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.pynew file mode 100644index 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.pyindex 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 = []