Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Allow ModelSamplingDiscrete to be instantiated without a model config.

comfyanonymousecb80abb58d5
diff --git a/comfy/model_base.py b/comfy/model_base.py
index 37a52debf..41d464e52 100644
--- a/comfy/model_base.py
+++ b/comfy/model_base.py
@@ -32,9 +32,12 @@ class V_PREDICTION(EPS):
class ModelSamplingDiscrete(torch.nn.Module):
- def __init__(self, model_config):
+ def __init__(self, model_config=None):
super().__init__()
- self._register_schedule(given_betas=None, beta_schedule=model_config.beta_schedule, timesteps=1000, linear_start=0.00085, linear_end=0.012, cosine_s=8e-3)
+ beta_schedule = "linear"
+ if model_config is not None:
+ beta_schedule = model_config.beta_schedule
+ self._register_schedule(given_betas=None, beta_schedule=beta_schedule, timesteps=1000, linear_start=0.00085, linear_end=0.012, cosine_s=8e-3)
self.sigma_data = 1.0
def _register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000,