diff --git a/fine_tune.py b/fine_tune.py
index c9102f6c0..fb6b3ed69 100644
--- a/fine_tune.py
+++ b/fine_tune.py
@@ -337,6 +337,9 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
     train_util.sample_images(
         accelerator, args, 0, global_step, accelerator.device, vae, tokenize_strategy.tokenizer, text_encoder, unet
     )
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     loss_recorder = train_util.LossRecorder()
     for epoch in range(num_train_epochs):
@@ -456,7 +459,7 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
                         )
 
             current_loss = loss.detach().item()  # 平均なのでbatch sizeは関係ないはず
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss}
                 train_util.append_lr_to_logs(logs, lr_scheduler, args.optimizer_type, including_unet=True)
                 accelerator.log(logs, step=global_step)
@@ -469,7 +472,7 @@ def fn_recursive_set_mem_eff(module: torch.nn.Module):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/flux_train.py b/flux_train.py
index 32a36f036..573b1911e 100644
--- a/flux_train.py
+++ b/flux_train.py
@@ -629,6 +629,9 @@ def optimizer_hook(parameter: torch.Tensor):
 
     # For --sample_at_first
     flux_train_utils.sample_images(accelerator, args, 0, global_step, flux, ae, [clip_l, t5xxl], sample_prompts_te_outputs)
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     loss_recorder = train_util.LossRecorder()
     epoch = 0  # avoid error when max_train_steps is 0
@@ -777,7 +780,7 @@ def optimizer_hook(parameter: torch.Tensor):
                         )
 
             current_loss = loss.detach().item()  # 平均なのでbatch sizeは関係ないはず
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss}
                 train_util.append_lr_to_logs(logs, lr_scheduler, args.optimizer_type, including_unet=True)
 
@@ -791,7 +794,7 @@ def optimizer_hook(parameter: torch.Tensor):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/library/flux_train_utils.py b/library/flux_train_utils.py
index a8e94ac00..b42cb0105 100644
--- a/library/flux_train_utils.py
+++ b/library/flux_train_utils.py
@@ -241,17 +241,19 @@ def sample_image_inference(
     img_filename = f"{'' if args.output_name is None else args.output_name + '_'}{num_suffix}_{i:02d}_{ts_str}{seed_suffix}.png"
     image.save(os.path.join(save_dir, img_filename))
 
-    # wandb有効時のみログを送信
-    try:
+    # send images to wandb if enabled
+    if "wandb" in [tracker.name for tracker in accelerator.trackers]:
         wandb_tracker = accelerator.get_tracker("wandb")
-        try:
-            import wandb
-        except ImportError:  # 事前に一度確認するのでここはエラー出ないはず
-            raise ImportError("No wandb / wandb がインストールされていないようです")
 
-        wandb_tracker.log({f"sample_{i}": wandb.Image(image)})
-    except:  # wandb 無効時
-        pass
+        import wandb
+        # not to commit images to avoid inconsistency between training and logging steps
+        wandb_tracker.log(
+            {f"sample_{i}": wandb.Image(
+                image,
+                caption=prompt # positive prompt as a caption
+            )}, 
+            commit=False
+        )
 
 
 def time_shift(mu: float, sigma: float, t: torch.Tensor):
diff --git a/library/sd3_train_utils.py b/library/sd3_train_utils.py
index da0729506..e819d440c 100644
--- a/library/sd3_train_utils.py
+++ b/library/sd3_train_utils.py
@@ -604,17 +604,19 @@ def sample_image_inference(
     img_filename = f"{'' if args.output_name is None else args.output_name + '_'}{num_suffix}_{i:02d}_{ts_str}{seed_suffix}.png"
     image.save(os.path.join(save_dir, img_filename))
 
-    # wandb有効時のみログを送信
-    try:
+    # send images to wandb if enabled
+    if "wandb" in [tracker.name for tracker in accelerator.trackers]:
         wandb_tracker = accelerator.get_tracker("wandb")
-        try:
-            import wandb
-        except ImportError:  # 事前に一度確認するのでここはエラー出ないはず
-            raise ImportError("No wandb / wandb がインストールされていないようです")
 
-        wandb_tracker.log({f"sample_{i}": wandb.Image(image)})
-    except:  # wandb 無効時
-        pass
+        import wandb
+        # not to commit images to avoid inconsistency between training and logging steps
+        wandb_tracker.log(
+            {f"sample_{i}": wandb.Image(
+                image,
+                caption=prompt # positive prompt as a caption
+            )}, 
+            commit=False
+        )
 
 
 # region Diffusers
diff --git a/library/train_util.py b/library/train_util.py
index 989758ad5..e23676209 100644
--- a/library/train_util.py
+++ b/library/train_util.py
@@ -5745,17 +5745,19 @@ def sample_image_inference(
     img_filename = f"{'' if args.output_name is None else args.output_name + '_'}{num_suffix}_{i:02d}_{ts_str}{seed_suffix}.png"
     image.save(os.path.join(save_dir, img_filename))
 
-    # wandb有効時のみログを送信
-    try:
+    # send images to wandb if enabled
+    if "wandb" in [tracker.name for tracker in accelerator.trackers]:
         wandb_tracker = accelerator.get_tracker("wandb")
-        try:
-            import wandb
-        except ImportError:  # 事前に一度確認するのでここはエラー出ないはず
-            raise ImportError("No wandb / wandb がインストールされていないようです")
 
-        wandb_tracker.log({f"sample_{i}": wandb.Image(image)})
-    except:  # wandb 無効時
-        pass
+        import wandb
+        # not to commit images to avoid inconsistency between training and logging steps
+        wandb_tracker.log(
+            {f"sample_{i}": wandb.Image(
+                image,
+                caption=prompt # positive prompt as a caption
+            )}, 
+            commit=False
+        )
 
 
 # endregion
diff --git a/sd3_train.py b/sd3_train.py
index 3b6c8a118..91ccc0953 100644
--- a/sd3_train.py
+++ b/sd3_train.py
@@ -662,6 +662,9 @@ def optimizer_hook(parameter: torch.Tensor):
 
     # For --sample_at_first
     sd3_train_utils.sample_images(accelerator, args, 0, global_step, mmdit, vae, [clip_l, clip_g, t5xxl], sample_prompts_te_outputs)
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     # following function will be moved to sd3_train_utils
 
@@ -881,7 +884,7 @@ def compute_loss_weighting_for_sd3(weighting_scheme: str, sigmas=None):
                         )
 
             current_loss = loss.detach().item()  # 平均なのでbatch sizeは関係ないはず
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss}
                 train_util.append_lr_to_logs(logs, lr_scheduler, args.optimizer_type, including_unet=train_mmdit)
 
@@ -895,7 +898,7 @@ def compute_loss_weighting_for_sd3(weighting_scheme: str, sigmas=None):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/sdxl_train.py b/sdxl_train.py
index b2c62dd11..7291ddd2f 100644
--- a/sdxl_train.py
+++ b/sdxl_train.py
@@ -617,6 +617,9 @@ def optimizer_hook(parameter: torch.Tensor):
     sdxl_train_util.sample_images(
         accelerator, args, 0, global_step, accelerator.device, vae, tokenizers, [text_encoder1, text_encoder2], unet
     )
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     loss_recorder = train_util.LossRecorder()
     for epoch in range(num_train_epochs):
@@ -797,7 +800,7 @@ def optimizer_hook(parameter: torch.Tensor):
                         )
 
             current_loss = loss.detach().item()  # 平均なのでbatch sizeは関係ないはず
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss}
                 if block_lrs is None:
                     train_util.append_lr_to_logs(logs, lr_scheduler, args.optimizer_type, including_unet=train_unet)
@@ -814,7 +817,7 @@ def optimizer_hook(parameter: torch.Tensor):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/sdxl_train_control_net_lllite.py b/sdxl_train_control_net_lllite.py
index 0eaec29b8..9d1cfc63e 100644
--- a/sdxl_train_control_net_lllite.py
+++ b/sdxl_train_control_net_lllite.py
@@ -541,14 +541,14 @@ def remove_model(old_ckpt_name):
             logs = {"avr_loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
             progress_bar.set_postfix(**logs)
 
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = generate_step_logs(args, current_loss, avr_loss, lr_scheduler)
                 accelerator.log(logs, step=global_step)
 
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/sdxl_train_control_net_lllite_old.py b/sdxl_train_control_net_lllite_old.py
index 292a0463a..6fa1d6096 100644
--- a/sdxl_train_control_net_lllite_old.py
+++ b/sdxl_train_control_net_lllite_old.py
@@ -480,14 +480,14 @@ def remove_model(old_ckpt_name):
             logs = {"avr_loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
             progress_bar.set_postfix(**logs)
 
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = generate_step_logs(args, current_loss, avr_loss, lr_scheduler)
                 accelerator.log(logs, step=global_step)
 
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/train_controlnet.py b/train_controlnet.py
index c9ac6c5a8..57f0d263f 100644
--- a/train_controlnet.py
+++ b/train_controlnet.py
@@ -409,6 +409,9 @@ def remove_model(old_ckpt_name):
     train_util.sample_images(
         accelerator, args, 0, global_step, accelerator.device, vae, tokenizer, text_encoder, unet, controlnet=controlnet
     )
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     # training loop
     for epoch in range(num_train_epochs):
@@ -542,14 +545,14 @@ def remove_model(old_ckpt_name):
             logs = {"avr_loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
             progress_bar.set_postfix(**logs)
 
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = generate_step_logs(args, current_loss, avr_loss, lr_scheduler)
                 accelerator.log(logs, step=global_step)
 
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/train_db.py b/train_db.py
index 7caee6647..d42afd89a 100644
--- a/train_db.py
+++ b/train_db.py
@@ -315,6 +315,9 @@ def train(args):
     train_util.sample_images(
         accelerator, args, 0, global_step, accelerator.device, vae, tokenize_strategy.tokenizer, text_encoder, unet
     )
+    if len(accelerator.trackers) > 0:
+        # log empty object to commit the sample images to wandb
+        accelerator.log({}, step=0)
 
     loss_recorder = train_util.LossRecorder()
     for epoch in range(num_train_epochs):
@@ -445,7 +448,7 @@ def train(args):
                         )
 
             current_loss = loss.detach().item()
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss}
                 train_util.append_lr_to_logs(logs, lr_scheduler, args.optimizer_type, including_unet=True)
                 accelerator.log(logs, step=global_step)
@@ -458,7 +461,7 @@ def train(args):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_recorder.moving_average}
             accelerator.log(logs, step=epoch + 1)
 
diff --git a/train_network.py b/train_network.py
index 048c7e7bd..b83890b94 100644
--- a/train_network.py
+++ b/train_network.py
@@ -1018,6 +1018,9 @@ def remove_model(old_ckpt_name):
 
         # For --sample_at_first
         self.sample_images(accelerator, args, 0, global_step, accelerator.device, vae, tokenizers, text_encoder, unet)
+        if len(accelerator.trackers) > 0:
+            # log empty object to commit the sample images to wandb
+            accelerator.log({}, step=0)
 
         # training loop
         if initial_step > 0:  # only if skip_until_initial_step is specified
@@ -1200,7 +1203,7 @@ def remove_model(old_ckpt_name):
                 if args.scale_weight_norms:
                     progress_bar.set_postfix(**{**max_mean_logs, **logs})
 
-                if args.logging_dir is not None:
+                if len(accelerator.trackers) > 0:
                     logs = self.generate_step_logs(
                         args, current_loss, avr_loss, lr_scheduler, lr_descriptions, keys_scaled, mean_norm, maximum_norm
                     )
@@ -1209,7 +1212,7 @@ def remove_model(old_ckpt_name):
                 if global_step >= args.max_train_steps:
                     break
 
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss/epoch": loss_recorder.moving_average}
                 accelerator.log(logs, step=epoch + 1)
 
diff --git a/train_textual_inversion.py b/train_textual_inversion.py
index 9044f50df..956c78603 100644
--- a/train_textual_inversion.py
+++ b/train_textual_inversion.py
@@ -550,6 +550,9 @@ def remove_model(old_ckpt_name):
             unet,
             prompt_replacement,
         )
+        if len(accelerator.trackers) > 0:
+            # log empty object to commit the sample images to wandb
+            accelerator.log({}, step=0)
 
         # training loop
         for epoch in range(num_train_epochs):
@@ -684,7 +687,7 @@ def remove_model(old_ckpt_name):
                                 remove_model(remove_ckpt_name)
 
                 current_loss = loss.detach().item()
-                if args.logging_dir is not None:
+                if len(accelerator.trackers) > 0:
                     logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
                     if (
                         args.optimizer_type.lower().startswith("DAdapt".lower()) or args.optimizer_type.lower() == "Prodigy".lower()
@@ -702,7 +705,7 @@ def remove_model(old_ckpt_name):
                 if global_step >= args.max_train_steps:
                     break
 
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss/epoch": loss_total / len(train_dataloader)}
                 accelerator.log(logs, step=epoch + 1)
 
@@ -739,6 +742,7 @@ def remove_model(old_ckpt_name):
                 unet,
                 prompt_replacement,
             )
+            accelerator.log({})
 
             # end of epoch
 
diff --git a/train_textual_inversion_XTI.py b/train_textual_inversion_XTI.py
index efb59137b..ca0b603fb 100644
--- a/train_textual_inversion_XTI.py
+++ b/train_textual_inversion_XTI.py
@@ -538,7 +538,7 @@ def remove_model(old_ckpt_name):
                             remove_model(remove_ckpt_name)
 
             current_loss = loss.detach().item()
-            if args.logging_dir is not None:
+            if len(accelerator.trackers) > 0:
                 logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
                 if (
                     args.optimizer_type.lower().startswith("DAdapt".lower()) or args.optimizer_type.lower() == "Prodigy".lower()
@@ -556,7 +556,7 @@ def remove_model(old_ckpt_name):
             if global_step >= args.max_train_steps:
                 break
 
-        if args.logging_dir is not None:
+        if len(accelerator.trackers) > 0:
             logs = {"loss/epoch": loss_total / len(train_dataloader)}
             accelerator.log(logs, step=epoch + 1)