Add Support for Z-Image.

Author: JerryWu-code

src/diffusers/__init__.py

@@ -271,6 +271,7 @@
             "WanAnimateTransformer3DModel",
             "WanTransformer3DModel",
             "WanVACETransformer3DModel",
+            "ZImageTransformer2DModel",
             "attention_backend",
         ]
     )
@@ -647,6 +648,7 @@
             "WuerstchenCombinedPipeline",
             "WuerstchenDecoderPipeline",
             "WuerstchenPriorPipeline",
+            "ZImagePipeline",
         ]
     )
 
@@ -1329,6 +1331,7 @@
             WuerstchenCombinedPipeline,
             WuerstchenDecoderPipeline,
             WuerstchenPriorPipeline,
+            ZImagePipeline,
         )
 
     try:

src/diffusers/hooks/_helpers.py

@@ -111,6 +111,7 @@ def _register_attention_processors_metadata():
     from ..models.transformers.transformer_hunyuanimage import HunyuanImageAttnProcessor
     from ..models.transformers.transformer_qwenimage import QwenDoubleStreamAttnProcessor2_0
     from ..models.transformers.transformer_wan import WanAttnProcessor2_0
+    from ..models.transformers.transformer_z_image import ZSingleStreamAttnProcessor
 
     # AttnProcessor2_0
     AttentionProcessorRegistry.register(
@@ -158,6 +159,14 @@ def _register_attention_processors_metadata():
         ),
     )
 
+    # ZSingleStreamAttnProcessor
+    AttentionProcessorRegistry.register(
+        model_class=ZSingleStreamAttnProcessor,
+        metadata=AttentionProcessorMetadata(
+            skip_processor_output_fn=_skip_proc_output_fn_Attention_ZSingleStreamAttnProcessor,
+        ),
+    )
+
 
 def _register_transformer_blocks_metadata():
     from ..models.attention import BasicTransformerBlock
@@ -179,6 +188,7 @@ def _register_transformer_blocks_metadata():
     from ..models.transformers.transformer_mochi import MochiTransformerBlock
     from ..models.transformers.transformer_qwenimage import QwenImageTransformerBlock
     from ..models.transformers.transformer_wan import WanTransformerBlock
+    from ..models.transformers.transformer_z_image import ZImageTransformerBlock
 
     # BasicTransformerBlock
     TransformerBlockRegistry.register(
@@ -312,6 +322,15 @@ def _register_transformer_blocks_metadata():
         ),
     )
 
+    # ZImage
+    TransformerBlockRegistry.register(
+        model_class=ZImageTransformerBlock,
+        metadata=TransformerBlockMetadata(
+            return_hidden_states_index=0,
+            return_encoder_hidden_states_index=None,
+        ),
+    )
+
 
 # fmt: off
 def _skip_attention___ret___hidden_states(self, *args, **kwargs):
@@ -338,4 +357,5 @@ def _skip_attention___ret___hidden_states___encoder_hidden_states(self, *args, *
 _skip_proc_output_fn_Attention_FluxAttnProcessor = _skip_attention___ret___hidden_states
 _skip_proc_output_fn_Attention_QwenDoubleStreamAttnProcessor2_0 = _skip_attention___ret___hidden_states
 _skip_proc_output_fn_Attention_HunyuanImageAttnProcessor = _skip_attention___ret___hidden_states
+_skip_proc_output_fn_Attention_ZSingleStreamAttnProcessor = _skip_attention___ret___hidden_states
 # fmt: on

src/diffusers/models/__init__.py

@@ -110,6 +110,7 @@
     _import_structure["transformers.transformer_wan"] = ["WanTransformer3DModel"]
     _import_structure["transformers.transformer_wan_animate"] = ["WanAnimateTransformer3DModel"]
     _import_structure["transformers.transformer_wan_vace"] = ["WanVACETransformer3DModel"]
+    _import_structure["transformers.transformer_z_image"] = ["ZImageTransformer2DModel"]
     _import_structure["unets.unet_1d"] = ["UNet1DModel"]
     _import_structure["unets.unet_2d"] = ["UNet2DModel"]
     _import_structure["unets.unet_2d_condition"] = ["UNet2DConditionModel"]
@@ -218,6 +219,7 @@
             WanAnimateTransformer3DModel,
             WanTransformer3DModel,
             WanVACETransformer3DModel,
+            ZImageTransformer2DModel,
         )
         from .unets import (
             I2VGenXLUNet,

src/diffusers/models/transformers/__init__.py

@@ -44,3 +44,4 @@
     from .transformer_wan import WanTransformer3DModel
     from .transformer_wan_animate import WanAnimateTransformer3DModel
     from .transformer_wan_vace import WanVACETransformer3DModel
+    from .transformer_z_image import ZImageTransformer2DModel