Accept tensor ip_adapter_image_embeds in SD3 ControlNet inpainting

src/diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py

@@ -777,15 +777,10 @@ def check_inputs(
                 "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
             )
 
-        if ip_adapter_image_embeds is not None:
-            if not isinstance(ip_adapter_image_embeds, list):
-                raise ValueError(
-                    f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
-                )
-            elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
-                raise ValueError(
-                    f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
-                )
+        if ip_adapter_image_embeds is not None and ip_adapter_image_embeds.ndim not in [3, 4]:
+            raise ValueError(
+                f"`ip_adapter_image_embeds` has to be a 3D or 4D tensor but is {ip_adapter_image_embeds.ndim}D"
+            )
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3.StableDiffusion3Pipeline.prepare_latents
     def prepare_latents(

tests/pipelines/controlnet_sd3/test_controlnet_inpaint_sd3.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import pytest
 import torch
 from transformers import (
     AutoConfig,
@@ -195,6 +196,35 @@ def test_controlnet_inpaint_sd3(self):
 
         assert_tensors_close(image_slice.flatten().cpu(), expected_slice, atol=1e-2)
 
+    def _check_inputs_with_ip_adapter_embeds(self, pipe, ip_adapter_image_embeds):
+        pipe.check_inputs(
+            height=32,
+            width=32,
+            image=torch.zeros(1, 3, 32, 32),
+            prompt=None,
+            prompt_2=None,
+            prompt_3=None,
+            prompt_embeds=torch.zeros(1, 2, 32),
+            pooled_prompt_embeds=torch.zeros(1, 64),
+            ip_adapter_image_embeds=ip_adapter_image_embeds,
+            control_guidance_start=[0.0],
+            control_guidance_end=[1.0],
+        )
+
+    def test_check_inputs_accepts_ip_adapter_image_embeds_tensor(self):
+        # Regression: `check_inputs` required a `list`, but the `__call__` docstring and
+        # `prepare_ip_adapter_image_embeds` (which calls `.chunk(2)` on it) both document and use a
+        # tensor, so the documented argument could never be passed.
+        pipe = self.get_pipeline()
+
+        self._check_inputs_with_ip_adapter_embeds(pipe, torch.zeros(1, 2, 32))
+
+    def test_check_inputs_rejects_ip_adapter_image_embeds_with_bad_ndim(self):
+        pipe = self.get_pipeline()
+
+        with pytest.raises(ValueError, match="has to be a 3D or 4D tensor"):
+            self._check_inputs_with_ip_adapter_embeds(pipe, torch.zeros(1, 32))
+
 
 class TestStableDiffusion3ControlNetInpaintingPipelineMemory(
     StableDiffusion3ControlNetInpaintingPipelineTesterConfig, MemoryTesterMixin