diff --git a/internnav/model/basemodel/internvla_n1/nextdit_traj.py b/internnav/model/basemodel/internvla_n1/nextdit_traj.py
index 1c18a63f..30142a6e 100644
--- a/internnav/model/basemodel/internvla_n1/nextdit_traj.py
+++ b/internnav/model/basemodel/internvla_n1/nextdit_traj.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Any, Dict, Optional
+from typing import Any, Callable, Dict, Optional
 
 import torch
 import torch.nn as nn
@@ -292,9 +292,38 @@ def __init__(
 
         assert (hidden_size // num_attention_heads) % 4 == 0, "2d rope needs head dim to be divisible by 4"
 
-    def _set_gradient_checkpointing(self, module, value=False):
-        if hasattr(module, "gradient_checkpointing"):
-            module.gradient_checkpointing = value
+    def _set_gradient_checkpointing(
+        self,
+        module: Optional[nn.Module] = None,
+        value: Optional[bool] = None,
+        *,
+        enable: Optional[bool] = None,
+        gradient_checkpointing_func: Optional[Callable] = None,
+    ) -> None:
+        """Set checkpointing flags across diffusers API generations.
+
+        Diffusers versions up to 0.33 call this hook once per module as
+        ``_set_gradient_checkpointing(module, value=True)``.  Newer versions
+        call it on the model with ``enable=`` and a checkpoint function.  A
+        single adapter keeps both call conventions functional and propagates
+        the function supplied by newer versions to every checkpointable
+        submodule.
+        """
+        if enable is not None:
+            value = enable
+        if value is None:
+            value = False
+
+        if module is None:
+            modules = self.modules()
+        else:
+            modules = (module,)
+
+        for child in modules:
+            if hasattr(child, "gradient_checkpointing"):
+                child.gradient_checkpointing = value
+                if gradient_checkpointing_func is not None:
+                    child._gradient_checkpointing_func = gradient_checkpointing_func
 
     def forward(
         self,
diff --git a/tests/unit_test/test_gradient_checkpointing_compat.py b/tests/unit_test/test_gradient_checkpointing_compat.py
new file mode 100644
index 00000000..8124aa4e
--- /dev/null
+++ b/tests/unit_test/test_gradient_checkpointing_compat.py
@@ -0,0 +1,36 @@
+"""Regression coverage for the NextDiT checkpointing hook."""
+
+import pytest
+
+
+torch = pytest.importorskip("torch")
+pytest.importorskip("diffusers")
+pytest.importorskip("transformers")
+
+from internnav.model.basemodel.internvla_n1.nextdit_crossattn_traj import (
+    NextDiTCrossAttn,
+    NextDiTCrossAttnConfig,
+)
+
+
+def test_nextdit_enables_and_disables_gradient_checkpointing():
+    config = NextDiTCrossAttnConfig(
+        input_size=4,
+        patch_size=1,
+        in_channels=4,
+        dim=32,
+        n_layers=1,
+        n_heads=2,
+        n_kv_heads=2,
+        multiple_of=8,
+        latent_embedding_size=16,
+        _gradient_checkpointing=True,
+    )
+
+    model = NextDiTCrossAttn(config)
+    assert model.model.gradient_checkpointing is True
+    assert model.model.is_gradient_checkpointing is True
+
+    model.model.disable_gradient_checkpointing()
+    assert model.model.gradient_checkpointing is False
+    assert model.model.is_gradient_checkpointing is False