diff --git a/python/sglang/multimodal_gen/test/run_suite.py b/python/sglang/multimodal_gen/test/run_suite.py
index 9f47799cdfa4..26fd0213d068 100644
--- a/python/sglang/multimodal_gen/test/run_suite.py
+++ b/python/sglang/multimodal_gen/test/run_suite.py
@@ -22,12 +22,14 @@
     "1-gpu": [
         "test_server_a.py",
         "test_server_b.py",
+        "test_server_correctness.py",
         "test_lora_format_adapter.py",
         # add new 1-gpu test files here
     ],
     "2-gpu": [
         "test_server_2_gpu_a.py",
         "test_server_2_gpu_b.py",
+        "test_server_2_gpu_correctness.py",
         # add new 2-gpu test files here
     ],
 }
diff --git a/python/sglang/multimodal_gen/test/server/test_server_2_gpu_correctness.py b/python/sglang/multimodal_gen/test/server/test_server_2_gpu_correctness.py
new file mode 100644
index 000000000000..a2891a07c6ed
--- /dev/null
+++ b/python/sglang/multimodal_gen/test/server/test_server_2_gpu_correctness.py
@@ -0,0 +1,33 @@
+"""
+End-to-end functional correctness tests for 2-GPU diffusion server.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
+from sglang.multimodal_gen.test.server.test_server_common import (  # noqa: F401
+    diffusion_server,
+)
+from sglang.multimodal_gen.test.server.test_server_correctness import (
+    CorrectnessTestMixin,
+)
+from sglang.multimodal_gen.test.server.testcase_configs import (
+    CORRECTNESS_2_GPU_CASES,
+    DiffusionTestCase,
+)
+
+logger = init_logger(__name__)
+
+
+class TestDiffusionServerTwoGpuCorrectness(CorrectnessTestMixin):
+    """
+    Functional correctness tests for 2-GPU diffusion cases.
+    Inherits shared functional logic from CorrectnessTestMixin.
+    """
+
+    @pytest.fixture(params=CORRECTNESS_2_GPU_CASES, ids=lambda c: c.id)
+    def case(self, request) -> DiffusionTestCase:
+        """Provide a DiffusionTestCase for each 2-GPU correctness test."""
+        return request.param
diff --git a/python/sglang/multimodal_gen/test/server/test_server_correctness.py b/python/sglang/multimodal_gen/test/server/test_server_correctness.py
new file mode 100644
index 000000000000..5120c83feb60
--- /dev/null
+++ b/python/sglang/multimodal_gen/test/server/test_server_correctness.py
@@ -0,0 +1,104 @@
+"""
+End-to-end functional correctness tests for diffusion server.
+Verifies functional logic, parameter propagation, and error robustness.
+"""
+
+from __future__ import annotations
+
+import pytest
+import requests
+from openai import OpenAI
+
+from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
+from sglang.multimodal_gen.test.server.test_server_common import (  # noqa: F401
+    diffusion_server,
+)
+from sglang.multimodal_gen.test.server.test_server_utils import get_generate_fn
+from sglang.multimodal_gen.test.server.testcase_configs import (
+    CORRECTNESS_1_GPU_CASES,
+    DiffusionTestCase,
+)
+
+logger = init_logger(__name__)
+
+
+class CorrectnessTestMixin:
+    """
+    Mixin containing functional verification logic shared across GPU suites.
+    """
+
+    def _client(self, ctx):
+        """Get OpenAI client for the server."""
+        return OpenAI(
+            api_key="sglang-anything",
+            base_url=f"http://localhost:{ctx.port}/v1",
+        )
+
+    def test_functional_success(self, case: DiffusionTestCase, diffusion_server):
+        """
+        Verify that the model generates output successfully for its modality.
+        Reuses project-standard polling and validation logic from get_generate_fn.
+        """
+        client = self._client(diffusion_server)
+        generate_fn = get_generate_fn(
+            model_path=case.server_args.model_path,
+            modality=case.server_args.modality,
+            sampling_params=case.sampling_params,
+        )
+
+        # functional success check
+        generate_fn(case.id, client)
+        logger.info(f"Functional success verified for {case.id}")
+
+    def test_seed_determinism(self, case: DiffusionTestCase, diffusion_server):
+        """
+        Verify bit-identical results for identical seeds (Image only).
+        """
+        if case.server_args.modality != "image":
+            pytest.skip("Seed determinism check restricted to image modality")
+
+        client = self._client(diffusion_server)
+        payload = {
+            "model": case.server_args.model_path,
+            "prompt": case.sampling_params.prompt or "A dog with sunglasses",
+            "size": case.sampling_params.output_size,
+            "response_format": "b64_json",
+            "seed": 42,
+        }
+
+        resp1 = client.images.generate(**payload)
+        resp2 = client.images.generate(**payload)
+
+        assert (
+            resp1.data[0].b64_json == resp2.data[0].b64_json
+        ), "Seed determinism failed: outputs are not bit-identical"
+        logger.info(f"Seed determinism verified for {case.id}")
+
+    def test_api_error_codes(self, case: DiffusionTestCase, diffusion_server):
+        """
+        Verify server returns correct HTTP error codes for malformed requests.
+        """
+        if case.server_args.modality != "image":
+            pytest.skip(
+                "Error code boundary check restricted to image generations endpoint"
+            )
+
+        client = self._client(diffusion_server)
+        # Use client.base_url to construct the target endpoint dynamically
+        base_url = f"{client.base_url}/images/generations"
+
+        # Verify 422 for missing mandatory prompt
+        payload = {"model": case.server_args.model_path}
+        resp = requests.post(base_url, json=payload)
+        assert resp.status_code == 422
+        logger.info(f"Error handling verified for {case.id}")
+
+
+class TestDiffusionCorrectness(CorrectnessTestMixin):
+    """
+    Functional correctness tests for 1-GPU diffusion cases.
+    """
+
+    @pytest.fixture(params=CORRECTNESS_1_GPU_CASES, ids=lambda c: c.id)
+    def case(self, request) -> DiffusionTestCase:
+        return request.param
diff --git a/python/sglang/multimodal_gen/test/server/testcase_configs.py b/python/sglang/multimodal_gen/test/server/testcase_configs.py
index 0bb2b5a5a273..bbef43052c47 100644
--- a/python/sglang/multimodal_gen/test/server/testcase_configs.py
+++ b/python/sglang/multimodal_gen/test/server/testcase_configs.py
@@ -569,5 +569,49 @@ def from_req_perf_record(
     ),
 ]
 
+# Correctness Cases
+CORRECTNESS_1_GPU_CASES = [
+    # 1. Standard Image Correctness (Seed & Functional Check)
+    DiffusionTestCase(
+        "qwen_image_correctness_seed",
+        DiffusionServerArgs(
+            model_path="Qwen/Qwen-Image", modality="image", warmup_text=1
+        ),
+        T2I_sampling_params,
+    ),
+    # Multimodal Image Edit Correctness (Exercises /edits and Vision encoders)
+    DiffusionTestCase(
+        "qwen_image_edit_correctness",
+        DiffusionServerArgs(
+            model_path="Qwen/Qwen-Image-Edit", modality="image", warmup_edit=1
+        ),
+        TI2I_sampling_params,
+    ),
+    # Video generation correctness
+    DiffusionTestCase(
+        "wan_video_correctness",
+        DiffusionServerArgs(
+            model_path="Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
+            modality="video",
+            warmup_text=0,
+        ),
+        DiffusionSamplingParams(prompt="A dog with sunglasses", num_frames=8),
+    ),
+]
+
+CORRECTNESS_2_GPU_CASES = [
+    # Multimodal Image to Video
+    DiffusionTestCase(
+        "wan_video_i2v_correctness_2gpu",
+        DiffusionServerArgs(
+            model_path="Wan-AI/Wan2.1-I2V-14B-480P-Diffusers",
+            modality="video",
+            num_gpus=2,
+            warmup_edit=0,
+        ),
+        TI2V_sampling_params,
+    ),
+]
+
 # Load global configuration
 BASELINE_CONFIG = BaselineConfig.load(Path(__file__).with_name("perf_baselines.json"))