feat: add parameter for ControlNet selection

api/onnx_web/chain/blend_controlnet.py

@@ -28,7 +28,7 @@ def blend_controlnet(
     params = params.with_args(**kwargs)
     source = stage_source or source
     logger.info(
-        "blending image using controlnet, %s steps: %s", params.steps, params.prompt
+        "blending image using ControlNet, %s steps: %s", params.steps, params.prompt
     )
 
     pipe = load_pipeline(

api/onnx_web/convert/diffusion/diffusers.py

@@ -174,7 +174,6 @@ def convert_diffusion_diffusers(
     if is_torch_2_0:
         pipe_cnet.set_attn_processor(CrossAttnProcessor())
 
-
     cnet_path = output_path / "cnet" / ONNX_MODEL
     onnx_export(
         pipe_cnet,

api/onnx_web/diffusers/load.py

@@ -116,14 +116,13 @@ def load_pipeline(
     scheduler_name: str,
     device: DeviceParams,
     lpw: bool,
+    control: Optional[str] = None,
     inversions: Optional[List[Tuple[str, float]]] = None,
     loras: Optional[List[Tuple[str, float]]] = None,
 ):
     inversions = inversions or []
     loras = loras or []
 
-    controlnet = "canny"  # TODO; from params
-
     torch_dtype = (
         torch.float16 if "torch-fp16" in server.optimizations else torch.float32
     )
@@ -186,6 +185,8 @@ def load_pipeline(
         }
 
         text_encoder = None
+
+        # Textual Inversion blending
         if inversions is not None and len(inversions) > 0:
             logger.debug("blending Textual Inversions from %s", inversions)
             inversion_names, inversion_weights = zip(*inversions)
@@ -225,7 +226,7 @@ def load_pipeline(
                     )
                 )
 
-        # test LoRA blending
+        # LoRA blending
         if loras is not None and len(loras) > 0:
             lora_names, lora_weights = zip(*loras)
             lora_models = [
@@ -278,8 +279,12 @@ def load_pipeline(
                 )
             )
 
-        if controlnet is not None:
-            components["controlnet"] = OnnxRuntimeModel.from_pretrained(controlnet)
+        if control is not None:
+            components["controlnet"] = OnnxRuntimeModel(OnnxRuntimeModel.load_model(
+                control,
+                provider=device.ort_provider(),
+                sess_options=device.sess_options(),
+            ))
 
         pipe = pipeline.from_pretrained(
             model,
@@ -360,7 +365,7 @@ def __init__(self, server, wrapped):
         self.server = server
         self.wrapped = wrapped
 
-    def __call__(self, sample=None, timestep=None, encoder_hidden_states=None):
+    def __call__(self, sample=None, timestep=None, encoder_hidden_states=None, **kwargs):
         global timestep_dtype
         timestep_dtype = timestep.dtype
 
@@ -382,6 +387,7 @@ def __call__(self, sample=None, timestep=None, encoder_hidden_states=None):
             sample=sample,
             timestep=timestep,
             encoder_hidden_states=encoder_hidden_states,
+            **kwargs,
         )
 
     def __getattr__(self, attr):
@@ -393,15 +399,15 @@ def __init__(self, server, wrapped):
         self.server = server
         self.wrapped = wrapped
 
-    def __call__(self, latent_sample=None):
+    def __call__(self, latent_sample=None, **kwargs):
         global timestep_dtype
 
         logger.trace("VAE parameter types: %s", latent_sample.dtype)
         if latent_sample.dtype != timestep_dtype:
             logger.info("converting VAE sample dtype")
             latent_sample = latent_sample.astype(timestep_dtype)
 
-        return self.wrapped(latent_sample=latent_sample)
+        return self.wrapped(latent_sample=latent_sample, **kwargs)
 
     def __getattr__(self, attr):
         return getattr(self.wrapped, attr)

api/onnx_web/diffusers/run.py

@@ -239,8 +239,9 @@ def run_img2img_pipeline(
         params.scheduler,
         job.get_device(),
         params.lpw,
-        inversions,
-        loras,
+        control=params.control,
+        inversions=inversions,
+        loras=loras,
     )
     progress = job.get_progress_callback()
     if params.lpw:

api/onnx_web/params.py

@@ -160,6 +160,18 @@ def torch_str(self) -> str:
 
 
 class ImageParams:
+    model: str
+    scheduler: str
+    prompt: str
+    cfg: float
+    steps: int
+    seed: int
+    negative_prompt: Optional[str]
+    lpw: bool
+    eta: float
+    batch: int
+    control: Optional[str]
+
     def __init__(
         self,
         model: str,
@@ -172,6 +184,7 @@ def __init__(
         lpw: bool = False,
         eta: float = 0.0,
         batch: int = 1,
+        control: Optional[str] = None,
     ) -> None:
         self.model = model
         self.scheduler = scheduler
@@ -183,6 +196,7 @@ def __init__(
         self.lpw = lpw or False
         self.eta = eta
         self.batch = batch
+        self.control = control
 
     def tojson(self) -> Dict[str, Optional[Param]]:
         return {
@@ -196,6 +210,7 @@ def tojson(self) -> Dict[str, Optional[Param]]:
             "lpw": self.lpw,
             "eta": self.eta,
             "batch": self.batch,
+            "control": self.control,
         }
 
     def with_args(self, **kwargs):
@@ -210,6 +225,7 @@ def with_args(self, **kwargs):
             kwargs.get("lpw", self.lpw),
             kwargs.get("eta", self.eta),
             kwargs.get("batch", self.batch),
+            kwargs.get("control", self.control),
         )
 
 

api/onnx_web/server/params.py

@@ -42,6 +42,7 @@ def pipeline_from_request(
                 device = platform
 
     # pipeline stuff
+    control = get_not_empty(request.args, "control", get_config_value("control"))
     lpw = get_not_empty(request.args, "lpw", "false") == "true"
     model = get_not_empty(request.args, "model", get_config_value("model"))
     model_path = get_model_path(server, model)
@@ -132,6 +133,7 @@ def pipeline_from_request(
         lpw=lpw,
         negative_prompt=negative_prompt,
         batch=batch,
+        control=control,
     )
     size = Size(width, height)
     return (device, params, size)

api/params.json

@@ -18,6 +18,10 @@
     "max": 30,
     "step": 0.1
   },
+  "control": {
+    "default": "",
+    "keys": []
+  },
   "correction": {
     "default": "",
     "keys": []

gui/examples/config.json

@@ -22,6 +22,10 @@
       "max": 30,
       "step": 0.1
     },
+    "control": {
+      "default": "",
+      "keys": []
+    },
     "correction": {
       "default": "",
       "keys": []

gui/src/client/api.ts

@@ -32,6 +32,11 @@ export interface ModelParams {
    * Use the long prompt weighting pipeline.
    */
   lpw: boolean;
+
+  /**
+   * ControlNet to be used.
+   */
+  control: string;
 }
 
 /**
@@ -191,7 +196,7 @@ export interface ReadyResponse {
 
 export interface NetworkModel {
   name: string;
-  type: 'inversion' | 'lora';
+  type: 'control' | 'inversion' | 'lora';
   // TODO: add token
   // TODO: add layer/token count
 }
@@ -392,6 +397,7 @@ export function appendModelToURL(url: URL, params: ModelParams) {
   url.searchParams.append('upscaling', params.upscaling);
   url.searchParams.append('correction', params.correction);
   url.searchParams.append('lpw', String(params.lpw));
+  url.searchParams.append('control', params.control);
 }
 
 /**

gui/src/components/control/ModelControl.tsx

@@ -116,6 +116,20 @@ export function ModelControl() {
           });
         }}
       />
+      <QueryMenu
+        id='control'
+        labelKey='model'
+        name={t('modelType.control')}
+        query={{
+          result: models,
+          selector: (result) => result.networks.filter((network) => network.type === 'control').map((network) => network.name),
+        }}
+        onSelect={(control) => {
+          setModel({
+            control,
+          });
+        }}
+      />
     </Stack>
     <Stack direction='row' spacing={2}>
       <FormControlLabel

gui/src/state.ts

@@ -500,11 +500,12 @@ export function createStateSlices(server: ServerParams) {
 
   const createModelSlice: Slice<ModelSlice> = (set) => ({
     model: {
+      control: server.control.default,
+      correction: server.correction.default,
+      lpw: false,
       model: server.model.default,
       platform: server.platform.default,
       upscaling: server.upscaling.default,
-      correction: server.correction.default,
-      lpw: false,
     },
     setModel(params) {
       set((prev) => ({