skip loading component models for XL

api/onnx_web/chain/blend_img2img.py

@@ -51,20 +51,20 @@ class BlendImg2ImgStage(BaseStage):
         )
 
         pipe_params = {}
-        if pipe_type == "controlnet":
+        if params.is_control():
             pipe_params["controlnet_conditioning_scale"] = strength
+        elif params.is_lpw():
+            pipe_params["strength"] = strength
+        elif params.is_panorama():
+            pipe_params["strength"] = strength
         elif pipe_type == "img2img":
             pipe_params["strength"] = strength
-        elif pipe_type == "lpw":
-            pipe_params["strength"] = strength
-        elif pipe_type == "panorama":
-            pipe_params["strength"] = strength
         elif pipe_type == "pix2pix":
             pipe_params["image_guidance_scale"] = strength
 
         outputs = []
         for source in sources:
-            if params.lpw():
+            if params.is_lpw():
                 logger.debug("using LPW pipeline for img2img")
                 rng = torch.manual_seed(params.seed)
                 result = pipe.img2img(
@@ -82,7 +82,9 @@ class BlendImg2ImgStage(BaseStage):
                 prompt_embeds = encode_prompt(
                     pipe, prompt_pairs, params.batch, params.do_cfg()
                 )
-                pipe.unet.set_prompts(prompt_embeds)
+
+                if not params.xl():
+                    pipe.unet.set_prompts(prompt_embeds)
 
                 rng = np.random.RandomState(params.seed)
                 result = pipe(

api/onnx_web/chain/source_txt2img.py

@@ -74,7 +74,7 @@ class SourceTxt2ImgStage(BaseStage):
             loras=loras,
         )
 
-        if params.lpw():
+        if params.is_lpw():
             logger.debug("using LPW pipeline for txt2img")
             rng = torch.manual_seed(params.seed)
             result = pipe.text2img(
@@ -95,7 +95,9 @@ class SourceTxt2ImgStage(BaseStage):
             prompt_embeds = encode_prompt(
                 pipe, prompt_pairs, params.batch, params.do_cfg()
             )
-            pipe.unet.set_prompts(prompt_embeds)
+
+            if not params.is_xl():
+                pipe.unet.set_prompts(prompt_embeds)
 
             rng = np.random.RandomState(params.seed)
             result = pipe(

api/onnx_web/chain/upscale_outpaint.py

@@ -81,7 +81,7 @@ class UpscaleOutpaintStage(BaseStage):
             else:
                 latents = get_tile_latents(latents, params.seed, latent_size, dims)
 
-            if params.lpw():
+            if params.is_lpw():
                 logger.debug("using LPW pipeline for inpaint")
                 rng = torch.manual_seed(params.seed)
                 result = pipe.inpaint(

api/onnx_web/diffusers/load.py

@@ -171,7 +171,7 @@ def load_pipeline(
         unet_type = "unet"
 
         # ControlNet component
-        if pipeline == "controlnet" and params.control is not None:
+        if params.is_control() and params.control is not None:
             cnet_path = path.join(
                 server.model_path, "control", f"{params.control.name}.onnx"
             )
@@ -282,7 +282,7 @@ def load_pipeline(
             )
 
         # make sure a UNet has been loaded
-        if "unet" not in components:
+        if not params.is_xl() and "unet" not in components:
             unet = path.join(model, unet_type, ONNX_MODEL)
             logger.debug("loading UNet (%s) from %s", unet_type, unet)
             components["unet"] = OnnxRuntimeModel(
@@ -298,7 +298,7 @@ def load_pipeline(
         vae_decoder = path.join(model, "vae_decoder", ONNX_MODEL)
         vae_encoder = path.join(model, "vae_encoder", ONNX_MODEL)
 
-        if path.exists(vae):
+        if not params.is_xl() and path.exists(vae):
             logger.debug("loading VAE from %s", vae)
             components["vae"] = OnnxRuntimeModel(
                 OnnxRuntimeModel.load_model(
@@ -307,7 +307,9 @@ def load_pipeline(
                     sess_options=device.sess_options(),
                 )
             )
-        elif path.exists(vae_decoder) and path.exists(vae_encoder):
+        elif (
+            not params.is_xl() and path.exists(vae_decoder) and path.exists(vae_encoder)
+        ):
             logger.debug("loading VAE decoder from %s", vae_decoder)
             components["vae_decoder"] = OnnxRuntimeModel(
                 OnnxRuntimeModel.load_model(
@@ -327,7 +329,7 @@ def load_pipeline(
             )
 
         # additional options for panorama pipeline
-        if pipeline == "panorama":
+        if params.is_panorama():
             components["window"] = params.tiles // 8
             components["stride"] = params.stride // 8
 
@@ -346,18 +348,21 @@ def load_pipeline(
             pipe.set_progress_bar_config(disable=True)
 
         optimize_pipeline(server, pipe)
-        patch_pipeline(server, pipe, pipeline, pipeline_class, params)
+
+        if not params.is_xl():
+            patch_pipeline(server, pipe, pipeline, pipeline_class, params)
 
         server.cache.set(ModelTypes.diffusion, pipe_key, pipe)
         server.cache.set(ModelTypes.scheduler, scheduler_key, components["scheduler"])
 
-    if hasattr(pipe, "vae_decoder"):
+    if not params.is_xl() and hasattr(pipe, "vae_decoder"):
         pipe.vae_decoder.set_tiled(tiled=params.tiled_vae)
-    if hasattr(pipe, "vae_encoder"):
+
+    if not params.is_xl() and hasattr(pipe, "vae_encoder"):
         pipe.vae_encoder.set_tiled(tiled=params.tiled_vae)
 
     # update panorama params
-    if pipeline == "panorama":
+    if params.is_panorama():
         latent_window = params.tiles // 8
         latent_stride = params.stride // 8
 

api/onnx_web/diffusers/run.py

@@ -43,8 +43,7 @@ def run_txt2img_pipeline(
     highres: HighresParams,
 ) -> None:
     # if using panorama, the pipeline will tile itself (views)
-    pipe_type = params.get_valid_pipeline("txt2img")
-    if pipe_type == "panorama":
+    if params.is_panorama():
         tile_size = max(params.tiles, size.width, size.height)
     else:
         tile_size = params.tiles

api/onnx_web/params.py

@@ -271,9 +271,18 @@ class ImageParams:
         logger.debug("pipeline %s is not valid for %s", pipeline, group)
         return group
 
-    def lpw(self):
+    def is_control(self):
+        return self.pipeline == "controlnet"
+
+    def is_lpw(self):
         return self.pipeline == "lpw"
 
+    def is_panorama(self):
+        return self.pipeline == "panorama"
+
+    def is_xl(self):
+        return self.pipeline.endswith("-sdxl")
+
     def tojson(self) -> Dict[str, Optional[Param]]:
         return {
             "model": self.model,