diff --git a/tests/diffusion/test_diffusion_scheduler.py b/tests/diffusion/test_diffusion_scheduler.py
index 0a317f1fb..d6d17c18f 100644
--- a/tests/diffusion/test_diffusion_scheduler.py
+++ b/tests/diffusion/test_diffusion_scheduler.py
@@ -276,12 +276,16 @@ class TestGetRequestBatchSamplingParamsKey:
         generator: torch.Generator | None = None,
         extra_args: dict | None = None,
         condition_key: tuple | None = None,
+        guidance_scale: float | None = None,
+        guidance_scale_2: float | None = None,
     ) -> OmniDiffusionRequest:
         sp = OmniDiffusionSamplingParams(
             num_inference_steps=num_inference_steps,
             seed=seed,
             generator=generator,
             extra_args=extra_args or {},
+            guidance_scale=guidance_scale,
+            guidance_scale_2=guidance_scale_2,
         )
         return OmniDiffusionRequest(
             prompt="prompt",
@@ -356,6 +360,21 @@ class TestGetRequestBatchSamplingParamsKey:
             self._make(condition_key=("wan22_s2v_condition", True))
         ) != scheduler._build_sampling_params_key(self._make(condition_key=("wan22_s2v_condition", False)))
 
+    def test_distinguishes_explicit_guidance_scale_2(self) -> None:
+        # An omitted guidance_scale_2 is auto-filled from guidance_scale, so a
+        # request that omits it and one that passes the same value explicitly end
+        # up with an identical numeric guidance_scale_2 but different
+        # guidance_scale_2_provided. Pipelines read guidance_scale_2_provided from
+        # the batch's first request to gate image guidance, so the two must not
+        # share a request batch.
+        scheduler = RequestScheduler()
+        omitted = self._make(guidance_scale=2.0)
+        explicit = self._make(guidance_scale=2.0, guidance_scale_2=2.0)
+
+        assert omitted.sampling_params.guidance_scale_2 == explicit.sampling_params.guidance_scale_2
+        assert omitted.sampling_params.guidance_scale_2_provided != explicit.sampling_params.guidance_scale_2_provided
+        assert scheduler._build_sampling_params_key(omitted) != scheduler._build_sampling_params_key(explicit)
+
 
 class TestRequestScheduler:
     def setup_method(self) -> None:
diff --git a/vllm_omni/diffusion/sched/interface.py b/vllm_omni/diffusion/sched/interface.py
index d97b5cbd0..122a97ec2 100644
--- a/vllm_omni/diffusion/sched/interface.py
+++ b/vllm_omni/diffusion/sched/interface.py
@@ -119,6 +119,7 @@ class RequestBatchSamplingParamsKey:
     guidance_scale: float = 0.0
     guidance_scale_provided: bool = False
     guidance_scale_2: float | None = None
+    guidance_scale_2_provided: bool = False
     guidance_rescale: float = 0.0
     true_cfg_scale: float | None = None
     cfg_normalize: bool = False