Update app.py

app.py

@@ -5,6 +5,10 @@ By effectively treating each residue as its own reference frame, we shift the eq
 representation space itself; this allows us to use a vanilla transformer model as our model. Here, we provide a simple
 online interface for generating single backbones with a given length, starting from a given random seed. 
 
+Tips for generating proteins:
+* The maximum sequence sequence length this model has been trained on is 128 residues. The shorter a sequence is, the more likely it will be "designable" (see our preprint).
+* FoldingDiff does *not* generate the amino acid sequence for its structures, it simply fills the structure with Glycine residues; use a tool like ESM-IF1 to generate amino acids corresponding to generated structure.
+
 See our preprint at https://arxiv.org/abs/2209.15611 and our full codebase at https://github.com/microsoft/foldingdiff
 """