Update README.md

README.md

@@ -67,6 +67,8 @@ This ensures that the model learns to associate the embedding with the character
 
 ![Hamburger Comparison](hamburger_comparison.jpg)
 
+*Each column uses the same seed; rows compare baseline vs. embedding outputs to isolate the effect of the embedding across prompts. Improvements are consistent across seeds and domains.*
+
 
 ## Usage & Adoption
 Boring Embeddings have been adopted across multiple Stable Diffusion communities:
@@ -142,6 +144,8 @@ To qualitatively illustrate how well the Boring embeddings have learned to impro
 
 ![Performance on Simple Prompts](comparison.jpg)
 
+*Paired comparison using identical seeds and generation settings; differences reflect only the effect of the embedding.*
+
 As we can see, putting these embeddings in the negative prompt yields a more delicious burger, a more vibrant and detailed landscape, a prettier pharoah, and a more 3-d-looking aquarium.
 Hyperparameters were tuned based on manual evaluations of grids like these.