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.

Add linear_start and linear_end to model_config.sampling_settings

comfyanonymousa4ec54a40d97
diff --git a/comfy/model_sampling.py b/comfy/model_sampling.py
index 69c8b1f01..cc8745c10 100644
--- a/comfy/model_sampling.py
+++ b/comfy/model_sampling.py
@@ -22,10 +22,17 @@ class V_PREDICTION(EPS):
class ModelSamplingDiscrete(torch.nn.Module):
def __init__(self, model_config=None):
super().__init__()
- beta_schedule = "linear"
+
if model_config is not None:
- beta_schedule = model_config.sampling_settings.get("beta_schedule", 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)
+ sampling_settings = model_config.sampling_settings
+ else:
+ sampling_settings = {}
+
+ beta_schedule = sampling_settings.get("beta_schedule", "linear")
+ linear_start = sampling_settings.get("linear_start", 0.00085)
+ linear_end = sampling_settings.get("linear_end", 0.012)
+
+ self._register_schedule(given_betas=None, beta_schedule=beta_schedule, timesteps=1000, linear_start=linear_start, linear_end=linear_end, cosine_s=8e-3)
self.sigma_data = 1.0
def _register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000,