diff --git a/src/diffusers/models/attention_dispatch.py b/src/diffusers/models/attention_dispatch.py
index 310c44457c27..3aed728b50d1 100644
--- a/src/diffusers/models/attention_dispatch.py
+++ b/src/diffusers/models/attention_dispatch.py
@@ -235,6 +235,10 @@ def decorator(func):
     def get_active_backend(cls):
         return cls._active_backend, cls._backends[cls._active_backend]
 
+    @classmethod
+    def set_active_backend(cls, backend: str):
+        cls._active_backend = backend
+
     @classmethod
     def list_backends(cls):
         return list(cls._backends.keys())
@@ -294,12 +298,12 @@ def attention_backend(backend: Union[str, AttentionBackendName] = AttentionBacke
     _maybe_download_kernel_for_backend(backend)
 
     old_backend = _AttentionBackendRegistry._active_backend
-    _AttentionBackendRegistry._active_backend = backend
+    _AttentionBackendRegistry.set_active_backend(backend)
 
     try:
         yield
     finally:
-        _AttentionBackendRegistry._active_backend = old_backend
+        _AttentionBackendRegistry.set_active_backend(old_backend)
 
 
 def dispatch_attention_fn(
@@ -348,6 +352,18 @@ def dispatch_attention_fn(
             check(**kwargs)
 
     kwargs = {k: v for k, v in kwargs.items() if k in _AttentionBackendRegistry._supported_arg_names[backend_name]}
+
+    if "_parallel_config" in kwargs and kwargs["_parallel_config"] is not None:
+        attention_backend = AttentionBackendName(backend_name)
+        if not _AttentionBackendRegistry._is_context_parallel_available(attention_backend):
+            compatible_backends = sorted(_AttentionBackendRegistry._supports_context_parallel)
+            raise ValueError(
+                f"Context parallelism is enabled but backend '{attention_backend.value}' "
+                f"which does not support context parallelism. "
+                f"Please set a compatible attention backend: {compatible_backends} using `model.set_attention_backend()` before "
+                f"calling `model.enable_parallelism()`."
+            )
+
     return backend_fn(**kwargs)
 
 
diff --git a/src/diffusers/models/modeling_utils.py b/src/diffusers/models/modeling_utils.py
index 41da95d3a2a2..197a0d78bc36 100644
--- a/src/diffusers/models/modeling_utils.py
+++ b/src/diffusers/models/modeling_utils.py
@@ -602,6 +602,7 @@ def set_attention_backend(self, backend: str) -> None:
         from .attention import AttentionModuleMixin
         from .attention_dispatch import (
             AttentionBackendName,
+            _AttentionBackendRegistry,
             _check_attention_backend_requirements,
             _maybe_download_kernel_for_backend,
         )
@@ -629,6 +630,9 @@ def set_attention_backend(self, backend: str) -> None:
                 continue
             processor._attention_backend = backend
 
+        # Important to set the active backend so that it propagates gracefully throughout.
+        _AttentionBackendRegistry.set_active_backend(backend)
+
     def reset_attention_backend(self) -> None:
         """
         Resets the attention backend for the model. Following calls to `forward` will use the environment default, if
@@ -1541,7 +1545,7 @@ def enable_parallelism(
                         f"Context parallelism is enabled but the attention processor '{processor.__class__.__name__}' "
                         f"is using backend '{attention_backend.value}' which does not support context parallelism. "
                         f"Please set a compatible attention backend: {compatible_backends} using `model.set_attention_backend()` before "
-                        f"calling `enable_parallelism()`."
+                        f"calling `model.enable_parallelism()`."
                     )
 
                 # All modules use the same attention processor and backend. We don't need to