Update README.md

README.md

@@ -25,6 +25,10 @@ from, as long as your dataset is rather large.
 **Note:** This checkpoint was also already trained on multi-aspect-ratios, meaning you can generate larger images than just 1024x1024. Sometimes generations up to 2048x2048
 even work. Feel free to try it out!
 
+**Also Note:** The base checkpoint usually requires a higher classifier-free-guidance value (`guidance_scale=8.0`) and also a negative caption in order to make good
+looking images. The [interpolated model](https://huggingface.co/warp-ai/wuerstchen-prior-model-interpolated) and [finetuned model](https://huggingface.co/warp-ai/wuerstchen-prior-model-finetuned)
+usually don't need a negative caption and work better with a lower classifier-free-guidance value (`guidance_scale=4.0`).
+
 ### Image Sizes
 Würstchen was trained on image resolutions between 1024x1024 & 1536x1536. We sometimes also observe good outputs at resolutions like 1024x2048. Feel free to try it out.
 We also observed that the Prior (Stage C) adapts extremely fast to new resolutions. So finetuning it at 2048x2048 should be computationally cheap.