-longer generate non-standard sized outputs - the outputs will be one of the standard aspect ratios (the closest to your start image)