Add files via upload

nodes.py

@@ -6,7 +6,7 @@ class uncondZeroNode:
         return {"required": {
                                 "model": ("MODEL",),
                                 "scale": ("FLOAT", {"default": 1, "min": 0.0, "max": 10.0, "step": 0.01, "round": 0.01}),
-                                "method":(["uncond_zero","rescale_cfg"],),
+                                "method":(["uncond_zero_v1","uncond_zero_v2","uncond_zero_v3"], {"default": "uncond_zero_v3"},),
                               }}
     RETURN_TYPES = ("MODEL",)
     FUNCTION = "patch"
@@ -19,10 +19,27 @@ class uncondZeroNode:
             x_orig = args["input"]
             x_orig -= x_orig.mean()
             cond   -= cond.mean()
-            return x_orig - (cond / cond.std() ** .5) * scale
-        
+            return x_orig - cond / cond.std() ** .5 * scale
+
+        def uncond_zero_v2(args):
+            cond   = args["cond_denoised"]
+            x_orig = args["input"]
+            cond  -= cond.mean()
+            result = torch.zeros_like(x_orig)
+
+            for b in range(len(x_orig)):
+                for c in range(len(x_orig[b])):
+                    x_orig[b][c] -= x_orig[b][c].mean()
+                    cond_c_mean   = cond[b][c].mean()
+                    cond[b][c]   -= cond_c_mean
+                    result[b][c]  = x_orig[b][c] - cond[b][c] / cond[b][c].std() ** .5 * scale + cond_c_mean
+
+            return result
+
+
         new_scale = 1 / (model.model.latent_format.scale_factor * 8)
-        #Taken and adapted from comfy_extras/nodes_model_advanced
+
+        #Taken and modified from comfy_extras/nodes_model_advanced
         def rescale_cfg(args):
             x_orig = args["input"]
             x_orig -= x_orig.mean()
@@ -30,7 +47,7 @@ class uncondZeroNode:
             cond = args["cond_denoised"]
             cond -= cond.mean()
 
-            cond = x_orig - (cond / cond.std() ** .5) * scale
+            cond = x_orig - cond / cond.std() ** .5
 
             sigma = args["sigma"]
             sigma = sigma.view(sigma.shape[:1] + (1,) * (cond.ndim - 1))
@@ -41,7 +58,7 @@ class uncondZeroNode:
             cond = ((x - (x_orig - cond)) * (sigma ** 2 + 1.0) ** 0.5) / (sigma)
 
             #rescalecfg
-            x_cfg = uncond + new_scale * (cond - uncond)
+            x_cfg = uncond + new_scale * scale * (cond - uncond)
             ro_pos = torch.std(cond, dim=(1,2,3), keepdim=True)
             ro_cfg = torch.std(x_cfg, dim=(1,2,3), keepdim=True)
 
@@ -50,5 +67,5 @@ class uncondZeroNode:
             return x_orig - (x - x_rescaled * sigma / (sigma * sigma + 1.0) ** 0.5)
 
         m = model.clone()
-        m.set_model_sampler_cfg_function({"uncond_zero":uncond_zero,"rescale_cfg":rescale_cfg}[method])
+        m.set_model_sampler_cfg_function({"uncond_zero_v1":uncond_zero,"uncond_zero_v2":uncond_zero_v2,"uncond_zero_v3":rescale_cfg}[method])
         return (m, )