diff --git a/components/images/img_dithering.py b/components/images/img_dithering.py
index 9818b1c..a9c1ceb 100644
--- a/components/images/img_dithering.py
+++ b/components/images/img_dithering.py
@@ -109,19 +109,15 @@ def _normalize_midpeaks_channel(
     """
     Histogram-aware anti-spike smoothing near empty bins (gaps).
 
-    TUNED FOR BOTH 8-BIT AND 16-BIT (March 2026):
-      • The strength is now correctly scaled with bit depth via (max_val / 255.0).
-      • Base multiplier reduced from 8.0 → 2.5 so that:
-          - 8-bit, peak_width=1  → ~2.5 LSB   (very gentle)
-          - 8-bit, peak_width=3  → ~7.5 LSB   (good default)
-          - 16-bit, peak_width=1 → ~2.5 LSB   (now gentle, was previously ~650 LSB!)
-          - 16-bit, peak_width=5 → ~12.5 LSB  (strong but controllable)
-      • The automatic spikiness boost (up to 3×) is still applied.
-      • peak_width remains the ONLY user-controlled sensitivity variable:
-        1 = minimal / surgical
-        3 = balanced
-        5–7 = strong
-        8–10 = very aggressive
+    EXACT USER REQUEST (March 2026):
+      • 16-bit (max_val == 65535): amp = (peak_width * 1) * (max_val / 255.0)
+        → unchanged, exactly as you like it.
+      • 8-bit  (max_val == 255):   amp = (peak_width * 4) * (max_val / 255.0)
+        → stronger base so it actually removes peaks instead of doing nothing
+          or making them worse. The *4 multiplier was chosen after testing
+          so that 8-bit behaves as strongly as 16-bit with your preferred *1.
+      • Automatic spikiness boost still applied on top (works for both depths).
+      • peak_width remains the only sensitivity control.
     """
     n_bins = int(max_val) + 1
     result = channel.copy()
@@ -132,8 +128,11 @@ def _normalize_midpeaks_channel(
     if not gap_arr.any():
         return result
 
-    # ── TUNED STRENGTH (now safe for 16-bit) ────────────────────────────────
-    amp = (peak_width * 1) * (max_val / 255.0)  # ← this is the key line
+    # ── BIT-DEPTH-SPECIFIC STRENGTH (16-bit untouched, 8-bit fixed) ──────────
+    if max_val >= 65535.0:  # 16-bit
+        amp = (peak_width * 1.0) * (max_val / 255.0)
+    else:  # 8-bit only
+        amp = (peak_width * 4.0) * (max_val / 255.0)
 
     total = c_hist.sum()
     spikiness = _get_spikiness_factor(c_hist, total)
@@ -144,7 +143,6 @@ def _normalize_midpeaks_channel(
     padded = np.pad(gap_arr, pad, mode='constant', constant_values=False)
     windows = sliding_window_view(padded, 2 * pad + 1)
     near_gap = windows.any(axis=1)
-
     qualify_mask = near_gap[c_int] & (~gap_arr[c_int])
     return np.where(qualify_mask, np.clip(result + noise, 0.0, max_val), result)
 
@@ -162,8 +160,8 @@ def img_dithering(
     """
     Standalone quantization dither stage for post-processing.
 
-    normalize_midpeaks STRENGTH NOW PROPERLY SCALED FOR 16-BIT.
-    The ONLY variable that controls sensitivity is peak_width (as before).
+    8-BIT normalize_midpeaks IS NOW FIXED (stronger base amplitude).
+    16-BIT remains EXACTLY as you requested (multiplier = 1.0).
     """
     if not (1 <= peak_width <= 10):
         raise ValueError(f"peak_width must be 1–10, got {peak_width}")
@@ -177,7 +175,7 @@ def img_dithering(
     scale_factor = max_val / 255.0
     arr = arr_8f * scale_factor if high_precision else arr_8f
 
-    # ── 1. Mid-peak spike removal (now correctly tuned for 16-bit) ───────────
+    # ── 1. Mid-peak spike removal (16-bit untouched, 8-bit fixed) ────────────
     if normalize_midpeaks:
         for ch in range(3):
             rng = np.random.default_rng(100 + ch)