Update README.md

README.md

@@ -6,4 +6,20 @@ tags:
 - art
 size_categories:
 - 100K<n<1M
----
+---
+
+Used to detect severely stylistically corrupted anime images generated by Stable Diffusion.
+
+There are two classes: `corrupted` and `normal`, comprising 61,100 and 63,004 images, respectively.
+
+For the `corrupted` type, it includes several cases such as:
+* Entirely black images
+* Mosaics or stylistic anomalies due to low sampling steps
+* Stylistic anomalies due to excessively weighted tags
+* Stylistic anomalies due to embedding with excessive weights
+
+It's important to note that the following content is not included:
+* Distortions in facial and hand details caused by resolution issues
+* Distortions in human and object structures
+
+For the `normal` type, it contains regular AI-generated images as well as approximately 15,000 images hand-drawn by humans.