readme

Author: lucidrains

README.md

@@ -147,6 +147,8 @@ You can also use the pretrained VAE offered by the authors of <a href="https://g
 
 In contrast to OpenAI's VAE, it also has an extra layer of downsampling, so the image sequence length is 256 instead of 1024 (this will lead to a 16 reduction in training costs, when you do the math). Whether it will generalize as well as the original DALL-E is up to the citizen scientists out there to discover.
 
+Update - <a href="https://github.com/lucidrains/DALLE-pytorch/discussions/131">it works!</a>
+
 ```python
 from dalle_pytorch import VQGanVAE1024