Merge branch 'master' into beta

comfy/controlnet.py

@@ -132,6 +132,7 @@ def __init__(self, control_model, global_average_pooling=False, device=None):
         self.control_model = control_model
         self.control_model_wrapped = comfy.model_patcher.ModelPatcher(self.control_model, load_device=comfy.model_management.get_torch_device(), offload_device=comfy.model_management.unet_offload_device())
         self.global_average_pooling = global_average_pooling
+        self.model_sampling_current = None
 
     def get_control(self, x_noisy, t, cond, batched_number):
         control_prev = None
@@ -159,7 +160,10 @@ def get_control(self, x_noisy, t, cond, batched_number):
         y = cond.get('y', None)
         if y is not None:
             y = y.to(self.control_model.dtype)
-        control = self.control_model(x=x_noisy.to(self.control_model.dtype), hint=self.cond_hint, timesteps=t, context=context.to(self.control_model.dtype), y=y)
+        timestep = self.model_sampling_current.timestep(t)
+        x_noisy = self.model_sampling_current.calculate_input(t, x_noisy)
+
+        control = self.control_model(x=x_noisy.to(self.control_model.dtype), hint=self.cond_hint, timesteps=timestep.float(), context=context.to(self.control_model.dtype), y=y)
         return self.control_merge(None, control, control_prev, output_dtype)
 
     def copy(self):
@@ -172,6 +176,14 @@ def get_models(self):
         out.append(self.control_model_wrapped)
         return out
 
+    def pre_run(self, model, percent_to_timestep_function):
+        super().pre_run(model, percent_to_timestep_function)
+        self.model_sampling_current = model.model_sampling
+
+    def cleanup(self):
+        self.model_sampling_current = None
+        super().cleanup()
+
 class ControlLoraOps:
     class Linear(torch.nn.Module):
         def __init__(self, in_features: int, out_features: int, bias: bool = True,

comfy/extra_samplers/uni_pc.py

@@ -852,6 +852,12 @@ def marginal_lambda(self, t):
         log_std = 0.5 * torch.log(1. - torch.exp(2. * log_mean_coeff))
         return log_mean_coeff - log_std
 
+def predict_eps_sigma(model, input, sigma_in, **kwargs):
+    sigma = sigma_in.view(sigma_in.shape[:1] + (1,) * (input.ndim - 1))
+    input = input * ((sigma ** 2 + 1.0) ** 0.5)
+    return  (input - model(input, sigma_in, **kwargs)) / sigma
+
+
 def sample_unipc(model, noise, image, sigmas, sampling_function, max_denoise, extra_args=None, callback=None, disable=False, noise_mask=None, variant='bh1'):
         timesteps = sigmas.clone()
         if sigmas[-1] == 0:
@@ -874,7 +880,7 @@ def sample_unipc(model, noise, image, sigmas, sampling_function, max_denoise, ex
         model_type = "noise"
 
         model_fn = model_wrapper(
-            model.predict_eps_sigma,
+            lambda input, sigma, **kwargs: predict_eps_sigma(model, input, sigma, **kwargs),
             ns,
             model_type=model_type,
             guidance_type="uncond",