[wip] Text to Image Pipeline for wan

src/diffusers/__init__.py

@@ -585,6 +585,7 @@
             "VQDiffusionPipeline",
             "WanImageToVideoPipeline",
             "WanPipeline",
+            "WanTextToImagePipeline",
             "WanVACEPipeline",
             "WanVideoToVideoPipeline",
             "WuerstchenCombinedPipeline",
@@ -1221,6 +1222,7 @@
             VQDiffusionPipeline,
             WanImageToVideoPipeline,
             WanPipeline,
+            WanTextToImagePipeline,
             WanVACEPipeline,
             WanVideoToVideoPipeline,
             WuerstchenCombinedPipeline,

src/diffusers/pipelines/__init__.py

@@ -379,7 +379,7 @@
         "WuerstchenDecoderPipeline",
         "WuerstchenPriorPipeline",
     ]
-    _import_structure["wan"] = ["WanPipeline", "WanImageToVideoPipeline", "WanVideoToVideoPipeline", "WanVACEPipeline"]
+    _import_structure["wan"] = ["WanPipeline", "WanImageToVideoPipeline", "WanVideoToVideoPipeline", "WanVACEPipeline", "WanTextToImagePipeline"]
     _import_structure["skyreels_v2"] = [
         "SkyReelsV2DiffusionForcingPipeline",
         "SkyReelsV2DiffusionForcingImageToVideoPipeline",
@@ -764,7 +764,7 @@
             UniDiffuserTextDecoder,
         )
         from .visualcloze import VisualClozeGenerationPipeline, VisualClozePipeline
-        from .wan import WanImageToVideoPipeline, WanPipeline, WanVACEPipeline, WanVideoToVideoPipeline
+        from .wan import WanImageToVideoPipeline, WanPipeline, WanVACEPipeline, WanVideoToVideoPipeline, WanTextToImagePipeline
         from .wuerstchen import (
             WuerstchenCombinedPipeline,
             WuerstchenDecoderPipeline,

src/diffusers/pipelines/wan/__init__.py

@@ -22,10 +22,12 @@
 
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
+    _import_structure["pipeline_output"] = ["WanPipelineOutput", "WanImagePipelineOutput"]
     _import_structure["pipeline_wan"] = ["WanPipeline"]
     _import_structure["pipeline_wan_i2v"] = ["WanImageToVideoPipeline"]
     _import_structure["pipeline_wan_vace"] = ["WanVACEPipeline"]
     _import_structure["pipeline_wan_video2video"] = ["WanVideoToVideoPipeline"]
+    _import_structure["pipeline_wan_t2i"] = ["WanTextToImagePipeline"]
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
         if not (is_transformers_available() and is_torch_available()):
@@ -34,10 +36,12 @@
     except OptionalDependencyNotAvailable:
         from ...utils.dummy_torch_and_transformers_objects import *
     else:
+        from .pipeline_output import WanPipelineOutput, WanImagePipelineOutput
         from .pipeline_wan import WanPipeline
         from .pipeline_wan_i2v import WanImageToVideoPipeline
         from .pipeline_wan_vace import WanVACEPipeline
         from .pipeline_wan_video2video import WanVideoToVideoPipeline
+        from .pipeline_wan_t2i import WanTextToImagePipeline
 
 else:
     import sys

src/diffusers/pipelines/wan/pipeline_output.py

@@ -18,3 +18,18 @@ class WanPipelineOutput(BaseOutput):
     """
 
     frames: torch.Tensor
+
+
+@dataclass
+class WanImagePipelineOutput(BaseOutput):
+    r"""
+    Output class for Wan text-to-image pipelines.
+
+    Args:
+        images (`torch.Tensor`, `np.ndarray`, or List[PIL.Image.Image]):
+            List of image outputs - It can be a list of PIL images or a NumPy array or Torch tensor of shape
+            `(batch_size, channels, height, width)`. The tensor dimensions are automatically squeezed to remove
+            the temporal dimension for single-frame outputs.
+    """
+
+    images: torch.Tensor