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 CFGZeroStar node.

comfyanonymous84fdaf7b0ef4
diff --git a/comfy_extras/nodes_cfg.py b/comfy_extras/nodes_cfg.py
new file mode 100644
index 000000000..1fb686644
--- /dev/null
+++ b/comfy_extras/nodes_cfg.py
@@ -0,0 +1,45 @@
+import torch
+
+# https://github.com/WeichenFan/CFG-Zero-star
+def optimized_scale(positive, negative):
+ positive_flat = positive.reshape(positive.shape[0], -1)
+ negative_flat = negative.reshape(negative.shape[0], -1)
+
+ # Calculate dot production
+ dot_product = torch.sum(positive_flat * negative_flat, dim=1, keepdim=True)
+
+ # Squared norm of uncondition
+ squared_norm = torch.sum(negative_flat ** 2, dim=1, keepdim=True) + 1e-8
+
+ # st_star = v_cond^T * v_uncond / ||v_uncond||^2
+ st_star = dot_product / squared_norm
+
+ return st_star.reshape([positive.shape[0]] + [1] * (positive.ndim - 1))
+
+class CFGZeroStar:
+ @classmethod
+ def INPUT_TYPES(s):
+ return {"required": {"model": ("MODEL",),
+ }}
+ RETURN_TYPES = ("MODEL",)
+ RETURN_NAMES = ("patched_model",)
+ FUNCTION = "patch"
+ CATEGORY = "advanced/guidance"
+
+ def patch(self, model):
+ m = model.clone()
+ def cfg_zero_star(args):
+ guidance_scale = args['cond_scale']
+ x = args['input']
+ cond_p = args['cond_denoised']
+ uncond_p = args['uncond_denoised']
+ out = args["denoised"]
+ alpha = optimized_scale(x - cond_p, x - uncond_p)
+
+ return out + uncond_p * (alpha - 1.0) + guidance_scale * uncond_p * (1.0 - alpha)
+ m.set_model_sampler_post_cfg_function(cfg_zero_star)
+ return (m, )
+
+NODE_CLASS_MAPPINGS = {
+ "CFGZeroStar": CFGZeroStar
+}
diff --git a/nodes.py b/nodes.py
index 27ef743b3..272c2a25e 100644
--- a/nodes.py
+++ b/nodes.py
@@ -2267,6 +2267,7 @@ def init_builtin_extra_nodes():
"nodes_lotus.py",
"nodes_hunyuan3d.py",
"nodes_primitive.py",
+ "nodes_cfg.py",
]
import_failed = []