fix inpainting conversion via torch

api/onnx_web/convert/diffusion/checkpoint.py

@@ -1377,6 +1377,7 @@ def extract_checkpoint(
     conversion: ConversionContext,
     new_model_name: str,
     checkpoint_file: str,
+    is_inpainting: Optional[bool] = False,
     scheduler_type="ddim",
     extract_ema=False,
     train_unfrozen=False,
@@ -1537,6 +1538,9 @@ def extract_checkpoint(
             original_config, image_size=image_size
         )
         unet_config["upcast_attention"] = upcast_attention
+        if is_inpainting:
+            unet_config["in_channels "] = 9
+            
         unet = UNet2DConditionModel(**unet_config)
 
         converted_unet_checkpoint, has_ema = convert_ldm_unet_checkpoint(
@@ -1685,6 +1689,7 @@ def convert_extract_checkpoint(
     conversion: ConversionContext,
     source: str,
     dest: str,
+    is_inpainting: Optional[bool] = False,
     config_file: Optional[str] = None,
     vae_file: Optional[str] = None,
 ) -> Tuple[bool, str]:

api/onnx_web/convert/diffusion/diffusers.py

@@ -312,9 +312,11 @@ def convert_diffusion_diffusers(
     v2, pipe_args = get_model_version(
         source, conversion.map_location, size=image_size, version=version
     )
-
+    
+    is_inpainting = False
     if pipe_type == "inpaint":
         pipe_args["num_in_channels"] = 9
+        is_inpainting = True
 
     if format == "safetensors":
         pipe_args["from_safetensors"] = True
@@ -334,6 +336,7 @@ def convert_diffusion_diffusers(
                 conversion,
                 source,
                 f"{name}-torch",
+                is_inpainting=is_inpainting
                 config_file=config,
                 vae_file=replace_vae,
             )