Add custom clip type cogvideox

comfy/sd.py

@@ -64,6 +64,7 @@ import comfy.text_encoders.ace15
 import comfy.text_encoders.longcat_image
 import comfy.text_encoders.qwen35
 import comfy.text_encoders.ernie
+import comfy.text_encoders.cogvideo
 
 import comfy.model_patcher
 import comfy.lora
@@ -1201,6 +1202,7 @@ class CLIPType(Enum):
     NEWBIE = 24
     FLUX2 = 25
     LONGCAT_IMAGE = 26
+    COGVIDEOX = 27
 
 
 
@@ -1396,6 +1398,9 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
                 clip_target.clip = comfy.text_encoders.hidream.hidream_clip(**t5xxl_detect(clip_data),
                                                                         clip_l=False, clip_g=False, t5=True, llama=False, dtype_llama=None)
                 clip_target.tokenizer = comfy.text_encoders.hidream.HiDreamTokenizer
+            elif clip_type == CLIPType.COGVIDEOX:
+                clip_target.clip = comfy.text_encoders.cogvideo.cogvideo_te(**t5xxl_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.cogvideo.CogVideoXTokenizer
             else: #CLIPType.MOCHI
                 clip_target.clip = comfy.text_encoders.genmo.mochi_te(**t5xxl_detect(clip_data))
                 clip_target.tokenizer = comfy.text_encoders.genmo.MochiT5Tokenizer

comfy/text_encoders/cogvideo.py

@@ -1,6 +1,48 @@
 import comfy.text_encoders.sd3_clip
+from comfy import sd1_clip
 
 
 class CogVideoXT5Tokenizer(comfy.text_encoders.sd3_clip.T5XXLTokenizer):
+    """Inner T5 tokenizer for CogVideoX.
+
+    CogVideoX was trained with T5 embeddings padded to 226 tokens (not 77 like SD3).
+    Used both directly by supported_models.CogVideoX_T2V.clip_target (paired with
+    the raw T5XXLModel) and by the CogVideoXTokenizer outer wrapper below.
+    """
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, min_length=226)
+
+
+class CogVideoXTokenizer(sd1_clip.SD1Tokenizer):
+    """Outer tokenizer wrapper for CLIPLoader (type="cogvideox")."""
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data,
+                         clip_name="t5xxl", tokenizer=CogVideoXT5Tokenizer)
+
+
+class CogVideoXT5XXL(sd1_clip.SD1ClipModel):
+    """Outer T5XXL model wrapper for CLIPLoader (type="cogvideox").
+
+    Wraps the raw T5XXL model in the SD1ClipModel interface so that CLIP.__init__
+    (which reads self.dtypes) works correctly. The inner model is the standard
+    sd3_clip.T5XXLModel (no attention_mask change needed for CogVideoX).
+    """
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="t5xxl",
+                         clip_model=comfy.text_encoders.sd3_clip.T5XXLModel,
+                         model_options=model_options)
+
+
+def cogvideo_te(dtype_t5=None, t5_quantization_metadata=None):
+    """Factory that returns a CogVideoXT5XXL class configured with the detected
+    T5 dtype and optional quantization metadata, for use in load_text_encoder_state_dicts.
+    """
+    class CogVideoXTEModel_(CogVideoXT5XXL):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if t5_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["t5xxl_quantization_metadata"] = t5_quantization_metadata
+            if dtype_t5 is not None:
+                dtype = dtype_t5
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return CogVideoXTEModel_

nodes.py

@@ -977,7 +977,7 @@ class CLIPLoader:
     @classmethod
     def INPUT_TYPES(s):
         return {"required": { "clip_name": (folder_paths.get_filename_list("text_encoders"), ),
-                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image"], ),
+                              "type": (["stable_diffusion", "stable_cascade", "sd3", "stable_audio", "mochi", "ltxv", "pixart", "cosmos", "lumina2", "wan", "hidream", "chroma", "ace", "omnigen2", "qwen_image", "hunyuan_image", "flux2", "ovis", "longcat_image", "cogvideox"], ),
                               },
                 "optional": {
                               "device": (["default", "cpu"], {"advanced": True}),
@@ -987,7 +987,7 @@ class CLIPLoader:
 
     CATEGORY = "advanced/loaders"
 
-    DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 xxl/ clip-g / clip-l\nstable_audio: t5 base\nmochi: t5 xxl\ncosmos: old t5 xxl\nlumina2: gemma 2 2B\nwan: umt5 xxl\n hidream: llama-3.1 (Recommend) or t5\nomnigen2: qwen vl 2.5 3B"
+    DESCRIPTION = "[Recipes]\n\nstable_diffusion: clip-l\nstable_cascade: clip-g\nsd3: t5 xxl/ clip-g / clip-l\nstable_audio: t5 base\nmochi: t5 xxl\ncogvideox: t5 xxl (226-token padding)\ncosmos: old t5 xxl\nlumina2: gemma 2 2B\nwan: umt5 xxl\n hidream: llama-3.1 (Recommend) or t5\nomnigen2: qwen vl 2.5 3B"
 
     def load_clip(self, clip_name, type="stable_diffusion", device="default"):
         clip_type = getattr(comfy.sd.CLIPType, type.upper(), comfy.sd.CLIPType.STABLE_DIFFUSION)