From a3864b2ee4028bf55605fec92dbf0500dc4998fa Mon Sep 17 00:00:00 2001
From: ljleb <lebel.louisjacob@gmail.com>
Date: Wed, 18 Oct 2023 17:49:22 -0400
Subject: [PATCH] update presets

---
 lib_free_u/global_state.py | 10 +++++-----
 lib_free_u/unet.py         |  8 ++++----
 2 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/lib_free_u/global_state.py b/lib_free_u/global_state.py
index 6bef786..bac0134 100644
--- a/lib_free_u/global_state.py
+++ b/lib_free_u/global_state.py
@@ -143,22 +143,22 @@ def apply_xyz():
 default_presets = {
     "SD1.4 Recommendations": State(
         stage_infos=[
-            StageInfo(1.2, 0.9),
+            StageInfo(1.3, 0.9),
             StageInfo(1.4, 0.2),
             StageInfo(1, 1),
         ],
     ),
     "SD2.1 Recommendations": State(
         stage_infos=[
-            StageInfo(1.1, 0.9),
-            StageInfo(1.2, 0.2),
+            StageInfo(1.4, 0.9),
+            StageInfo(1.6, 0.2),
             StageInfo(1, 1),
         ],
     ),
     "SDXL Recommendations": State(
         stage_infos=[
-            StageInfo(1.1, 0.6),
-            StageInfo(1.2, 0.4),
+            StageInfo(1.3, 0.9),
+            StageInfo(1.4, 0.2),
             StageInfo(1, 1),
         ],
     ),
diff --git a/lib_free_u/unet.py b/lib_free_u/unet.py
index dd5589d..6494429 100644
--- a/lib_free_u/unet.py
+++ b/lib_free_u/unet.py
@@ -62,7 +62,7 @@ def free_u_cat_hijack(hs, *args, original_function, **kwargs):
 
         h[:, mask] *= get_backbone_scale(
             h,
-            backbone_factor=lerp(1, stage_info.backbone_factor, schedule_ratio),
+            base_scale=lerp(1, stage_info.backbone_factor, schedule_ratio),
         )
         h_skip = filter_skip(
             h_skip,
@@ -74,9 +74,9 @@ def free_u_cat_hijack(hs, *args, original_function, **kwargs):
     return original_function([h, h_skip], *args, **kwargs)
 
 
-def get_backbone_scale(h, backbone_factor):
+def get_backbone_scale(h, base_scale):
     if global_state.instance.version == "1":
-        return backbone_factor
+        return base_scale
 
     #if global_state.instance.version == "2":
     features_mean = h.mean(1, keepdim=True)
@@ -84,7 +84,7 @@ def get_backbone_scale(h, backbone_factor):
     features_max, _ = torch.max(features_mean.view(batch_dims, -1), dim=-1, keepdim=True)
     features_min, _ = torch.min(features_mean.view(batch_dims, -1), dim=-1, keepdim=True)
     hidden_mean = (features_mean - features_min.unsqueeze(2).unsqueeze(3)) / (features_max - features_min).unsqueeze(2).unsqueeze(3)
-    return 1 + (backbone_factor - 1) * hidden_mean
+    return 1 + (base_scale - 1) * hidden_mean
 
 
 def filter_skip(x, threshold, scale, scale_high):