[Modular]z-image (#12808)

* initiL

* up up

* fix: z_image -> z-image

* style

* copy

* fix more

* some docstring fix

Author: yiyixuxu

src/diffusers/__init__.py

@@ -419,6 +419,8 @@
             "Wan22AutoBlocks",
             "WanAutoBlocks",
             "WanModularPipeline",
+            "ZImageAutoBlocks",
+            "ZImageModularPipeline",
         ]
     )
     _import_structure["pipelines"].extend(
@@ -1124,6 +1126,8 @@
             Wan22AutoBlocks,
             WanAutoBlocks,
             WanModularPipeline,
+            ZImageAutoBlocks,
+            ZImageModularPipeline,
         )
         from .pipelines import (
             AllegroPipeline,

src/diffusers/modular_pipelines/__init__.py

@@ -60,6 +60,10 @@
         "QwenImageEditPlusModularPipeline",
         "QwenImageEditPlusAutoBlocks",
     ]
+    _import_structure["z_image"] = [
+        "ZImageAutoBlocks",
+        "ZImageModularPipeline",
+    ]
     _import_structure["components_manager"] = ["ComponentsManager"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
@@ -91,6 +95,7 @@
         )
         from .stable_diffusion_xl import StableDiffusionXLAutoBlocks, StableDiffusionXLModularPipeline
         from .wan import Wan22AutoBlocks, WanAutoBlocks, WanModularPipeline
+        from .z_image import ZImageAutoBlocks, ZImageModularPipeline
 else:
     import sys
 

src/diffusers/modular_pipelines/modular_pipeline.py

@@ -61,6 +61,7 @@
         ("qwenimage", "QwenImageModularPipeline"),
         ("qwenimage-edit", "QwenImageEditModularPipeline"),
         ("qwenimage-edit-plus", "QwenImageEditPlusModularPipeline"),
+        ("z-image", "ZImageModularPipeline"),
     ]
 )
 

src/diffusers/modular_pipelines/wan/encoders.py

@@ -530,6 +530,7 @@ def __call__(self, components: WanModularPipeline, state: PipelineState) -> Pipe
 
         device = components._execution_device
         dtype = torch.float32
+        vae_dtype = components.vae.dtype
 
         height = block_state.height or components.default_height
         width = block_state.width or components.default_width
@@ -555,7 +556,7 @@ def __call__(self, components: WanModularPipeline, state: PipelineState) -> Pipe
             vae=components.vae,
             generator=block_state.generator,
             device=device,
-            dtype=dtype,
+            dtype=vae_dtype,
             latent_channels=components.num_channels_latents,
         )
 
@@ -627,6 +628,7 @@ def __call__(self, components: WanModularPipeline, state: PipelineState) -> Pipe
 
         device = components._execution_device
         dtype = torch.float32
+        vae_dtype = components.vae.dtype
 
         height = block_state.height or components.default_height
         width = block_state.width or components.default_width
@@ -659,7 +661,7 @@ def __call__(self, components: WanModularPipeline, state: PipelineState) -> Pipe
             vae=components.vae,
             generator=block_state.generator,
             device=device,
-            dtype=dtype,
+            dtype=vae_dtype,
             latent_channels=components.num_channels_latents,
         )
 

src/diffusers/modular_pipelines/z_image/__init__.py

@@ -0,0 +1,57 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["decoders"] = ["ZImageVaeDecoderStep"]
+    _import_structure["encoders"] = ["ZImageTextEncoderStep", "ZImageVaeImageEncoderStep"]
+    _import_structure["modular_blocks"] = [
+        "ALL_BLOCKS",
+        "ZImageAutoBlocks",
+    ]
+    _import_structure["modular_pipeline"] = ["ZImageModularPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
+    else:
+        from .decoders import ZImageVaeDecoderStep
+        from .encoders import ZImageTextEncoderStep
+        from .modular_blocks import (
+            ALL_BLOCKS,
+            ZImageAutoBlocks,
+        )
+        from .modular_pipeline import ZImageModularPipeline
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)