[GLM-Image] New Models Support

docs/source/en/_toctree.yml

@@ -353,6 +353,8 @@
         title: Flux2Transformer2DModel
       - local: api/models/flux_transformer
         title: FluxTransformer2DModel
+      - local: api/models/glm_image_transformer2d
+        title: GlmImageTransformer2DModel
       - local: api/models/hidream_image_transformer
         title: HiDreamImageTransformer2DModel
       - local: api/models/hunyuan_transformer2d
@@ -547,6 +549,8 @@
         title: Flux2
       - local: api/pipelines/control_flux_inpaint
         title: FluxControlInpaint
+      - local: api/pipelines/glm_image
+        title: GLM-Image
       - local: api/pipelines/hidream
         title: HiDream-I1
       - local: api/pipelines/hunyuandit

docs/source/en/api/models/glm_image_transformer2d.md

@@ -0,0 +1,18 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License. -->
+
+# GlmImageTransformer2DModel
+
+A Diffusion Transformer model for 2D data from [GlmImageTransformer2DModel]()
+
+## GlmImageTransformer2DModel
+
+[[autodoc]] GlmImageTransformer2DModel

docs/source/en/api/pipelines/glm_image.md

@@ -0,0 +1,31 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+-->
+
+# GLM-Image
+
+> [!TIP]
+> Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.
+
+This pipeline was contributed by [zRzRzRzRzRzRzR](https://github.com/zRzRzRzRzRzRzR). The original codebase can be found [here](https://huggingface.co/zai-org). The original weights can be found under [hf.co/zai-org](https://huggingface.co/zai-org).
+
+## GlmImagePipeline
+
+[[autodoc]] GlmImagePipeline
+  - all
+  - __call__
+
+## GlmImagePipelineOutput
+
+[[autodoc]] pipelines.cogview4.pipeline_output.GlmImagePipelineOutput

src/diffusers/__init__.py

@@ -225,6 +225,7 @@
             "FluxControlNetModel",
             "FluxMultiControlNetModel",
             "FluxTransformer2DModel",
+            "GlmImageTransformer2DModel",
             "HiDreamImageTransformer2DModel",
             "HunyuanDiT2DControlNetModel",
             "HunyuanDiT2DModel",
@@ -490,6 +491,7 @@
             "FluxKontextPipeline",
             "FluxPipeline",
             "FluxPriorReduxPipeline",
+            "GlmImagePipeline",
             "HiDreamImagePipeline",
             "HunyuanDiTControlNetPipeline",
             "HunyuanDiTPAGPipeline",
@@ -977,6 +979,7 @@
             FluxControlNetModel,
             FluxMultiControlNetModel,
             FluxTransformer2DModel,
+            GlmImageTransformer2DModel,
             HiDreamImageTransformer2DModel,
             HunyuanDiT2DControlNetModel,
             HunyuanDiT2DModel,
@@ -1212,6 +1215,7 @@
             FluxKontextPipeline,
             FluxPipeline,
             FluxPriorReduxPipeline,
+            GlmImagePipeline,
             HiDreamImagePipeline,
             HunyuanDiTControlNetPipeline,
             HunyuanDiTPAGPipeline,

src/diffusers/models/__init__.py

@@ -98,6 +98,7 @@
     _import_structure["transformers.transformer_easyanimate"] = ["EasyAnimateTransformer3DModel"]
     _import_structure["transformers.transformer_flux"] = ["FluxTransformer2DModel"]
     _import_structure["transformers.transformer_flux2"] = ["Flux2Transformer2DModel"]
+    _import_structure["transformers.transformer_glm_image"] = ["GlmImageTransformer2DModel"]
     _import_structure["transformers.transformer_hidream_image"] = ["HiDreamImageTransformer2DModel"]
     _import_structure["transformers.transformer_hunyuan_video"] = ["HunyuanVideoTransformer3DModel"]
     _import_structure["transformers.transformer_hunyuan_video15"] = ["HunyuanVideo15Transformer3DModel"]
@@ -208,6 +209,7 @@
             EasyAnimateTransformer3DModel,
             Flux2Transformer2DModel,
             FluxTransformer2DModel,
+            GlmImageTransformer2DModel,
             HiDreamImageTransformer2DModel,
             HunyuanDiT2DModel,
             HunyuanImageTransformer2DModel,

src/diffusers/models/embeddings.py

@@ -1658,6 +1658,37 @@ def forward(
         return conditioning
 
 
+class GlmImageCombinedTimestepSizeEmbeddings(nn.Module):
+    def __init__(self, embedding_dim: int, condition_dim: int, pooled_projection_dim: int, timesteps_dim: int = 256):
+        super().__init__()
+
+        self.time_proj = Timesteps(num_channels=timesteps_dim, flip_sin_to_cos=True, downscale_freq_shift=0)
+        self.condition_proj = Timesteps(num_channels=condition_dim, flip_sin_to_cos=True, downscale_freq_shift=0)
+        self.timestep_embedder = TimestepEmbedding(in_channels=timesteps_dim, time_embed_dim=embedding_dim)
+        self.condition_embedder = PixArtAlphaTextProjection(pooled_projection_dim, embedding_dim, act_fn="silu")
+
+    def forward(
+        self,
+        timestep: torch.Tensor,
+        target_size: torch.Tensor,
+        crop_coords: torch.Tensor,
+        hidden_dtype: torch.dtype,
+    ) -> torch.Tensor:
+        timesteps_proj = self.time_proj(timestep)
+
+        crop_coords_proj = self.condition_proj(crop_coords.flatten()).view(crop_coords.size(0), -1)
+        target_size_proj = self.condition_proj(target_size.flatten()).view(target_size.size(0), -1)
+
+        # (B, 2 * condition_dim)
+        condition_proj = torch.cat([crop_coords_proj, target_size_proj], dim=1)
+
+        timesteps_emb = self.timestep_embedder(timesteps_proj.to(dtype=hidden_dtype))  # (B, embedding_dim)
+        condition_emb = self.condition_embedder(condition_proj.to(dtype=hidden_dtype))  # (B, embedding_dim)
+
+        conditioning = timesteps_emb + condition_emb
+        return conditioning
+
+
 class HunyuanDiTAttentionPool(nn.Module):
     # Copied from https://github.com/Tencent/HunyuanDiT/blob/cb709308d92e6c7e8d59d0dff41b74d35088db6a/hydit/modules/poolers.py#L6
 

src/diffusers/models/transformers/__init__.py

@@ -27,6 +27,7 @@
     from .transformer_easyanimate import EasyAnimateTransformer3DModel
     from .transformer_flux import FluxTransformer2DModel
     from .transformer_flux2 import Flux2Transformer2DModel
+    from .transformer_glm_image import GlmImageTransformer2DModel
     from .transformer_hidream_image import HiDreamImageTransformer2DModel
     from .transformer_hunyuan_video import HunyuanVideoTransformer3DModel
     from .transformer_hunyuan_video15 import HunyuanVideo15Transformer3DModel