diff --git a/README.md b/README.md
index 2565273..a4472b8 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,6 @@
+# Notice
+As of the SD XL update the workaround used to patch the hardcoded transformer model from the HuggingFace library no longer works. I might update it in the future, but as of now, the contribution of this extension is not great enough to spend time on. The directional prompt attention sounds good on paper, but models are in no way finetuned to deal with this forced attention view, and the quality sufffers from this. Furthermore, this extension only affects the CLIP part of the framework, but as the SD part is conditioned on a "summarized" representation of the prompt, the SD part of the framework gets to see all input prompts anyway, making it even harder for this method to work consistently.
+
 # CLIP Directional Prompt Attention for ComfyUI
 ### What is Directional Prompt Attention?
 Direction prompt attention tries to solve the problem of contextual words (or parts of the prompt) having an effect on much later or irrelevant parts of the prompt. For example, this happens often when something is described as a color which makes subsequent parts of the prompt also have this color. [Cutoff for ComfyUI](https://github.com/BlenderNeko/ComfyUI_Cutoff) is script/extension which tries to solve this through isolated prompt masking. However, this can be achieved much easier by simple using an already built-in feature of the CLIP transformer: attention masks. Using attention masks the transformer is limited to only apply attention on certain tokens (words) in the prompt.