Log WanAnimate interpolation resizing as a warning

src/diffusers/pipelines/wan/pipeline_wan_animate.py

@@ -569,7 +569,7 @@ def prepare_prev_segment_cond_latents(
  latent_height = height // self.vae_scale_factor_spatial
  latent_width = width // self.vae_scale_factor_spatial
  if segment_height != height or segment_width != width:
-print(
+logger.warning(
  f"Interpolating prev segment cond video from ({segment_width}, {segment_height}) to ({width}, {height})"
  )
  # Perform a 4D (spatial) rather than a 5D (spatiotemporal) reshape, following the original code

tests/pipelines/wan/test_wan_animate.py

@@ -12,6 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
+
 import pytest
 import torch
 from PIL import Image
@@ -178,6 +180,28 @@ def test_inference_replacement(self):
  video = pipe(**inputs).frames[0]
  assert video.shape == self.output_shape
 
+ def test_prepare_prev_segment_cond_latents_logs_interpolation_warning(self, caplog):
+ pipe = self.get_pipeline()
+ prev_segment_cond_video = torch.zeros((1, 3, 1, 8, 8), dtype=torch.float32)
+
+ with caplog.at_level(logging.WARNING, logger="diffusers.pipelines.wan.pipeline_wan_animate"):
+ pipe.prepare_prev_segment_cond_latents(
+ prev_segment_cond_video=prev_segment_cond_video,
+ batch_size=1,
+ segment_frame_length=5,
+ height=16,
+ width=16,
+ prev_segment_cond_frames=1,
+ task="animate",
+ dtype=torch.float32,
+ device=torch.device("cpu"),
+ )
+
+ assert any(
+ "Interpolating prev segment cond video from (8, 8) to (16, 16)" in record.message
+ for record in caplog.records
+ )
+
  @pytest.mark.skip(
  reason="Setting the Wan Animate latents to zero at the last denoising step does not guarantee that the output"
  " will be zero. I believe this is because the latents are further processed in the outer loop where we loop"