[diffusion] model: sync with upstream z-Image (#17822)

python/sglang/multimodal_gen/registry.py

@@ -98,7 +98,10 @@ from sglang.multimodal_gen.configs.sample.wan import (
     WanT2V_1_3B_SamplingParams,
     WanT2V_14B_SamplingParams,
 )
-from sglang.multimodal_gen.configs.sample.zimage import ZImageSamplingParams
+from sglang.multimodal_gen.configs.sample.zimage import (
+    ZImageSamplingParams,
+    ZImageTurboSamplingParams,
+)
 from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
@@ -598,12 +601,22 @@ def _register_configs():
         ],
     )
     register_configs(
-        sampling_param_cls=ZImageSamplingParams,
+        sampling_param_cls=ZImageTurboSamplingParams,
         pipeline_config_cls=ZImagePipelineConfig,
         hf_model_paths=[
             "Tongyi-MAI/Z-Image-Turbo",
         ],
-        model_detectors=[lambda hf_id: "z-image" in hf_id.lower()],
+        model_detectors=[lambda hf_id: "z-image-turbo" in hf_id.lower()],
+    )
+    register_configs(
+        sampling_param_cls=ZImageSamplingParams,
+        pipeline_config_cls=ZImagePipelineConfig,
+        hf_model_paths=[
+            "Tongyi-MAI/Z-Image",
+        ],
+        model_detectors=[
+            lambda hf_id: "z-image" in hf_id.lower() and "turbo" not in hf_id.lower()
+        ],
     )
     # Qwen-Image
     register_configs(