Use sqrt(cfg) scaling for K + store raw prev_eps

Full K corrupts at high CFG (output correction = cfg*K = 2.4 at cfg=12).
K/cfg was too weak (0.2 at cfg=12). The paper only tested up to cfg=7.5
where output corrections range 0.5-1.5.

K/sqrt(cfg) keeps output correction = sqrt(cfg)*K growing sub-linearly,
giving 0.69 at cfg=12 — within the paper's working range.

Also store raw (pre-correction) guidance as prev_eps to prevent
correction accumulation through the sliding surface.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

nodes.py

@@ -37,24 +37,17 @@ class SMCCFGCtrl:
     CATEGORY = "sampling/custom_sampling"
 
     def patch(self, model, smc_cfg_lambda, smc_cfg_K, warmup_steps):
-        # Mutable state persisted across denoising steps via closure
+        state = {"prev_eps": None, "step": 0, "prev_sigma": None}
-        state = {
-            "prev_eps": None,
-            "step": 0,
-            "prev_sigma": None,
-        }
-
         lam = smc_cfg_lambda
         K = smc_cfg_K
 
         def smc_cfg_function(args):
-            cond = args["cond"]           # x - cond_denoised (sigma-scaled noise)
+            cond = args["cond"]
-            uncond = args["uncond"]       # x - uncond_denoised (sigma-scaled noise)
+            uncond = args["uncond"]
             cond_scale = args["cond_scale"]
             sigma = args["sigma"]
 
-            # Detect new generation: sigma should decrease monotonically during
+            # Detect new generation (sigma jumps up = new sampling run)
-            # denoising. If it jumps up, a new sampling run has started.
             curr_sigma = sigma.max().item() if torch.is_tensor(sigma) else float(sigma)
             if state["prev_sigma"] is not None and curr_sigma > state["prev_sigma"] * 1.1:
                 state["prev_eps"] = None
@@ -64,21 +57,13 @@ class SMCCFGCtrl:
             step = state["step"]
             state["step"] = step + 1
 
-            # Warmup: pure conditional prediction (no guidance)
             if warmup_steps > 0 and step < warmup_steps:
                 return cond
 
-            # Normalize to noise-prediction space by dividing out sigma.
+            # Normalize to noise-prediction space (divide out sigma).
-            # The paper's K is calibrated for unit-variance noise predictions.
-            # ComfyUI's cond/uncond are (x - denoised) ≈ sigma * epsilon,
-            # so dividing by sigma recovers epsilon-space where K=0.2 is correct.
-            # Crucially, when converting back, the sigma factor naturally dampens
-            # the correction at late steps (small sigma), preventing noise injection.
             sigma_val = max(curr_sigma, 1e-8)
             guidance_eps = (cond - uncond) / sigma_val
 
-            # Initialize prev_eps on first SMC step (matches original paper
-            # where SMC correction is applied from the very first step)
             if state["prev_eps"] is None:
                 state["prev_eps"] = guidance_eps.detach().clone()
 
@@ -87,29 +72,26 @@ class SMCCFGCtrl:
             # Sliding surface: s_t = (e_t - e_{t-1}) + lambda * e_{t-1}
             s = (guidance_eps - prev_eps) + lam * prev_eps
 
-            # Smooth switching via tanh(s/phi) instead of hard sign(s).
+            # Scale K so the output correction (cond_scale * K_eff) stays
-            # The paper uses sign(s) which works in DiffSynth but creates
+            # in the range the paper tested (0.5–1.5). The paper only
-            # salt-and-pepper artifacts in ComfyUI's latent space. tanh
+            # tested up to cfg=7.5; at cfg=12, full K gives cond_scale*K=2.4
-            # provides smooth spatial gradients: proportional near zero,
+            # which corrupts the image. sqrt scaling keeps the output
-            # saturating at ±K for large |s|.
+            # correction growing sub-linearly with cfg.
-            phi = s.std().clamp(min=1e-6)
+            K_eff = K / max(cond_scale, 1.0) ** 0.5
-            u_sw = -K * torch.tanh(s / phi)
 
-            # Spatial smoothing: blur the correction to remove per-element
+            # Smooth switching via tanh instead of sign to avoid
-            # grid artifacts at VAE patch boundaries (each latent = 8x8 px).
+            # salt-and-pepper artifacts in ComfyUI's latent space.
+            phi = s.std().clamp(min=1e-6)
+            u_sw = -K_eff * torch.tanh(s / phi)
+
+            # Spatial blur to smooth grid artifacts at VAE patch boundaries.
             if u_sw.ndim == 4:
                 u_sw = F.avg_pool2d(u_sw, kernel_size=5, stride=1, padding=2)
 
-            # Store RAW guidance (before correction) for the next step's
+            # Store RAW guidance (before correction) to prevent correction
-            # sliding surface. The paper stores corrected guidance, but in
+            # accumulation through the lambda * prev_eps term.
-            # ComfyUI the corrections accumulate through the surface's
-            # lambda * prev_eps term (amplified 4x per step at lambda=5),
-            # overwhelming the actual guidance signal after a few steps.
-            # Storing raw guidance keeps the surface tracking the model's
-            # actual guidance evolution while applying corrections fresh.
             state["prev_eps"] = guidance_eps.detach().clone()
 
-            # Apply correction and convert back to sigma-scaled space
             return uncond + cond_scale * (guidance_eps + u_sw) * sigma_val
 
         m = model.clone()