diff --git a/python/sglang/multimodal_gen/__init__.py b/python/sglang/multimodal_gen/__init__.py
index 11f4ecd4e240..751822218340 100644
--- a/python/sglang/multimodal_gen/__init__.py
+++ b/python/sglang/multimodal_gen/__init__.py
@@ -1,6 +1,5 @@
 # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
-
-from sglang.multimodal_gen.configs.pipelines import PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs import PipelineConfig
 from sglang.multimodal_gen.configs.sample import SamplingParams
 from sglang.multimodal_gen.runtime.entrypoints.diffusion_generator import DiffGenerator
 
diff --git a/python/sglang/multimodal_gen/configs/pipelines/__init__.py b/python/sglang/multimodal_gen/configs/pipeline_configs/__init__.py
similarity index 60%
rename from python/sglang/multimodal_gen/configs/pipelines/__init__.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/__init__.py
index 6611a9382abd..370ec38e95f0 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/__init__.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/__init__.py
@@ -1,16 +1,16 @@
 # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
 
-from sglang.multimodal_gen.configs.pipelines.base import (
+from sglang.multimodal_gen.configs.pipeline_configs.base import (
     PipelineConfig,
     SlidingTileAttnConfig,
 )
-from sglang.multimodal_gen.configs.pipelines.flux import FluxPipelineConfig
-from sglang.multimodal_gen.configs.pipelines.hunyuan import (
+from sglang.multimodal_gen.configs.pipeline_configs.flux import FluxPipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.hunyuan import (
     FastHunyuanConfig,
     HunyuanConfig,
 )
-from sglang.multimodal_gen.configs.pipelines.stepvideo import StepVideoT2VConfig
-from sglang.multimodal_gen.configs.pipelines.wan import (
+from sglang.multimodal_gen.configs.pipeline_configs.stepvideo import StepVideoT2VConfig
+from sglang.multimodal_gen.configs.pipeline_configs.wan import (
     SelfForcingWanT2V480PConfig,
     WanI2V480PConfig,
     WanI2V720PConfig,
diff --git a/python/sglang/multimodal_gen/configs/pipelines/base.py b/python/sglang/multimodal_gen/configs/pipeline_configs/base.py
similarity index 100%
rename from python/sglang/multimodal_gen/configs/pipelines/base.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/base.py
diff --git a/python/sglang/multimodal_gen/configs/pipelines/flux.py b/python/sglang/multimodal_gen/configs/pipeline_configs/flux.py
similarity index 97%
rename from python/sglang/multimodal_gen/configs/pipelines/flux.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/flux.py
index a187c4f153a5..9c377689bfd3 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/flux.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/flux.py
@@ -13,12 +13,12 @@
     T5Config,
 )
 from sglang.multimodal_gen.configs.models.vaes.flux import FluxVAEConfig
-from sglang.multimodal_gen.configs.pipelines.base import (
+from sglang.multimodal_gen.configs.pipeline_configs.base import (
     ModelTaskType,
     PipelineConfig,
     preprocess_text,
 )
-from sglang.multimodal_gen.configs.pipelines.hunyuan import (
+from sglang.multimodal_gen.configs.pipeline_configs.hunyuan import (
     clip_postprocess_text,
     clip_preprocess_text,
 )
diff --git a/python/sglang/multimodal_gen/configs/pipelines/hunyuan.py b/python/sglang/multimodal_gen/configs/pipeline_configs/hunyuan.py
similarity index 98%
rename from python/sglang/multimodal_gen/configs/pipelines/hunyuan.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/hunyuan.py
index 73ede7d07892..2483063ead36 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/hunyuan.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/hunyuan.py
@@ -15,7 +15,7 @@
     LlamaConfig,
 )
 from sglang.multimodal_gen.configs.models.vaes import HunyuanVAEConfig
-from sglang.multimodal_gen.configs.pipelines.base import PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs import PipelineConfig
 
 PROMPT_TEMPLATE_ENCODE_VIDEO = (
     "<|start_header_id|>system<|end_header_id|>\n\nDescribe the video by detailing the following aspects: "
diff --git a/python/sglang/multimodal_gen/configs/pipelines/qwen_image.py b/python/sglang/multimodal_gen/configs/pipeline_configs/qwen_image.py
similarity index 99%
rename from python/sglang/multimodal_gen/configs/pipelines/qwen_image.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/qwen_image.py
index 54c85bdcbdd9..10bd4e610b9f 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/qwen_image.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/qwen_image.py
@@ -9,7 +9,10 @@
 from sglang.multimodal_gen.configs.models.dits.qwenimage import QwenImageDitConfig
 from sglang.multimodal_gen.configs.models.encoders.qwen_image import Qwen2_5VLConfig
 from sglang.multimodal_gen.configs.models.vaes.qwenimage import QwenImageVAEConfig
-from sglang.multimodal_gen.configs.pipelines.base import ModelTaskType, PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.base import (
+    ModelTaskType,
+    PipelineConfig,
+)
 from sglang.multimodal_gen.utils import calculate_dimensions
 
 
diff --git a/python/sglang/multimodal_gen/configs/pipelines/stepvideo.py b/python/sglang/multimodal_gen/configs/pipeline_configs/stepvideo.py
similarity index 94%
rename from python/sglang/multimodal_gen/configs/pipelines/stepvideo.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/stepvideo.py
index 586e7542bac2..aff18e5cf8b8 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/stepvideo.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/stepvideo.py
@@ -6,7 +6,7 @@
 from sglang.multimodal_gen.configs.models import DiTConfig, VAEConfig
 from sglang.multimodal_gen.configs.models.dits import StepVideoConfig
 from sglang.multimodal_gen.configs.models.vaes import StepVideoVAEConfig
-from sglang.multimodal_gen.configs.pipelines.base import PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.base import PipelineConfig
 
 
 @dataclass
diff --git a/python/sglang/multimodal_gen/configs/pipelines/wan.py b/python/sglang/multimodal_gen/configs/pipeline_configs/wan.py
similarity index 98%
rename from python/sglang/multimodal_gen/configs/pipelines/wan.py
rename to python/sglang/multimodal_gen/configs/pipeline_configs/wan.py
index d7068fba7a05..9e7f83eca374 100644
--- a/python/sglang/multimodal_gen/configs/pipelines/wan.py
+++ b/python/sglang/multimodal_gen/configs/pipeline_configs/wan.py
@@ -14,7 +14,10 @@
     T5Config,
 )
 from sglang.multimodal_gen.configs.models.vaes import WanVAEConfig
-from sglang.multimodal_gen.configs.pipelines.base import ModelTaskType, PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.base import (
+    ModelTaskType,
+    PipelineConfig,
+)
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
 logger = init_logger(__name__)
diff --git a/python/sglang/multimodal_gen/registry.py b/python/sglang/multimodal_gen/registry.py
index 08c87917995a..9600531abb9e 100644
--- a/python/sglang/multimodal_gen/registry.py
+++ b/python/sglang/multimodal_gen/registry.py
@@ -15,7 +15,7 @@
 from functools import lru_cache
 from typing import Any, Callable, Dict, List, Optional, Tuple, Type
 
-from sglang.multimodal_gen.configs.pipelines import (
+from sglang.multimodal_gen.configs.pipeline_configs import (
     FastHunyuanConfig,
     FluxPipelineConfig,
     HunyuanConfig,
@@ -25,12 +25,12 @@
     WanT2V480PConfig,
     WanT2V720PConfig,
 )
-from sglang.multimodal_gen.configs.pipelines.base import PipelineConfig
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs.base import PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
     QwenImagePipelineConfig,
 )
-from sglang.multimodal_gen.configs.pipelines.wan import (
+from sglang.multimodal_gen.configs.pipeline_configs.wan import (
     FastWan2_1_T2V_480P_Config,
     FastWan2_2_TI2V_5B_Config,
     Wan2_2_I2V_A14B_Config,
@@ -55,7 +55,7 @@
     WanT2V_1_3B_SamplingParams,
     WanT2V_14B_SamplingParams,
 )
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
 from sglang.multimodal_gen.runtime.utils.hf_diffusers_utils import (
@@ -74,49 +74,37 @@
 def _discover_and_register_pipelines():
     """
     Automatically discover and register all ComposedPipelineBase subclasses.
-    This function scans the 'sglang.multimodal_gen.runtime.architectures' package,
+    This function scans the 'sglang.multimodal_gen.runtime.pipelines' package,
     finds modules with an 'EntryClass' attribute, and maps the class's 'pipeline_name'
     to the class itself in a global registry.
     """
-    if _PIPELINE_REGISTRY:  # E-run only once
+    if _PIPELINE_REGISTRY:  # run only once
         return
 
-    package_name = "sglang.multimodal_gen.runtime.architectures"
+    package_name = "sglang.multimodal_gen.runtime.pipelines"
     package = importlib.import_module(package_name)
 
-    for _, pipeline_type_str, ispkg in pkgutil.iter_modules(package.__path__):
+    for _, module_name, ispkg in pkgutil.walk_packages(
+        package.__path__, package.__name__ + "."
+    ):
         if not ispkg:
-            continue
-        pipeline_type_package_name = f"{package_name}.{pipeline_type_str}"
-        pipeline_type_package = importlib.import_module(pipeline_type_package_name)
-        for _, arch, ispkg_arch in pkgutil.iter_modules(pipeline_type_package.__path__):
-            if not ispkg_arch:
-                continue
-            arch_package_name = f"{pipeline_type_package_name}.{arch}"
-            arch_package = importlib.import_module(arch_package_name)
-            for _, module_name, ispkg_module in pkgutil.walk_packages(
-                arch_package.__path__, arch_package.__name__ + "."
-            ):
-                if not ispkg_module:
-                    pipeline_module = importlib.import_module(module_name)
-                    if hasattr(pipeline_module, "EntryClass"):
-                        entry_cls = pipeline_module.EntryClass
-                        if not isinstance(entry_cls, list):
-                            entry_cls_list = [entry_cls]
-                        else:
-                            entry_cls_list = entry_cls
-
-                        for cls in entry_cls_list:
-                            if hasattr(cls, "pipeline_name"):
-                                if cls.pipeline_name in _PIPELINE_REGISTRY:
-                                    logger.warning(
-                                        f"Duplicate pipeline name '{cls.pipeline_name}' found. Overwriting."
-                                    )
-                                _PIPELINE_REGISTRY[cls.pipeline_name] = cls
-                            # else:
-                            #     logger.warning(
-                            #         f"Pipeline class {cls.__name__} does not have a 'pipeline_name' attribute."
-                            #     )
+            pipeline_module = importlib.import_module(module_name)
+            if hasattr(pipeline_module, "EntryClass"):
+                entry_cls = pipeline_module.EntryClass
+                entry_cls_list = (
+                    [entry_cls] if not isinstance(entry_cls, list) else entry_cls
+                )
+
+                for cls in entry_cls_list:
+                    if hasattr(cls, "pipeline_name"):
+                        if cls.pipeline_name in _PIPELINE_REGISTRY:
+                            logger.warning(
+                                f"Duplicate pipeline name '{cls.pipeline_name}' found. Overwriting."
+                            )
+                        _PIPELINE_REGISTRY[cls.pipeline_name] = cls
+    logger.debug(
+        f"Registering pipelines complete, {len(_PIPELINE_REGISTRY)} pipelines registered"
+    )
 
 
 # --- Part 2: Config Registration ---
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/__init__.py
deleted file mode 100644
index 9f0e4bc62392..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/__init__.py
+++ /dev/null
@@ -1,8 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
-
-# SPDX-License-Identifier: Apache-2.0
-"""
-Basic inference pipelines for sglang.multimodal_gen.
-
-This package contains basic pipelines for video and image generation.
-"""
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/flux/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/flux/__init__.py
deleted file mode 100644
index af2eb7d103a8..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/flux/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/__init__.py
deleted file mode 100644
index af2eb7d103a8..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/__init__.py
deleted file mode 100644
index af2eb7d103a8..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/__init__.py
deleted file mode 100644
index af2eb7d103a8..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/__init__.py b/python/sglang/multimodal_gen/runtime/architectures/basic/wan/__init__.py
deleted file mode 100644
index af2eb7d103a8..000000000000
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/__init__.py
+++ /dev/null
@@ -1 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
diff --git a/python/sglang/multimodal_gen/runtime/entrypoints/diffusion_generator.py b/python/sglang/multimodal_gen/runtime/entrypoints/diffusion_generator.py
index 1c376634a9c3..787379f788ce 100644
--- a/python/sglang/multimodal_gen/runtime/entrypoints/diffusion_generator.py
+++ b/python/sglang/multimodal_gen/runtime/entrypoints/diffusion_generator.py
@@ -21,8 +21,8 @@
 import torchvision
 from einops import rearrange
 
-from sglang.multimodal_gen.runtime.pipelines import Req
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import OutputBatch
+from sglang.multimodal_gen.runtime.pipelines_core import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import OutputBatch
 
 # Suppress verbose logging from imageio, which is triggered when saving images.
 logging.getLogger("imageio").setLevel(logging.WARNING)
diff --git a/python/sglang/multimodal_gen/runtime/entrypoints/openai/image_api.py b/python/sglang/multimodal_gen/runtime/entrypoints/openai/image_api.py
index 07f67a5a1786..34ee3fc55389 100644
--- a/python/sglang/multimodal_gen/runtime/entrypoints/openai/image_api.py
+++ b/python/sglang/multimodal_gen/runtime/entrypoints/openai/image_api.py
@@ -24,7 +24,7 @@
     post_process_sample,
 )
 from sglang.multimodal_gen.runtime.entrypoints.utils import prepare_request
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
 from sglang.multimodal_gen.runtime.scheduler_client import scheduler_client
 from sglang.multimodal_gen.runtime.server_args import get_global_server_args
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
diff --git a/python/sglang/multimodal_gen/runtime/entrypoints/openai/video_api.py b/python/sglang/multimodal_gen/runtime/entrypoints/openai/video_api.py
index ad6b6f000694..267a6a213a39 100644
--- a/python/sglang/multimodal_gen/runtime/entrypoints/openai/video_api.py
+++ b/python/sglang/multimodal_gen/runtime/entrypoints/openai/video_api.py
@@ -34,7 +34,7 @@
     post_process_sample,
 )
 from sglang.multimodal_gen.runtime.entrypoints.utils import prepare_request
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
 from sglang.multimodal_gen.runtime.server_args import get_global_server_args
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/entrypoints/utils.py b/python/sglang/multimodal_gen/runtime/entrypoints/utils.py
index a4f8744697fb..7dc5f19e7e2c 100644
--- a/python/sglang/multimodal_gen/runtime/entrypoints/utils.py
+++ b/python/sglang/multimodal_gen/runtime/entrypoints/utils.py
@@ -16,7 +16,7 @@
 logging.getLogger("imageio_ffmpeg").setLevel(logging.WARNING)
 
 from sglang.multimodal_gen.configs.sample.base import DataType, SamplingParams
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 from sglang.multimodal_gen.utils import shallow_asdict
diff --git a/python/sglang/multimodal_gen/runtime/managers/forward_context.py b/python/sglang/multimodal_gen/runtime/managers/forward_context.py
index 0fcf747f44cd..e506929c6fdb 100644
--- a/python/sglang/multimodal_gen/runtime/managers/forward_context.py
+++ b/python/sglang/multimodal_gen/runtime/managers/forward_context.py
@@ -14,7 +14,7 @@
 
 if TYPE_CHECKING:
     from sglang.multimodal_gen.runtime.layers.attention import AttentionMetadata
-    from sglang.multimodal_gen.runtime.pipelines import Req
+    from sglang.multimodal_gen.runtime.pipelines_core import Req
 
 logger = init_logger(__name__)
 
diff --git a/python/sglang/multimodal_gen/runtime/managers/gpu_worker.py b/python/sglang/multimodal_gen/runtime/managers/gpu_worker.py
index 7d2152846581..ddf9dbc780fa 100644
--- a/python/sglang/multimodal_gen/runtime/managers/gpu_worker.py
+++ b/python/sglang/multimodal_gen/runtime/managers/gpu_worker.py
@@ -16,8 +16,8 @@
     get_cfg_group,
     get_tp_group,
 )
-from sglang.multimodal_gen.runtime.pipelines import Req, build_pipeline
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import OutputBatch
+from sglang.multimodal_gen.runtime.pipelines_core import Req, build_pipeline
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import OutputBatch
 from sglang.multimodal_gen.runtime.server_args import PortArgs, ServerArgs
 from sglang.multimodal_gen.runtime.utils.common import set_cuda_arch
 from sglang.multimodal_gen.runtime.utils.logging_utils import (
diff --git a/python/sglang/multimodal_gen/runtime/managers/scheduler.py b/python/sglang/multimodal_gen/runtime/managers/scheduler.py
index dffd5623921f..e7789a4b500b 100644
--- a/python/sglang/multimodal_gen/runtime/managers/scheduler.py
+++ b/python/sglang/multimodal_gen/runtime/managers/scheduler.py
@@ -6,7 +6,7 @@
 import zmq
 
 from sglang.multimodal_gen.runtime.managers.gpu_worker import GPUWorker
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import OutputBatch
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import OutputBatch
 from sglang.multimodal_gen.runtime.server_args import (
     PortArgs,
     ServerArgs,
diff --git a/python/sglang/multimodal_gen/runtime/managers/schedulerbase.py b/python/sglang/multimodal_gen/runtime/managers/schedulerbase.py
index 208ff5fbb6d0..a2da3cc75253 100644
--- a/python/sglang/multimodal_gen/runtime/managers/schedulerbase.py
+++ b/python/sglang/multimodal_gen/runtime/managers/schedulerbase.py
@@ -6,8 +6,8 @@
 
 import zmq
 
-from sglang.multimodal_gen.runtime.pipelines import Req
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import OutputBatch
+from sglang.multimodal_gen.runtime.pipelines_core import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import OutputBatch
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/__init__.py b/python/sglang/multimodal_gen/runtime/pipelines/__init__.py
index fdb1acbfa182..af2eb7d103a8 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/__init__.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/__init__.py
@@ -1,64 +1 @@
 # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
-
-# SPDX-License-Identifier: Apache-2.0
-"""
-Diffusion pipelines for sglang.multimodal_gen.
-
-This package contains diffusion pipelines for generating videos and images.
-"""
-
-from typing import cast
-
-from sglang.multimodal_gen.registry import get_model_info
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
-    ComposedPipelineBase,
-)
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.server_args import ServerArgs
-from sglang.multimodal_gen.runtime.utils.hf_diffusers_utils import (
-    maybe_download_model,
-    verify_model_config_and_directory,
-)
-from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
-
-logger = init_logger(__name__)
-
-
-class PipelineWithLoRA(LoRAPipeline, ComposedPipelineBase):
-    """Type for a pipeline that has both ComposedPipelineBase and LoRAPipeline functionality."""
-
-    pass
-
-
-def build_pipeline(
-    server_args: ServerArgs,
-) -> PipelineWithLoRA:
-    """
-    Only works with valid hf diffusers configs. (model_index.json)
-    We want to build a pipeline based on the inference args mode_path:
-    1. download the model from the hub if it's not already downloaded
-    2. verify the model config and directory
-    3. based on the config, determine the pipeline class
-    """
-    model_path = server_args.model_path
-    model_info = get_model_info(model_path)
-    if model_info is None:
-        raise ValueError(f"Unsupported model: {model_path}")
-
-    pipeline_cls = model_info.pipeline_cls
-
-    # instantiate the pipelines
-    pipeline = pipeline_cls(model_path, server_args)
-
-    logger.info("Pipelines instantiated")
-
-    return cast(PipelineWithLoRA, pipeline)
-
-
-__all__ = [
-    "build_pipeline",
-    "ComposedPipelineBase",
-    "Req",
-    "LoRAPipeline",
-]
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/flux/flux.py b/python/sglang/multimodal_gen/runtime/pipelines/flux.py
similarity index 93%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/flux/flux.py
rename to python/sglang/multimodal_gen/runtime/pipelines/flux.py
index b56947b42780..7ede4b388f1a 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/flux/flux.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/flux.py
@@ -1,11 +1,11 @@
 # Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
 
 # SPDX-License-Identifier: Apache-2.0
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ConditioningStage,
     DecodingStage,
     DenoisingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/hunyuan_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/hunyuan_pipeline.py
similarity index 94%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/hunyuan_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/hunyuan_pipeline.py
index 8edf842c079a..0be68fc4dd0c 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/hunyuan/hunyuan_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/hunyuan_pipeline.py
@@ -9,10 +9,10 @@
 """
 
 
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ConditioningStage,
     DecodingStage,
     DenoisingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/qwen_image.py b/python/sglang/multimodal_gen/runtime/pipelines/qwen_image.py
similarity index 94%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/qwen_image.py
rename to python/sglang/multimodal_gen/runtime/pipelines/qwen_image.py
index 0682c467fb57..e74ebb8ae25a 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/qwen_image/qwen_image.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/qwen_image.py
@@ -3,11 +3,11 @@
 # SPDX-License-Identifier: Apache-2.0
 from diffusers.image_processor import VaeImageProcessor
 
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     DecodingStage,
     DenoisingStage,
     ImageEncodingStage,
@@ -17,7 +17,7 @@
     TextEncodingStage,
     TimestepPreparationStage,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.conditioning import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages.conditioning import (
     ConditioningStage,
 )
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/__init__.py b/python/sglang/multimodal_gen/runtime/pipelines/stages/__init__.py
deleted file mode 100644
index 062d4cd8e6d0..000000000000
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/__init__.py
+++ /dev/null
@@ -1,59 +0,0 @@
-# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
-
-# SPDX-License-Identifier: Apache-2.0
-"""
-Pipeline stages for diffusion models.
-
-This package contains the various stages that can be composed to create
-complete diffusion pipelines.
-"""
-
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.causal_denoising import (
-    CausalDMDDenoisingStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.conditioning import (
-    ConditioningStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.decoding import DecodingStage
-from sglang.multimodal_gen.runtime.pipelines.stages.denoising import DenoisingStage
-from sglang.multimodal_gen.runtime.pipelines.stages.denoising_dmd import (
-    DmdDenoisingStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.encoding import EncodingStage
-from sglang.multimodal_gen.runtime.pipelines.stages.image_encoding import (
-    ImageEncodingStage,
-    ImageVAEEncodingStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.input_validation import (
-    InputValidationStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.latent_preparation import (
-    LatentPreparationStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.stepvideo_encoding import (
-    StepvideoPromptEncodingStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.text_encoding import (
-    TextEncodingStage,
-)
-from sglang.multimodal_gen.runtime.pipelines.stages.timestep_preparation import (
-    TimestepPreparationStage,
-)
-
-__all__ = [
-    "PipelineStage",
-    "InputValidationStage",
-    "TimestepPreparationStage",
-    "LatentPreparationStage",
-    "ConditioningStage",
-    "DenoisingStage",
-    "DmdDenoisingStage",
-    "CausalDMDDenoisingStage",
-    "EncodingStage",
-    "DecodingStage",
-    "ImageEncodingStage",
-    "ImageVAEEncodingStage",
-    "TextEncodingStage",
-    "StepvideoPromptEncodingStage",
-]
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/stepvideo_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/stepvideo_pipeline.py
similarity index 96%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/stepvideo_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/stepvideo_pipeline.py
index bb0e4ef9fe19..9d2e80c23673 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/stepvideo/stepvideo_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/stepvideo_pipeline.py
@@ -24,11 +24,11 @@
     HunyuanClip,  # type: ignore
 )
 from sglang.multimodal_gen.runtime.models.encoders.stepllm import STEP1TextEncoder
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     DecodingStage,
     DenoisingStage,
     InputValidationStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_causal_dmd_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/wan_causal_dmd_pipeline.py
similarity index 90%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_causal_dmd_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/wan_causal_dmd_pipeline.py
index 31b5bf04f03f..b103ee0a5e78 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_causal_dmd_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/wan_causal_dmd_pipeline.py
@@ -7,13 +7,13 @@
 This module wires the causal DMD denoising stage into the modular pipeline.
 """
 
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
 
 # isort: off
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ConditioningStage,
     DecodingStage,
     CausalDMDDenoisingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_dmd_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/wan_dmd_pipeline.py
similarity index 92%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_dmd_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/wan_dmd_pipeline.py
index 984812f1fd16..3c973834cafb 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_dmd_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/wan_dmd_pipeline.py
@@ -11,15 +11,15 @@
 from sglang.multimodal_gen.runtime.models.schedulers.scheduling_flow_match_euler_discrete import (
     FlowMatchEulerDiscreteScheduler,
 )
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
 # isort: off
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ConditioningStage,
     DecodingStage,
     DmdDenoisingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_dmd_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_dmd_pipeline.py
similarity index 93%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_dmd_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_dmd_pipeline.py
index ed3fdd4eaad8..7a49eef9b4b3 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_dmd_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_dmd_pipeline.py
@@ -8,15 +8,15 @@
 using the modular pipeline architecture.
 """
 
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
 # isort: off
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ImageEncodingStage,
     ConditioningStage,
     DecodingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_pipeline.py
similarity index 93%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_pipeline.py
index 9477d4009da5..93a1968704da 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_i2v_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/wan_i2v_pipeline.py
@@ -8,15 +8,15 @@
 using the modular pipeline architecture.
 """
 
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
 # isort: off
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ImageEncodingStage,
     ConditioningStage,
     DecodingStage,
diff --git a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines/wan_pipeline.py
similarity index 92%
rename from python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines/wan_pipeline.py
index 9d7c592bd88b..8f1cbfc26870 100644
--- a/python/sglang/multimodal_gen/runtime/architectures/basic/wan/wan_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines/wan_pipeline.py
@@ -11,11 +11,11 @@
 from sglang.multimodal_gen.runtime.models.schedulers.scheduling_flow_unipc_multistep import (
     FlowUniPCMultistepScheduler,
 )
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
-from sglang.multimodal_gen.runtime.pipelines.lora_pipeline import LoRAPipeline
-from sglang.multimodal_gen.runtime.pipelines.stages import (
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.stages import (
     ConditioningStage,
     DecodingStage,
     DenoisingStage,
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/README.md b/python/sglang/multimodal_gen/runtime/pipelines_core/README.md
similarity index 100%
rename from python/sglang/multimodal_gen/runtime/pipelines/README.md
rename to python/sglang/multimodal_gen/runtime/pipelines_core/README.md
diff --git a/python/sglang/multimodal_gen/runtime/pipelines_core/__init__.py b/python/sglang/multimodal_gen/runtime/pipelines_core/__init__.py
new file mode 100644
index 000000000000..feda424e54fa
--- /dev/null
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/__init__.py
@@ -0,0 +1,64 @@
+# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
+
+# SPDX-License-Identifier: Apache-2.0
+"""
+Diffusion pipelines for sglang.multimodal_gen.
+
+This package contains diffusion pipelines for generating videos and images.
+"""
+
+from typing import cast
+
+from sglang.multimodal_gen.registry import get_model_info
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
+    ComposedPipelineBase,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.lora_pipeline import LoRAPipeline
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.server_args import ServerArgs
+from sglang.multimodal_gen.runtime.utils.hf_diffusers_utils import (
+    maybe_download_model,
+    verify_model_config_and_directory,
+)
+from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
+
+logger = init_logger(__name__)
+
+
+class PipelineWithLoRA(LoRAPipeline, ComposedPipelineBase):
+    """Type for a pipeline that has both ComposedPipelineBase and LoRAPipeline functionality."""
+
+    pass
+
+
+def build_pipeline(
+    server_args: ServerArgs,
+) -> PipelineWithLoRA:
+    """
+    Only works with valid hf diffusers configs. (model_index.json)
+    We want to build a pipeline based on the inference args mode_path:
+    1. download the model from the hub if it's not already downloaded
+    2. verify the model config and directory
+    3. based on the config, determine the pipeline class
+    """
+    model_path = server_args.model_path
+    model_info = get_model_info(model_path)
+    if model_info is None:
+        raise ValueError(f"Unsupported model: {model_path}")
+
+    pipeline_cls = model_info.pipeline_cls
+
+    # instantiate the pipelines
+    pipeline = pipeline_cls(model_path, server_args)
+
+    logger.info("Pipelines instantiated")
+
+    return cast(PipelineWithLoRA, pipeline)
+
+
+__all__ = [
+    "build_pipeline",
+    "ComposedPipelineBase",
+    "Req",
+    "LoRAPipeline",
+]
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/composed_pipeline_base.py b/python/sglang/multimodal_gen/runtime/pipelines_core/composed_pipeline_base.py
similarity index 96%
rename from python/sglang/multimodal_gen/runtime/pipelines/composed_pipeline_base.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/composed_pipeline_base.py
index d3f2c8fae655..b9d22fc46346 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/composed_pipeline_base.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/composed_pipeline_base.py
@@ -15,15 +15,15 @@
 import torch
 from tqdm import tqdm
 
-from sglang.multimodal_gen.configs.pipelines import PipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs import PipelineConfig
 from sglang.multimodal_gen.runtime.loader.component_loader import (
     PipelineComponentLoader,
 )
-from sglang.multimodal_gen.runtime.pipelines.executors.pipeline_executor import (
+from sglang.multimodal_gen.runtime.pipelines_core.executors.pipeline_executor import (
     PipelineExecutor,
 )
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import PipelineStage
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.hf_diffusers_utils import (
     maybe_download_model,
@@ -90,7 +90,7 @@ def __init__(
 
     def build_executor(self, server_args: ServerArgs):
         # TODO
-        from sglang.multimodal_gen.runtime.pipelines.executors.parallel_executor import (
+        from sglang.multimodal_gen.runtime.pipelines_core.executors.parallel_executor import (
             ParallelExecutor,
         )
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/executors/parallel_executor.py b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/parallel_executor.py
similarity index 93%
rename from python/sglang/multimodal_gen/runtime/pipelines/executors/parallel_executor.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/executors/parallel_executor.py
index a1a9d88fa9ef..dabe531c666f 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/executors/parallel_executor.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/parallel_executor.py
@@ -9,12 +9,12 @@
     get_cfg_group,
     get_classifier_free_guidance_rank,
 )
-from sglang.multimodal_gen.runtime.pipelines import Req
-from sglang.multimodal_gen.runtime.pipelines.executors.pipeline_executor import (
+from sglang.multimodal_gen.runtime.pipelines_core import Req
+from sglang.multimodal_gen.runtime.pipelines_core.executors.pipeline_executor import (
     PipelineExecutor,
     Timer,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.base import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import (
     PipelineStage,
     StageParallelismType,
 )
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/executors/pipeline_executor.py b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/pipeline_executor.py
similarity index 91%
rename from python/sglang/multimodal_gen/runtime/pipelines/executors/pipeline_executor.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/executors/pipeline_executor.py
index d826f7cd919d..3a3ec051a764 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/executors/pipeline_executor.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/pipeline_executor.py
@@ -8,8 +8,8 @@
 from abc import ABC, abstractmethod
 from typing import List
 
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import PipelineStage
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/executors/sync_executor.py b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/sync_executor.py
similarity index 78%
rename from python/sglang/multimodal_gen/runtime/pipelines/executors/sync_executor.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/executors/sync_executor.py
index 6368593d2b7f..5e429b3e8768 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/executors/sync_executor.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/executors/sync_executor.py
@@ -6,13 +6,13 @@
 """
 from typing import List
 
-from sglang.multimodal_gen.runtime.pipelines.executors.pipeline_executor import (
+from sglang.multimodal_gen.runtime.pipelines_core.executors.pipeline_executor import (
     PipelineExecutor,
     Timer,
     logger,
 )
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import PipelineStage
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/lora_pipeline.py b/python/sglang/multimodal_gen/runtime/pipelines_core/lora_pipeline.py
similarity index 99%
rename from python/sglang/multimodal_gen/runtime/pipelines/lora_pipeline.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/lora_pipeline.py
index 4e7bc0901c6e..6f5c52a0494e 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/lora_pipeline.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/lora_pipeline.py
@@ -16,7 +16,7 @@
     replace_submodule,
 )
 from sglang.multimodal_gen.runtime.loader.utils import get_param_names_mapping
-from sglang.multimodal_gen.runtime.pipelines.composed_pipeline_base import (
+from sglang.multimodal_gen.runtime.pipelines_core.composed_pipeline_base import (
     ComposedPipelineBase,
 )
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/schedule_batch.py b/python/sglang/multimodal_gen/runtime/pipelines_core/schedule_batch.py
similarity index 100%
rename from python/sglang/multimodal_gen/runtime/pipelines/schedule_batch.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/schedule_batch.py
diff --git a/python/sglang/multimodal_gen/runtime/pipelines_core/stages/__init__.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/__init__.py
new file mode 100644
index 000000000000..3c962b0b970d
--- /dev/null
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/__init__.py
@@ -0,0 +1,59 @@
+# Copied and adapted from: https://github.com/hao-ai-lab/FastVideo
+
+# SPDX-License-Identifier: Apache-2.0
+"""
+Pipeline stages for diffusion models.
+
+This package contains the various stages that can be composed to create
+complete diffusion pipelines.
+"""
+
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.causal_denoising import (
+    CausalDMDDenoisingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.conditioning import (
+    ConditioningStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.decoding import DecodingStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.denoising import DenoisingStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.denoising_dmd import (
+    DmdDenoisingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.encoding import EncodingStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.image_encoding import (
+    ImageEncodingStage,
+    ImageVAEEncodingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.input_validation import (
+    InputValidationStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.latent_preparation import (
+    LatentPreparationStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.stepvideo_encoding import (
+    StepvideoPromptEncodingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.text_encoding import (
+    TextEncodingStage,
+)
+from sglang.multimodal_gen.runtime.pipelines_core.stages.timestep_preparation import (
+    TimestepPreparationStage,
+)
+
+__all__ = [
+    "PipelineStage",
+    "InputValidationStage",
+    "TimestepPreparationStage",
+    "LatentPreparationStage",
+    "ConditioningStage",
+    "DenoisingStage",
+    "DmdDenoisingStage",
+    "CausalDMDDenoisingStage",
+    "EncodingStage",
+    "DecodingStage",
+    "ImageEncodingStage",
+    "ImageVAEEncodingStage",
+    "TextEncodingStage",
+    "StepvideoPromptEncodingStage",
+]
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/base.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/base.py
similarity index 98%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/base.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/base.py
index 8601eb1e1d13..5f66ddaa9d37 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/base.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/base.py
@@ -16,8 +16,10 @@
 import torch
 
 import sglang.multimodal_gen.envs as envs
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs, get_global_server_args
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/causal_denoising.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/causal_denoising.py
similarity index 98%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/causal_denoising.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/causal_denoising.py
index 27b5ac2a6016..a00ef3f8dcf2 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/causal_denoising.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/causal_denoising.py
@@ -5,12 +5,14 @@
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
 from sglang.multimodal_gen.runtime.managers.forward_context import set_forward_context
 from sglang.multimodal_gen.runtime.models.utils import pred_noise_to_pred_video
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.denoising import DenoisingStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.denoising import DenoisingStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/conditioning.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/conditioning.py
similarity index 90%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/conditioning.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/conditioning.py
index 38a6fd3892e5..284dac49e3ae 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/conditioning.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/conditioning.py
@@ -7,12 +7,14 @@
 
 import torch
 
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/decoding.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/decoding.py
similarity index 96%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/decoding.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/decoding.py
index f4198bf5dcfe..d54e64f6a130 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/decoding.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/decoding.py
@@ -10,19 +10,21 @@
 import torch
 
 from sglang.multimodal_gen.configs.models.vaes.base import VAEArchConfig
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
     QwenImagePipelineConfig,
 )
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
 from sglang.multimodal_gen.runtime.loader.component_loader import VAELoader
 from sglang.multimodal_gen.runtime.models.vaes.common import ParallelTiledVAE
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import OutputBatch, Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import OutputBatch, Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import (
     PipelineStage,
     StageParallelismType,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs, get_global_server_args
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 from sglang.multimodal_gen.utils import PRECISION_TO_TYPE
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/denoising.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising.py
similarity index 99%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/denoising.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising.py
index 1c99da75e34f..bb0c6c3a5745 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/denoising.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising.py
@@ -19,7 +19,7 @@
 from einops import rearrange
 from tqdm.auto import tqdm
 
-from sglang.multimodal_gen.configs.pipelines.base import ModelTaskType, STA_Mode
+from sglang.multimodal_gen.configs.pipeline_configs.base import ModelTaskType, STA_Mode
 from sglang.multimodal_gen.runtime.distributed import (
     cfg_model_parallel_all_reduce,
     get_local_torch_device,
@@ -44,15 +44,17 @@
 )
 from sglang.multimodal_gen.runtime.loader.component_loader import TransformerLoader
 from sglang.multimodal_gen.runtime.managers.forward_context import set_forward_context
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import (
     PipelineStage,
     StageParallelismType,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.platforms.interface import AttentionBackendEnum
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/denoising_dmd.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising_dmd.py
similarity index 97%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/denoising_dmd.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising_dmd.py
index 47516fa70c92..1dd64a67cc61 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/denoising_dmd.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising_dmd.py
@@ -22,9 +22,9 @@
     FlowMatchEulerDiscreteScheduler,
 )
 from sglang.multimodal_gen.runtime.models.utils import pred_noise_to_pred_video
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages import DenoisingStage
-from sglang.multimodal_gen.runtime.pipelines.stages.denoising import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages import DenoisingStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.denoising import (
     st_attn_available,
     vsa_available,
 )
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/encoding.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/encoding.py
similarity index 91%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/encoding.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/encoding.py
index e06937f7a24e..91fe9c43d8df 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/encoding.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/encoding.py
@@ -9,12 +9,14 @@
 
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
 from sglang.multimodal_gen.runtime.models.vaes.common import ParallelTiledVAE
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     V,  # Import validators
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 from sglang.multimodal_gen.utils import PRECISION_TO_TYPE
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/image_encoding.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/image_encoding.py
similarity index 97%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/image_encoding.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/image_encoding.py
index 421eff15c12a..53ec3d1c2248 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/image_encoding.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/image_encoding.py
@@ -10,7 +10,7 @@
 import PIL
 import torch
 
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
     QwenImagePipelineConfig,
     _pack_latents,
@@ -25,12 +25,14 @@
     pil_to_numpy,
     resize,
 )
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ExecutionMode, ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 from sglang.multimodal_gen.utils import PRECISION_TO_TYPE
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/input_validation.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py
similarity index 94%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/input_validation.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py
index b992c7791b74..c92942e46502 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/input_validation.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py
@@ -9,15 +9,15 @@
 import torchvision.transforms.functional as TF
 from PIL import Image
 
-from sglang.multimodal_gen.configs.pipelines import WanI2V480PConfig
-from sglang.multimodal_gen.configs.pipelines.base import ModelTaskType
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs import WanI2V480PConfig
+from sglang.multimodal_gen.configs.pipeline_configs.base import ModelTaskType
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
 )
 from sglang.multimodal_gen.runtime.models.vision_utils import load_image, load_video
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators,
     VerificationResult,
 )
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/latent_preparation.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation.py
similarity index 94%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/latent_preparation.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation.py
index 1bfccf0f4892..34d7d82d45dc 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/latent_preparation.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/latent_preparation.py
@@ -7,12 +7,14 @@
 from diffusers.utils.torch_utils import randn_tensor
 
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/stepvideo_encoding.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/stepvideo_encoding.py
similarity index 90%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/stepvideo_encoding.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/stepvideo_encoding.py
index 81e39d7b6116..eed0edd99875 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/stepvideo_encoding.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/stepvideo_encoding.py
@@ -5,12 +5,14 @@
 import torch
 
 from sglang.multimodal_gen.runtime.managers.forward_context import set_forward_context
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/text_encoding.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/text_encoding.py
similarity index 96%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/text_encoding.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/text_encoding.py
index cf6c342f2a77..47a5ef3c6518 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/text_encoding.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/text_encoding.py
@@ -10,15 +10,17 @@
 import torch
 
 from sglang.multimodal_gen.configs.models.encoders import BaseEncoderOutput
-from sglang.multimodal_gen.configs.pipelines import FluxPipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs import FluxPipelineConfig
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
 from sglang.multimodal_gen.runtime.managers.forward_context import set_forward_context
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import PipelineStage
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import PipelineStage
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/timestep_preparation.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/timestep_preparation.py
similarity index 92%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/timestep_preparation.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/timestep_preparation.py
index 28275f11ceb5..91bc6e69532a 100644
--- a/python/sglang/multimodal_gen/runtime/pipelines/stages/timestep_preparation.py
+++ b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/timestep_preparation.py
@@ -12,21 +12,23 @@
 
 import numpy as np
 
-from sglang.multimodal_gen.configs.pipelines import FluxPipelineConfig
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs import FluxPipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
     QwenImagePipelineConfig,
 )
 from sglang.multimodal_gen.runtime.distributed import get_local_torch_device
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
-from sglang.multimodal_gen.runtime.pipelines.stages.base import (
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.stages.base import (
     PipelineStage,
     StageParallelismType,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import (
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
     StageValidators as V,
 )
-from sglang.multimodal_gen.runtime.pipelines.stages.validators import VerificationResult
+from sglang.multimodal_gen.runtime.pipelines_core.stages.validators import (
+    VerificationResult,
+)
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/pipelines/stages/validators.py b/python/sglang/multimodal_gen/runtime/pipelines_core/stages/validators.py
similarity index 100%
rename from python/sglang/multimodal_gen/runtime/pipelines/stages/validators.py
rename to python/sglang/multimodal_gen/runtime/pipelines_core/stages/validators.py
diff --git a/python/sglang/multimodal_gen/runtime/scheduler_client.py b/python/sglang/multimodal_gen/runtime/scheduler_client.py
index 5569ae209f8d..4b70de3fb031 100644
--- a/python/sglang/multimodal_gen/runtime/scheduler_client.py
+++ b/python/sglang/multimodal_gen/runtime/scheduler_client.py
@@ -5,7 +5,7 @@
 import zmq
 import zmq.asyncio
 
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger
 
diff --git a/python/sglang/multimodal_gen/runtime/server_args.py b/python/sglang/multimodal_gen/runtime/server_args.py
index 5fc85b3d4e59..a6d51434bc86 100644
--- a/python/sglang/multimodal_gen/runtime/server_args.py
+++ b/python/sglang/multimodal_gen/runtime/server_args.py
@@ -15,9 +15,9 @@
 from enum import Enum
 from typing import Any, Optional
 
-from sglang.multimodal_gen.configs.pipelines import FluxPipelineConfig
-from sglang.multimodal_gen.configs.pipelines.base import PipelineConfig, STA_Mode
-from sglang.multimodal_gen.configs.pipelines.qwen_image import (
+from sglang.multimodal_gen.configs.pipeline_configs import FluxPipelineConfig
+from sglang.multimodal_gen.configs.pipeline_configs.base import PipelineConfig, STA_Mode
+from sglang.multimodal_gen.configs.pipeline_configs.qwen_image import (
     QwenImageEditPipelineConfig,
     QwenImagePipelineConfig,
 )
diff --git a/python/sglang/multimodal_gen/runtime/sync_scheduler_client.py b/python/sglang/multimodal_gen/runtime/sync_scheduler_client.py
index 2e246fb82027..8e89728f6ebe 100644
--- a/python/sglang/multimodal_gen/runtime/sync_scheduler_client.py
+++ b/python/sglang/multimodal_gen/runtime/sync_scheduler_client.py
@@ -2,7 +2,7 @@
 
 import zmq
 
-from sglang.multimodal_gen.runtime.pipelines.schedule_batch import Req
+from sglang.multimodal_gen.runtime.pipelines_core.schedule_batch import Req
 from sglang.multimodal_gen.runtime.server_args import ServerArgs
 from sglang.multimodal_gen.runtime.utils.logging_utils import init_logger