Update.

Author: james77777778

keras_hub/src/models/stable_diffusion_3/stable_diffusion_3_text_to_image_preprocessor.py

@@ -50,8 +50,12 @@ def build(self, input_shape):
 
     def generate_preprocess(self, x):
         token_ids = {}
-        token_ids["clip_l"] = self.clip_l_preprocessor(x)["token_ids"]
-        token_ids["clip_g"] = self.clip_g_preprocessor(x)["token_ids"]
+        token_ids["clip_l"] = self.clip_l_preprocessor(
+            {"prompts": x, "images": None}
+        )["token_ids"]
+        token_ids["clip_g"] = self.clip_g_preprocessor(
+            {"prompts": x, "images": None}
+        )["token_ids"]
         if self.t5_preprocessor is not None:
             token_ids["t5"] = self.t5_preprocessor(x)["token_ids"]
         return token_ids

tools/checkpoint_conversion/convert_stable_diffusion_3_checkpoints.py

@@ -649,6 +649,9 @@ def validate_output(preset, keras_model, keras_preprocessor, output_dir):
     config = PRESET_MAP[preset]
     dtype = config["dtype"]
     hf_repo_id = config["root"].replace("hf://", "", 1)
+    if preset == "stable_diffusion_3_medium":
+        hf_repo_id += "-diffusers"
+
     if dtype == "float16":
         torch_dtype = torch.float16
     elif dtype == "bfloat16":