fix: correct boolean logic in zero_rank_print that prevented all output

[Mr-Neutr0n]

Summary

The zero_rank_print function in animatediff/utils/util.py contains a boolean logic error that causes it to never print anything, regardless of the distributed process state.

The Bug

The current condition is:

if (not dist.is_initialized()) and (dist.is_initialized() and dist.get_rank() == 0):

This follows the pattern (NOT A) AND (A AND B), which is a logical contradiction — it requires dist.is_initialized() to be both False and True simultaneously. The condition evaluates to False for every possible input, so the print statement is dead code.

The Fix

Change the and between the two clauses to or:

if (not dist.is_initialized()) or (dist.is_initialized() and dist.get_rank() == 0):

This correctly implements the intended behavior:

• If distributed training is NOT initialized: print the message (single-process mode).
• If distributed training IS initialized and this is rank 0: print the message (only the main process prints).

This is the standard pattern used across PyTorch distributed training codebases for rank-guarded logging.