Announcing the upcoming deprecation of fine-tuned SDXL models and SDXL 0.9 & BETA

Announcing the upcoming deprecation of fine-tuned SDXL models and SDXL 0.9 & BETA

Some features on NightCafe are powered by Stability AI, the creators of Stable Diffusion. Stability AI have announced that they're deprecating the following features:
  1. Fine-tuning for SDXL will be deprecated on October 18
  2. Stable Diffusion XL v0.9 & SDXL BETA will be deprecated on October 11
Unfortunately we've been given very little notice about these major changes.

What this means for NightCafe users:

  1. SDXL v0.9 & SDXL BETA will be removed on October 11
  2. SDXL and Stable Diffusion 1.5 fine-tuning has been disabled
  3. Existing fine-tuned SDXL models will become unusable on October 18
  4. Existing fine-tuned Stable Diffusion 1.5 models will continue to work
  5. Creations that used fine-tuned models will be unaffected
  6. Datasets will be unaffected (so it's pretty simple to use the same dataset to train a Flux model)

Compensation for model creators

This news means that anyone who has fine-tuned an SDXL model will lose access to that model. We've been attempting to convince Stability AI for a long time (long before they made this announcement) to give us access to the fine-tuned model (LoRA) files so that we can continue to support existing LoRAs on another provider, but they've refused.

Thankfully, Flux fine-tuning is now here and is better than SDXL fine-tuning ever was (albeit more expensive due to the larger base model), especially for faces. To compensate users who have created a fine-tuned model, we'll be refunding the cost of any SDXL fine-tuned models created or used in the last 3 months at a rate of 2 fine-tuning credits per model. If you've created a fine-tuned SDXL model that has been used in the last 3 months, you will be refunded by Saturday the 12th of October. You will be notified when your refund has been processed.