diff --git a/fine_tune.py b/fine_tune.py
index b556672d..19a35229 100644
--- a/fine_tune.py
+++ b/fine_tune.py
@@ -383,10 +383,10 @@ def train(args):
 
                     if args.min_snr_gamma:
                         loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                    if args.scale_v_pred_loss_like_noise_pred:
+                    if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                         loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                     if args.debiased_estimation_loss:
-                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                     loss = loss.mean()  # mean over batch dimension
                 else:
diff --git a/library/custom_train_functions.py b/library/custom_train_functions.py
index 2a513dc5..faf44304 100644
--- a/library/custom_train_functions.py
+++ b/library/custom_train_functions.py
@@ -96,10 +96,13 @@ def add_v_prediction_like_loss(loss, timesteps, noise_scheduler, v_pred_like_los
     return loss
 
 
-def apply_debiased_estimation(loss, timesteps, noise_scheduler):
+def apply_debiased_estimation(loss, timesteps, noise_scheduler, v_prediction=False):
     snr_t = torch.stack([noise_scheduler.all_snr[t] for t in timesteps])  # batch_size
     snr_t = torch.minimum(snr_t, torch.ones_like(snr_t) * 1000)  # if timestep is 0, snr_t is inf, so limit it to 1000
-    weight = 1 / torch.sqrt(snr_t)
+    if v_prediction:
+        weight = 1 / (snr_t + 1)
+    else:
+        weight = 1 / torch.sqrt(snr_t)
     loss = weight * loss
     return loss
 
diff --git a/library/train_util.py b/library/train_util.py
index 27910dc9..adb983d2 100644
--- a/library/train_util.py
+++ b/library/train_util.py
@@ -3731,6 +3731,11 @@ def verify_training_args(args: argparse.Namespace):
         raise ValueError(
             "scale_v_pred_loss_like_noise_pred can be enabled only with v_parameterization / scale_v_pred_loss_like_noise_predはv_parameterizationが有効なときのみ有効にできます"
         )
+    
+    if args.scale_v_pred_loss_like_noise_pred and args.zero_terminal_snr:
+        raise ValueError(
+            "zero_terminal_snr enabled. scale_v_pred_loss_like_noise_pred will not be used / zero_terminal_snrが有効です。scale_v_pred_loss_like_noise_predは使用されません"
+        )
 
     if args.v_pred_like_loss and args.v_parameterization:
         raise ValueError(
diff --git a/sdxl_train.py b/sdxl_train.py
index e0a8f2b2..44ee9233 100644
--- a/sdxl_train.py
+++ b/sdxl_train.py
@@ -725,12 +725,12 @@ def train(args):
 
                     if args.min_snr_gamma:
                         loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                    if args.scale_v_pred_loss_like_noise_pred:
+                    if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                         loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                     if args.v_pred_like_loss:
                         loss = add_v_prediction_like_loss(loss, timesteps, noise_scheduler, args.v_pred_like_loss)
                     if args.debiased_estimation_loss:
-                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                     loss = loss.mean()  # mean over batch dimension
                 else:
diff --git a/sdxl_train_control_net_lllite.py b/sdxl_train_control_net_lllite.py
index 5ff060a9..436f0e19 100644
--- a/sdxl_train_control_net_lllite.py
+++ b/sdxl_train_control_net_lllite.py
@@ -474,12 +474,12 @@ def train(args):
 
                 if args.min_snr_gamma:
                     loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                if args.scale_v_pred_loss_like_noise_pred:
+                if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                     loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                 if args.v_pred_like_loss:
                     loss = add_v_prediction_like_loss(loss, timesteps, noise_scheduler, args.v_pred_like_loss)
                 if args.debiased_estimation_loss:
-                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                 loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
 
diff --git a/sdxl_train_control_net_lllite_old.py b/sdxl_train_control_net_lllite_old.py
index 292a0463..8fba9eba 100644
--- a/sdxl_train_control_net_lllite_old.py
+++ b/sdxl_train_control_net_lllite_old.py
@@ -434,12 +434,12 @@ def train(args):
 
                 if args.min_snr_gamma:
                     loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                if args.scale_v_pred_loss_like_noise_pred:
+                if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                     loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                 if args.v_pred_like_loss:
                     loss = add_v_prediction_like_loss(loss, timesteps, noise_scheduler, args.v_pred_like_loss)
                 if args.debiased_estimation_loss:
-                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                 loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
 
diff --git a/train_db.py b/train_db.py
index 2c7f0258..d5a94a56 100644
--- a/train_db.py
+++ b/train_db.py
@@ -370,10 +370,10 @@ def train(args):
 
                 if args.min_snr_gamma:
                     loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                if args.scale_v_pred_loss_like_noise_pred:
+                if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                     loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                 if args.debiased_estimation_loss:
-                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                 loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
 
diff --git a/train_network.py b/train_network.py
index 044ec3aa..790fbfc9 100644
--- a/train_network.py
+++ b/train_network.py
@@ -993,12 +993,12 @@ class NetworkTrainer:
 
                     if args.min_snr_gamma:
                         loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                    if args.scale_v_pred_loss_like_noise_pred:
+                    if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                         loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                     if args.v_pred_like_loss:
                         loss = add_v_prediction_like_loss(loss, timesteps, noise_scheduler, args.v_pred_like_loss)
                     if args.debiased_estimation_loss:
-                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                     loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
 
diff --git a/train_textual_inversion.py b/train_textual_inversion.py
index 96e7bd50..10b34db5 100644
--- a/train_textual_inversion.py
+++ b/train_textual_inversion.py
@@ -598,12 +598,12 @@ class TextualInversionTrainer:
 
                     if args.min_snr_gamma:
                         loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                    if args.scale_v_pred_loss_like_noise_pred:
+                    if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                         loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                     if args.v_pred_like_loss:
                         loss = add_v_prediction_like_loss(loss, timesteps, noise_scheduler, args.v_pred_like_loss)
                     if args.debiased_estimation_loss:
-                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                        loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                     loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
 
diff --git a/train_textual_inversion_XTI.py b/train_textual_inversion_XTI.py
index efb59137..084b90c6 100644
--- a/train_textual_inversion_XTI.py
+++ b/train_textual_inversion_XTI.py
@@ -483,10 +483,10 @@ def train(args):
                 loss = loss * loss_weights
                 if args.min_snr_gamma:
                     loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma, args.v_parameterization)
-                if args.scale_v_pred_loss_like_noise_pred:
+                if args.scale_v_pred_loss_like_noise_pred and not args.zero_terminal_snr:
                     loss = scale_v_prediction_loss_like_noise_prediction(loss, timesteps, noise_scheduler)
                 if args.debiased_estimation_loss:
-                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler)
+                    loss = apply_debiased_estimation(loss, timesteps, noise_scheduler, args.v_parameterization)
 
                 loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし