Publish v0.1.0

Add Github Action for Publishing to Comfy Registry

.github/workflows/publish.yml

@@ -0,0 +1,21 @@
+name: Publish to Comfy registry
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - master
+    paths:
+      - "pyproject.toml"
+
+jobs:
+  publish-node:
+    name: Publish Custom Node to registry
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Publish Custom Node
+        uses: Comfy-Org/publish-node-action@main
+        with:
+          ## Add your own personal access token to your Github Repository secrets and reference it here.
+          personal_access_token: ${{ secrets.REGISTRY_ACCESS_TOKEN }}

README.md

@@ -11,3 +11,11 @@ There's an `AdaptiveGuidance` node (under `sampling/custom_sampling/guiders`) th
 The node calculates the cosine similarity between the u-net's conditional and unconditional output ("positive" and "negative" prompts) and once the similarity crosses the specified threshold, it sets CFG to 1.0, effectively skipping negative prompt calculations and speeding up inference.
 
 I'm not sure if the cosine similarity calculation matches the original paper since I had to translate from maths to Python, but it appears to work.
+
+### Uncond zero
+
+Set uncond_zero_scale to > 0 to enable "uncond zero" CFG *after* the normal CFG gets disabled. Stolen from https://github.com/Extraltodeus/Uncond-Zero-for-ComfyUI
+
+It seems to work slightly better than just running without CFG, but YMMV
+
+Note: this functionality is unstable and will probably change, so using it means your workflows likely won't be perfectly reproducible.

__init__.py

@@ -6,40 +6,56 @@ cos = torch.nn.CosineSimilarity(dim=1)
 
 class AdaptiveGuider(comfy.samplers.CFGGuider):
     threshold_timestep = 0
+    uz_scale = 0.0
 
-    def set_cfg(self, cfg, threshold):
+    def set_cfg(self, cfg):
         self.cfg = cfg
+
+    def set_threshold(self, threshold):
         self.threshold = threshold
 
+    def set_uncond_zero_scale(self, scale):
+        self.uz_scale = scale
+
+    def zero_cond(self, args):
+        cond = args["cond_denoised"]
+        x = args["input"]
+        x -= x.mean()
+        cond -= cond.mean()
+        return x - (cond / cond.std() ** 0.5) * self.uz_scale
+
     def predict_noise(self, x, timestep, model_options={}, seed=None):
         cond = self.conds.get("positive")
         uncond = self.conds.get("negative")
         ts = timestep[0].item()
         if self.threshold_timestep > ts:
+            if self.uz_scale > 0.0:
+                model_options = model_options.copy()
+                model_options["sampler_cfg_function"] = self.zero_cond
             return comfy.samplers.sampling_function(
                 self.inner_model, x, timestep, uncond, cond, 1.0, model_options=model_options, seed=seed
             )
-        else:
-            self.threshold_timestep = 0
-            uncond_pred, cond_pred = comfy.samplers.calc_cond_batch(
-                self.inner_model, [uncond, cond], x, timestep, model_options
-            )
+        self.threshold_timestep = 0
+        uncond_pred, cond_pred = comfy.samplers.calc_cond_batch(
+            self.inner_model, [uncond, cond], x, timestep, model_options
+        )
+        if not self.threshold >= 1.0:
             # Is this reshape correct? It at least gives a scalar value...
             sim = cos(cond_pred.reshape(1, -1), uncond_pred.reshape(1, -1)).item()
             if sim >= self.threshold:
                 print("AdaptiveGuidance: Cosine similarity", sim, "exceeds threshold, setting CFG to 1.0")
                 self.threshold_timestep = ts
-            return comfy.samplers.cfg_function(
-                self.inner_model,
-                cond_pred,
-                uncond_pred,
-                self.cfg,
-                x,
-                timestep,
-                model_options=model_options,
-                cond=cond,
-                uncond=uncond,
-            )
+        return comfy.samplers.cfg_function(
+            self.inner_model,
+            cond_pred,
+            uncond_pred,
+            self.cfg,
+            x,
+            timestep,
+            model_options=model_options,
+            cond=cond,
+            uncond=uncond,
+        )
 
 
 class AdaptiveGuidance:
@@ -52,7 +68,8 @@ class AdaptiveGuidance:
                 "negative": ("CONDITIONING",),
                 "threshold": ("FLOAT", {"default": 0.990, "min": 0.90, "max": 1.0, "step": 0.001, "round": 0.001}),
                 "cfg": ("FLOAT", {"default": 8.0, "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.01}),
-            }
+            },
+            "optional": {"uncond_zero_scale": ("FLOAT", {"default": 0.0, "max": 2.0, "step": 0.01})},
         }
 
     RETURN_TYPES = ("GUIDER",)
@@ -60,10 +77,12 @@ class AdaptiveGuidance:
 
     CATEGORY = "sampling/custom_sampling/guiders"
 
-    def patch(self, model, positive, negative, threshold, cfg):
+    def patch(self, model, positive, negative, threshold, cfg, uncond_zero_scale=0.0):
         g = AdaptiveGuider(model)
         g.set_conds(positive, negative)
-        g.set_cfg(cfg, threshold)
+        g.set_threshold(threshold)
+        g.set_uncond_zero_scale(uncond_zero_scale)
+        g.set_cfg(cfg)
 
         return (g,)
 

pyproject.toml

@@ -0,0 +1,14 @@
+[project]
+name = "comfyui-adaptive-guidance"
+description = "An implementation of adaptive guidance for ComfyUI\nSee https://bcv-uniandes.github.io/adaptiveguidance-wp/"
+version = "0.1.0"
+license = "GPL-3.0"
+
+[project.urls]
+Repository = "https://github.com/asagi4/ComfyUI-Adaptive-Guidance"
+#  Used by Comfy Registry https://comfyregistry.org
+
+[tool.comfy]
+PublisherId = "asagi4"
+DisplayName = "ComfyUI Adaptive Guidance"
+Icon = ""