diff --git a/src/llamafactory/model/model_utils/visual.py b/src/llamafactory/model/model_utils/visual.py
index cfbe6a22..4ca64569 100644
--- a/src/llamafactory/model/model_utils/visual.py
+++ b/src/llamafactory/model/model_utils/visual.py
@@ -355,7 +355,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_vl",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks"],
+    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
     language_model_keys=["language_model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )
@@ -364,7 +364,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_vl_moe",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks"],
+    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list"],
     language_model_keys=["language_model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )
@@ -373,7 +373,7 @@ _register_composite_model(
 _register_composite_model(
     model_type="qwen3_omni_moe_thinker",
     projector_key="visual.merger",
-    vision_model_keys=["visual.patch_embed", "visual.blocks", "audio_tower"],
+    vision_model_keys=["visual.patch_embed", "visual.blocks", "visual.deepstack_merger_list", "audio_tower"],
     language_model_keys=["model", "lm_head"],
     lora_conflict_keys=["patch_embed"],
 )