chore: unify filesystem usage across all frameworks and workers (#6391)

Signed-off-by: ayushag <ayushag@nvidia.com>

.gitignore

@@ -2,6 +2,8 @@
 .vs/
 .vscode/
 .helix
+issues/
+plans/
 *rendered.Dockerfile
 [Bb]inlog/
 [Bb][Uu][Ii][Ll][Dd]/

components/src/dynamo/common/configuration/groups/runtime_args.py

@@ -31,6 +31,8 @@ class DynamoRuntimeConfig(ConfigBase):
     dump_config_to: Optional[str] = None
     multimodal_embedding_cache_capacity_gb: float
     output_modalities: List[str]
+    media_output_fs_url: str = "file:///tmp/dynamo_media"
+    media_output_http_url: Optional[str] = None
 
     def validate(self) -> None:
         # TODO  get a better way for spot fixes like this.
@@ -176,3 +178,19 @@ class DynamoRuntimeArgGroup(ArgGroup):
             help="Output modalities for omni/diffusion mode (e.g., --output-modalities text image audio video).",
             nargs="*",
         )
+
+        # Media storage (generated images and videos)
+        add_argument(
+            g,
+            flag_name="--media-output-fs-url",
+            env_var="DYN_MEDIA_OUTPUT_FS_URL",
+            default="file:///tmp/dynamo_media",
+            help="Filesystem URL for storing generated images and videos (e.g. file:///tmp/dynamo_media, s3://bucket/path).",
+        )
+        add_argument(
+            g,
+            flag_name="--media-output-http-url",
+            env_var="DYN_MEDIA_OUTPUT_HTTP_URL",
+            default=None,
+            help="Base URL for rewriting media file paths in responses (e.g. http://localhost:8000/media). If unset, returns raw filesystem paths.",
+        )

components/src/dynamo/common/storage.py

@@ -32,6 +32,9 @@ S3:
 
 
 """
+import asyncio
+from typing import Optional
+
 import fsspec
 from fsspec.implementations.dirfs import DirFileSystem
 
@@ -64,3 +67,51 @@ def get_fs(fs_url: str) -> DirFileSystem:
         fs_opts = {"auto_mkdir": True}
 
     return DirFileSystem(fs=fsspec.filesystem(protocol, **fs_opts), path=root_path)
+
+
+def get_media_url(
+    fs: DirFileSystem, storage_path: str, base_url: Optional[str] = None
+) -> str:
+    """Build a public URL for a file stored in the media filesystem.
+
+    Args:
+        fs: The DirFileSystem returned by ``get_fs()``.
+        storage_path: Relative path within the filesystem (e.g. "videos/req-id.mp4").
+        base_url: Optional CDN / proxy base URL.  When set, the returned URL is
+            ``{base_url}/{storage_path}``.  When *None*, the URL is constructed
+            from the filesystem's protocol and root path.
+
+    Returns:
+        Public URL string for the uploaded file.
+    """
+    if base_url:
+        return f"{base_url.rstrip('/')}/{storage_path}"
+
+    protocol = fs.fs.protocol
+    if isinstance(protocol, (list, tuple)):
+        protocol = protocol[0]
+    return f"{protocol}://{fs.path}/{storage_path}"
+
+
+async def upload_to_fs(
+    fs: DirFileSystem,
+    storage_path: str,
+    data: bytes,
+    base_url: Optional[str] = None,
+) -> str:
+    """Upload bytes to the media filesystem and return the public URL.
+
+    This is the canonical helper for all backends (vLLM, SGLang, TRT-LLM)
+    to store generated images/videos and produce a response URL.
+
+    Args:
+        fs: The DirFileSystem returned by ``get_fs()``.
+        storage_path: Relative path within the filesystem (e.g. "images/req-id/file.png").
+        data: Raw bytes to upload.
+        base_url: Optional CDN / proxy base URL for URL rewriting.
+
+    Returns:
+        Public URL string for the uploaded file.
+    """
+    await asyncio.to_thread(fs.pipe, storage_path, data)
+    return get_media_url(fs, storage_path, base_url)

components/src/dynamo/common/tests/test_storage.py

+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Unit tests for dynamo.common.storage module."""
+
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from dynamo.common.storage import get_fs, get_media_url, upload_to_fs
+
+pytestmark = [
+    pytest.mark.unit,
+    pytest.mark.gpu_0,
+    pytest.mark.pre_merge,
+]
+
+
+class TestGetFs:
+    """Tests for get_fs() filesystem initialization."""
+
+    def test_local_file_url(self, tmp_path):
+        """Test file:// URL returns DirFileSystem with correct path."""
+        media_dir = tmp_path / "test_media"
+        fs = get_fs(f"file://{media_dir}")
+        assert fs.path == str(media_dir)
+        protocol = fs.fs.protocol
+        if isinstance(protocol, (list, tuple)):
+            protocol = protocol[0]
+        assert protocol == "file"
+
+    def test_local_file_url_auto_mkdir(self, tmp_path):
+        """Test file:// URL enables auto_mkdir on underlying filesystem."""
+        media_dir = tmp_path / "test_media"
+        fs = get_fs(f"file://{media_dir}")
+        assert fs.fs.auto_mkdir is True
+
+    def test_no_protocol_defaults_to_file(self, tmp_path):
+        """Test URL without protocol defaults to file."""
+        media_dir = tmp_path / "test_media"
+        fs = get_fs(str(media_dir))
+        protocol = fs.fs.protocol
+        if isinstance(protocol, (list, tuple)):
+            protocol = protocol[0]
+        assert protocol == "file"
+
+    def test_s3_url_protocol(self):
+        """Test s3:// URL extracts correct protocol and bucket path."""
+        with patch("dynamo.common.storage.fsspec.filesystem") as mock_fsspec, patch(
+            "dynamo.common.storage.DirFileSystem"
+        ) as mock_dirfs:
+            mock_inner_fs = MagicMock(protocol="s3")
+            mock_fsspec.return_value = mock_inner_fs
+            get_fs("s3://my-bucket/prefix")
+            mock_fsspec.assert_called_once_with("s3")
+            mock_dirfs.assert_called_once_with(
+                fs=mock_inner_fs, path="my-bucket/prefix"
+            )
+
+    def test_gs_url_protocol(self):
+        """Test gs:// URL extracts correct protocol and path."""
+        with patch("dynamo.common.storage.fsspec.filesystem") as mock_fsspec, patch(
+            "dynamo.common.storage.DirFileSystem"
+        ) as mock_dirfs:
+            mock_inner_fs = MagicMock(protocol="gs")
+            mock_fsspec.return_value = mock_inner_fs
+            get_fs("gs://my-gcs-bucket/data")
+            mock_fsspec.assert_called_once_with("gs")
+            mock_dirfs.assert_called_once_with(
+                fs=mock_inner_fs, path="my-gcs-bucket/data"
+            )
+
+
+class TestGetMediaUrl:
+    """Tests for get_media_url() URL construction."""
+
+    def _make_fs(self, protocol="file", path="/tmp/media"):  # noqa: S108
+        """Create a mock DirFileSystem."""
+        fs = MagicMock()
+        fs.fs.protocol = protocol
+        fs.path = path
+        return fs
+
+    def test_base_url_rewrite(self):
+        """Test that base_url takes precedence over protocol fallback."""
+        fs = self._make_fs()
+        url = get_media_url(
+            fs, "videos/req-123.mp4", base_url="https://cdn.example.com/media"
+        )
+        assert url == "https://cdn.example.com/media/videos/req-123.mp4"
+
+    def test_base_url_trailing_slash_stripped(self):
+        """Test that trailing slash on base_url is normalized."""
+        fs = self._make_fs()
+        url = get_media_url(fs, "images/test.png", base_url="https://cdn.example.com/")
+        assert url == "https://cdn.example.com/images/test.png"
+
+    def test_protocol_fallback_file(self):
+        """Test URL construction from file:// protocol when no base_url."""
+        fs = self._make_fs(protocol="file", path="/tmp/dynamo_media")  # noqa: S108
+        url = get_media_url(fs, "videos/req-123.mp4")
+        assert url == "file:///tmp/dynamo_media/videos/req-123.mp4"
+
+    def test_protocol_fallback_s3(self):
+        """Test URL construction from s3:// protocol when no base_url."""
+        fs = self._make_fs(protocol="s3", path="my-bucket/prefix")
+        url = get_media_url(fs, "images/img.png")
+        assert url == "s3://my-bucket/prefix/images/img.png"
+
+    def test_tuple_protocol_uses_first(self):
+        """Test that tuple protocol (e.g., ('s3', 's3a')) uses first element."""
+        fs = self._make_fs()
+        fs.fs.protocol = ("s3", "s3a")
+        fs.path = "my-bucket"
+        url = get_media_url(fs, "file.mp4")
+        assert url == "s3://my-bucket/file.mp4"
+
+    def test_list_protocol_uses_first(self):
+        """Test that list protocol uses first element."""
+        fs = self._make_fs()
+        fs.fs.protocol = ["gs", "gcs"]
+        fs.path = "my-gcs-bucket"
+        url = get_media_url(fs, "file.mp4")
+        assert url == "gs://my-gcs-bucket/file.mp4"
+
+    def test_base_url_none_falls_back_to_protocol(self):
+        """Test that None base_url triggers protocol fallback."""
+        fs = self._make_fs()
+        url = get_media_url(fs, "test.png", base_url=None)
+        assert url == "file:///tmp/media/test.png"
+
+    def test_base_url_empty_string_falls_back_to_protocol(self):
+        """Test that empty string base_url triggers protocol fallback."""
+        fs = self._make_fs()
+        url = get_media_url(fs, "test.png", base_url="")
+        assert url == "file:///tmp/media/test.png"
+
+
+class TestUploadToFs:
+    """Tests for upload_to_fs() async upload + URL construction."""
+
+    def _make_fs(self, protocol="file", path="/tmp/media"):  # noqa: S108
+        """Create a mock DirFileSystem."""
+        fs = MagicMock()
+        fs.fs.protocol = protocol
+        fs.path = path
+        fs.pipe = MagicMock()
+        return fs
+
+    @pytest.mark.asyncio
+    async def test_calls_pipe_with_correct_args(self):
+        """Test that fs.pipe is called with storage_path and data."""
+        fs = self._make_fs()
+        data = b"test image bytes"
+
+        await upload_to_fs(fs, "images/test.png", data)
+
+        fs.pipe.assert_called_once_with("images/test.png", data)
+
+    @pytest.mark.asyncio
+    async def test_returns_url_with_base_url(self):
+        """Test that returned URL uses base_url when provided."""
+        fs = self._make_fs()
+        url = await upload_to_fs(
+            fs, "videos/req-123.mp4", b"video", base_url="https://cdn.example.com"
+        )
+        assert url == "https://cdn.example.com/videos/req-123.mp4"
+
+    @pytest.mark.asyncio
+    async def test_returns_url_with_protocol_fallback(self):
+        """Test that returned URL uses protocol when no base_url."""
+        fs = self._make_fs(protocol="s3", path="my-bucket/prefix")
+        url = await upload_to_fs(fs, "images/img.png", b"image")
+        assert url == "s3://my-bucket/prefix/images/img.png"
+
+    @pytest.mark.asyncio
+    async def test_pipe_called_before_url_returned(self):
+        """Test that pipe() is called (data uploaded) before URL is returned."""
+        fs = self._make_fs()
+        call_order = []
+        original_pipe = fs.pipe
+
+        def tracking_pipe(*args, **kwargs):
+            call_order.append("pipe")
+            return original_pipe(*args, **kwargs)
+
+        fs.pipe = tracking_pipe
+
+        url = await upload_to_fs(fs, "test.png", b"data")
+        call_order.append("url_returned")
+
+        assert call_order == ["pipe", "url_returned"]
+        assert url

components/src/dynamo/common/utils/video_utils.py

@@ -51,7 +51,7 @@ def compute_num_frames(
 ) -> int:
     """Compute the number of video frames.
 
-    Priority: num_frames > seconds × fps > default_num_frames.
+    Priority: num_frames > seconds x fps > default_num_frames.
     """
     if num_frames is not None:
         return num_frames

components/src/dynamo/sglang/backend_args.py

@@ -73,20 +73,6 @@ class DynamoSGLangArgGroup(ArgGroup):
             default=False,
             help="Run as image diffusion worker for image generation.",
         )
-        add_argument(
-            g,
-            flag_name="--image-diffusion-fs-url",
-            env_var="DYN_SGL_IMAGE_DIFFUSION_FS_URL",
-            default=None,
-            help="Filesystem URL for storing generated images using fsspec (e.g., s3://bucket/path, gs://bucket/path, file:///local/path). Supports any fsspec-compatible filesystem.",
-        )
-        add_argument(
-            g,
-            flag_name="--image-diffusion-base-url",
-            env_var="DYN_SGL_IMAGE_DIFFUSION_BASE_URL",
-            default="http://localhost:8008/",
-            help="Base URL for rewriting image URLs in responses (e.g., http://localhost:8008/). When set, generated image URLs will use this base instead of filesystem URLs.",
-        )
         add_argument(
             g,
             flag_name="--disagg-config",
@@ -108,13 +94,6 @@ class DynamoSGLangArgGroup(ArgGroup):
             default=False,
             help="Run as video generation worker for video generation (T2V/I2V).",
         )
-        add_argument(
-            g,
-            flag_name="--video-generation-fs-url",
-            env_var="DYN_SGL_VIDEO_GENERATION_FS_URL",
-            default=None,
-            help="Filesystem URL for storing generated videos using fsspec (e.g., s3://bucket/path, gs://bucket/path, file:///local/path). Supports any fsspec-compatible filesystem.",
-        )
 
 
 class DynamoSGLangConfig(ConfigBase):
@@ -126,14 +105,11 @@ class DynamoSGLangConfig(ConfigBase):
     multimodal_worker: bool
     embedding_worker: bool
     image_diffusion_worker: bool
-    image_diffusion_fs_url: Optional[str] = None
-    image_diffusion_base_url: Optional[str] = None
 
     disagg_config: Optional[str] = None
     disagg_config_key: Optional[str] = None
 
     video_generation_worker: bool
-    video_generation_fs_url: Optional[str] = None
 
     def validate(self) -> None:
         if (self.disagg_config is not None) ^ (self.disagg_config_key is not None):

components/src/dynamo/sglang/main.py

@@ -583,12 +583,7 @@ async def init_image_diffusion(runtime: DistributedRuntime, config: Config):
         dist_timeout=dist_timeout,
     )
 
-    # Initialize fsspec filesystems for image storage
-    fs_url = dynamo_args.image_diffusion_fs_url
-
-    # Initialize primary filesystem
-    if not fs_url:
-        raise ValueError("--image-diffusion-fs-url is required for diffusion workers")
+    fs_url = dynamo_args.media_output_fs_url
 
     generate_endpoint = runtime.endpoint(
         f"{dynamo_args.namespace}.{dynamo_args.component}.{dynamo_args.endpoint}"
@@ -667,14 +662,7 @@ async def init_video_generation(runtime: DistributedRuntime, config: Config):
         dist_timeout=dist_timeout,
     )
 
-    # Initialize fsspec filesystems for video storage
-    fs_url = dynamo_args.video_generation_fs_url
-
-    # Initialize primary filesystem
-    if not fs_url:
-        raise ValueError(
-            "--video-generation-fs-url is required for video generation workers"
-        )
+    fs_url = dynamo_args.media_output_fs_url
 
     generate_endpoint = runtime.endpoint(
         f"{dynamo_args.namespace}.{dynamo_args.component}.{dynamo_args.endpoint}"

components/src/dynamo/sglang/request_handlers/image_diffusion/image_diffusion_handler.py

@@ -14,6 +14,7 @@ import torch
 from PIL import Image
 
 from dynamo._core import Component, Context
+from dynamo.common.storage import upload_to_fs
 from dynamo.sglang.args import Config
 from dynamo.sglang.protocol import CreateImageRequest, ImageData, ImagesResponse, NvExt
 from dynamo.sglang.publisher import DynamoSglangPublisher
@@ -52,8 +53,8 @@ class ImageDiffusionWorkerHandler(BaseGenerativeHandler):
 
         self.generator = generator  # DiffGenerator, not Engine
         self.fs = fs
-        self.fs_url = config.dynamo_args.image_diffusion_fs_url
-        self.base_url = config.dynamo_args.image_diffusion_base_url
+        self.fs_url = config.dynamo_args.media_output_fs_url
+        self.base_url = config.dynamo_args.media_output_http_url
 
         logger.info(
             f"Image diffusion worker handler initialized with fs_url={self.fs_url}, url_base={self.base_url}"
@@ -225,10 +226,7 @@ class ImageDiffusionWorkerHandler(BaseGenerativeHandler):
         # Per-user storage path
         storage_path = f"users/{user_id}/generations/{request_id}/{image_filename}"
 
-        # send image to filesystem
-        await asyncio.to_thread(self.fs.pipe, storage_path, image_bytes)
-
-        return f"{self.base_url}/{storage_path}"
+        return await upload_to_fs(self.fs, storage_path, image_bytes, self.base_url)
 
     def _encode_base64(self, image_bytes: bytes) -> str:
         """Encode image as base64 string"""

components/src/dynamo/sglang/request_handlers/video_generation/video_generation_handler.py

@@ -12,6 +12,7 @@ from typing import Any, AsyncGenerator, Optional
 import torch
 
 from dynamo._core import Component, Context
+from dynamo.common.storage import upload_to_fs
 from dynamo.sglang.args import Config
 from dynamo.sglang.protocol import (
     CreateVideoRequest,
@@ -56,7 +57,8 @@ class VideoGenerationWorkerHandler(BaseGenerativeHandler):
         self.generator = generator  # DiffGenerator, not Engine
         self._generate_lock = asyncio.Lock()  # Serialize generator access
         self.fs = fs
-        self.fs_url = config.dynamo_args.video_generation_fs_url
+        self.fs_url = config.dynamo_args.media_output_fs_url
+        self.base_url = config.dynamo_args.media_output_http_url
 
         logger.info(
             f"Video generation worker handler initialized with fs_url={self.fs_url}"
@@ -303,11 +305,7 @@ class VideoGenerationWorkerHandler(BaseGenerativeHandler):
             URL for the uploaded video.
         """
         storage_path = f"{request_id}.mp4"
-
-        # DirFileSystem handles root path and protocol internally
-        await asyncio.to_thread(self.fs.pipe, storage_path, video_bytes)
-
-        return f"{self.fs_url}/{storage_path}"
+        return await upload_to_fs(self.fs, storage_path, video_bytes, self.base_url)
 
     def _encode_base64(self, video_bytes: bytes) -> str:
         """Encode video as base64 string"""

components/src/dynamo/sglang/tests/test_sglang_image_diffusion_handler.py

@@ -42,8 +42,8 @@ def mock_config():
     """Mock Config object."""
     config = MagicMock()
     config.dynamo_args = MagicMock()
-    config.dynamo_args.image_diffusion_fs_url = "file:///tmp/images"
-    config.dynamo_args.image_diffusion_base_url = "file:///tmp/images"
+    config.dynamo_args.media_output_fs_url = "file:///tmp/images"
+    config.dynamo_args.media_output_http_url = "file:///tmp/images"
     return config
 
 
@@ -96,8 +96,8 @@ class TestImageDiffusionWorkerHandler:
         """Test handler initialization with URL base."""
         config = MagicMock()
         config.dynamo_args = MagicMock()
-        config.dynamo_args.image_diffusion_fs_url = "s3://my-bucket/images"
-        config.dynamo_args.image_diffusion_base_url = "http://localhost:8008/images"
+        config.dynamo_args.media_output_fs_url = "s3://my-bucket/images"
+        config.dynamo_args.media_output_http_url = "http://localhost:8008/images"
 
         handler = ImageDiffusionWorkerHandler(
             component=mock_component,

components/src/dynamo/trtllm/backend_args.py

@@ -195,13 +195,6 @@ class DynamoTrtllmArgGroup(ArgGroup):
             "Diffusion Options [Experimental]",
             "Options for video_diffusion modality",
         )
-        add_argument(
-            diffusion_group,
-            flag_name="--output-dir",
-            env_var="DYN_TRTLLM_OUTPUT_DIR",
-            default="/tmp/dynamo_videos",
-            help="Directory to store generated videos/images.",
-        )
         add_argument(
             diffusion_group,
             flag_name="--default-height",
@@ -377,7 +370,6 @@ class DynamoTrtllmConfig(ConfigBase):
     allowed_local_media_path: str
     max_file_size_mb: int
 
-    output_dir: str
     default_height: int
     default_width: int
     default_num_frames: int

components/src/dynamo/trtllm/configs/diffusion_config.py

@@ -41,8 +41,9 @@ class DiffusionConfig:
     # float16 can be used on older GPUs (V100, etc.)
     torch_dtype: str = "bfloat16"
 
-    # Output config
-    output_dir: str = "/tmp/dynamo_videos"
+    # Media storage
+    media_output_fs_url: str = "file:///tmp/dynamo_media"
+    media_output_http_url: Optional[str] = None
 
     # Default generation parameters
     default_height: int = 480
@@ -85,7 +86,7 @@ class DiffusionConfig:
             f"endpoint={self.endpoint}, "
             f"model_path={self.model_path}, "
             f"served_model_name={self.served_model_name}, "
-            f"output_dir={self.output_dir}, "
+            f"media_output_fs_url={self.media_output_fs_url}, "
             f"default_height={self.default_height}, "
             f"default_width={self.default_width}, "
             f"default_num_frames={self.default_num_frames}, "

components/src/dynamo/trtllm/request_handlers/video_diffusion/video_handler.py

@@ -20,7 +20,8 @@ from dynamo.common.protocols.video_protocol import (
     VideoData,
     VideoNvExt,
 )
-from dynamo.common.utils.video_utils import encode_to_mp4, encode_to_mp4_bytes
+from dynamo.common.storage import get_fs, upload_to_fs
+from dynamo.common.utils.video_utils import encode_to_mp4_bytes
 from dynamo.trtllm.configs.diffusion_config import DiffusionConfig
 from dynamo.trtllm.engines.diffusion_engine import DiffusionEngine
 from dynamo.trtllm.request_handlers.base_generative_handler import BaseGenerativeHandler
@@ -55,6 +56,12 @@ class VideoGenerationHandler(BaseGenerativeHandler):
         self.component = component
         self.engine = engine
         self.config = config
+        if not config.media_output_fs_url:
+            raise ValueError(
+                "media_output_fs_url must be set; use --media-output-fs-url or DYN_MEDIA_OUTPUT_FS_URL."
+            )
+        self.media_output_fs = get_fs(config.media_output_fs_url)
+        self.media_output_http_url = config.media_output_http_url
         # Serialize pipeline access — visual_gen is not thread-safe (global
         # singleton configs, mutable instance state, unprotected CUDA graph cache).
         # asyncio.Lock suspends waiting coroutines cooperatively so the event
@@ -218,21 +225,21 @@ class VideoGenerationHandler(BaseGenerativeHandler):
             response_format = req.response_format or "url"
             fps = nvext.fps or self.config.default_fps
 
+            # Encode frames to MP4 bytes in memory
+            video_bytes = await asyncio.to_thread(encode_to_mp4_bytes, frames, fps=fps)
+
             if response_format == "url":
-                # Encode to MP4 and save to file
-                output_path = await asyncio.to_thread(
-                    encode_to_mp4,
-                    frames,
-                    self.config.output_dir,
-                    request_id,
-                    fps=fps,
+                # Upload via filesystem
+                storage_path = f"videos/{request_id}.mp4"
+                video_url = await upload_to_fs(
+                    self.media_output_fs,
+                    storage_path,
+                    video_bytes,
+                    self.media_output_http_url,
                 )
-                video_data = VideoData(url=output_path)
+                video_data = VideoData(url=video_url)
             else:
                 # Encode to base64
-                video_bytes = await asyncio.to_thread(
-                    encode_to_mp4_bytes, frames, fps=fps
-                )
                 b64_video = base64.b64encode(video_bytes).decode("utf-8")
                 video_data = VideoData(b64_json=b64_video)
 

components/src/dynamo/trtllm/tests/test_trtllm_video_diffusion.py

@@ -98,8 +98,9 @@ class TestDiffusionConfig:
         assert config.default_num_inference_steps == 50
         assert config.default_guidance_scale == 5.0
 
-        # Model defaults
-        assert config.output_dir == "/tmp/dynamo_videos"
+        # Media storage defaults
+        assert config.media_output_fs_url == "file:///tmp/dynamo_media"
+        assert config.media_output_http_url is None
 
         # Optimization defaults
         assert config.enable_teacache is False
@@ -126,6 +127,16 @@ class TestDiffusionConfig:
         assert config.enable_teacache is True
         assert config.dit_tp_size == 2
 
+    def test_custom_media_storage(self):
+        """Test that media storage fields can be overridden."""
+        config = DiffusionConfig(
+            media_output_fs_url="s3://my-bucket/videos",
+            media_output_http_url="https://cdn.example.com/videos",
+        )
+
+        assert config.media_output_fs_url == "s3://my-bucket/videos"
+        assert config.media_output_http_url == "https://cdn.example.com/videos"
+
     def test_str_representation(self):
         """Test that __str__ includes key fields."""
         config = DiffusionConfig(
@@ -576,11 +587,15 @@ class TestVideoHandlerConcurrency:
         mock_engine.generate = tracker.generate
 
         config = DiffusionConfig(
-            output_dir="/tmp/test_videos",
+            media_output_fs_url="file:///tmp/test_media",
             default_fps=24,
             default_seconds=4,
         )
 
+        with patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.get_fs",
+            return_value=MagicMock(),
+        ):
             handler = VideoGenerationHandler(
                 component=MagicMock(),
                 engine=mock_engine,
@@ -616,8 +631,11 @@ class TestVideoHandlerConcurrency:
             requests = [self._make_request() for _ in range(3)]
 
             with patch(
-                "dynamo.trtllm.request_handlers.video_diffusion.video_handler.encode_to_mp4",
-                return_value="/tmp/test.mp4",
+                "dynamo.trtllm.request_handlers.video_diffusion.video_handler.encode_to_mp4_bytes",
+                return_value=b"fake_mp4_bytes",
+            ), patch(
+                "dynamo.trtllm.request_handlers.video_diffusion.video_handler.upload_to_fs",
+                return_value="http://fake/video.mp4",
             ):
                 await asyncio.gather(
                     *(self._drain_generator(handler, req) for req in requests)
@@ -631,3 +649,156 @@ class TestVideoHandlerConcurrency:
             f"Expected max_concurrent=1 (serialized), got {tracker.max_concurrent}. "
             "Pipeline was accessed concurrently — this would corrupt visual_gen state."
         )
+
+
+# =============================================================================
+# Part 6: VideoGenerationHandler Response Format Tests
+# =============================================================================
+
+
+class TestVideoHandlerResponseFormats:
+    """Tests for VideoGenerationHandler generate() response format branching."""
+
+    def _make_handler(self):
+        """Create a handler with mocked engine and fs."""
+        import numpy as np
+
+        from dynamo.trtllm.request_handlers.video_diffusion.video_handler import (
+            VideoGenerationHandler,
+        )
+
+        mock_engine = MagicMock()
+        mock_engine.generate = MagicMock(
+            return_value=np.zeros((4, 64, 64, 3), dtype=np.uint8)
+        )
+
+        config = DiffusionConfig(
+            media_output_fs_url="file:///tmp/test_media",
+            media_output_http_url="https://cdn.example.com/media",
+            default_fps=24,
+            default_seconds=4,
+        )
+
+        with patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.get_fs",
+            return_value=MagicMock(),
+        ):
+            handler = VideoGenerationHandler(
+                component=MagicMock(),
+                engine=mock_engine,
+                config=config,
+            )
+
+        return handler
+
+    @pytest.mark.asyncio
+    async def test_url_response_format(self):
+        """Test generate() with url response format calls upload_to_fs."""
+        handler = self._make_handler()
+
+        request = {
+            "prompt": "a test video",
+            "model": "test-model",
+            "response_format": "url",
+        }
+
+        with patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.encode_to_mp4_bytes",
+            return_value=b"fake_mp4",
+        ), patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.upload_to_fs",
+            return_value="https://cdn.example.com/media/videos/test.mp4",
+        ) as mock_upload:
+            results = []
+            async for result in handler.generate(request, MagicMock()):
+                results.append(result)
+
+        assert len(results) == 1
+        response = results[0]
+        assert response["status"] == "completed"
+        assert len(response["data"]) == 1
+        assert (
+            response["data"][0]["url"]
+            == "https://cdn.example.com/media/videos/test.mp4"
+        )
+        mock_upload.assert_called_once()
+
+    @pytest.mark.asyncio
+    async def test_b64_response_format(self):
+        """Test generate() with b64_json response format returns base64 encoded video."""
+        handler = self._make_handler()
+
+        request = {
+            "prompt": "a test video",
+            "model": "test-model",
+            "response_format": "b64_json",
+        }
+
+        with patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.encode_to_mp4_bytes",
+            return_value=b"fake_mp4_bytes",
+        ):
+            results = []
+            async for result in handler.generate(request, MagicMock()):
+                results.append(result)
+
+        assert len(results) == 1
+        response = results[0]
+        assert response["status"] == "completed"
+        assert len(response["data"]) == 1
+        assert response["data"][0]["b64_json"] is not None
+        assert response["data"][0].get("url") is None
+
+        # Verify valid base64
+        import base64
+
+        decoded = base64.b64decode(response["data"][0]["b64_json"])
+        assert decoded == b"fake_mp4_bytes"
+
+    @pytest.mark.asyncio
+    async def test_default_response_format_is_url(self):
+        """Test that generate() defaults to url response format."""
+        handler = self._make_handler()
+
+        request = {
+            "prompt": "a test video",
+            "model": "test-model",
+            # No response_format specified
+        }
+
+        with patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.encode_to_mp4_bytes",
+            return_value=b"fake_mp4",
+        ), patch(
+            "dynamo.trtllm.request_handlers.video_diffusion.video_handler.upload_to_fs",
+            return_value="https://cdn.example.com/media/videos/test.mp4",
+        ) as mock_upload:
+            results = []
+            async for result in handler.generate(request, MagicMock()):
+                results.append(result)
+
+        assert len(results) == 1
+        # Default should be "url" format, so upload_to_fs should be called
+        mock_upload.assert_called_once()
+        assert results[0]["data"][0]["url"] is not None
+
+    @pytest.mark.asyncio
+    async def test_error_response_on_failure(self):
+        """Test that generate() returns error response on engine failure."""
+        handler = self._make_handler()
+        handler.engine.generate = MagicMock(side_effect=RuntimeError("GPU OOM"))
+
+        request = {
+            "prompt": "a test video",
+            "model": "test-model",
+        }
+
+        results = []
+        async for result in handler.generate(request, MagicMock()):
+            results.append(result)
+
+        assert len(results) == 1
+        response = results[0]
+        assert response["status"] == "failed"
+        assert response["error"] == "GPU OOM"
+        assert response["data"] == []

components/src/dynamo/trtllm/workers/video_diffusion_worker.py

@@ -60,7 +60,8 @@ async def init_video_diffusion_worker(
         event_plane=config.event_plane,
         model_path=config.model,
         served_model_name=config.served_model_name,
-        output_dir=config.output_dir,
+        media_output_fs_url=config.media_output_fs_url,
+        media_output_http_url=config.media_output_http_url,
         default_height=config.default_height,
         default_width=config.default_width,
         default_num_frames=config.default_num_frames,

components/src/dynamo/vllm/backend_args.py

@@ -138,15 +138,7 @@ class DynamoVllmArgGroup(ArgGroup):
             help="Path to vLLM-Omni stage configuration YAML file for --omni mode (optional).",
         )
 
-        # Video diffusion output
-        # TODO: Propose an alternate design to switch to AsyncOmniEngine args while using vLLM-Omni
-        add_argument(
-            g,
-            flag_name="--video-output-dir",
-            env_var="DYN_VLLM_VIDEO_OUTPUT_DIR",
-            default="/tmp/dynamo_videos",  # noqa: S108
-            help="Directory to save generated video MP4 files.",
-        )
+        # Video encoding
         add_argument(
             g,
             flag_name="--default-video-fps",
@@ -240,6 +232,13 @@ class DynamoVllmArgGroup(ArgGroup):
             default=False,
             help="Enable CPU offloading for diffusion models to reduce GPU memory usage.",
         )
+        add_negatable_bool_argument(
+            g,
+            flag_name="--enforce-eager",
+            env_var="DYN_VLLM_ENFORCE_EAGER",
+            default=False,
+            help="Disable torch.compile and force eager execution for diffusion models.",
+        )
         # Diffusion parallel configuration
         add_argument(
             g,
@@ -300,8 +299,7 @@ class DynamoVllmConfig(ConfigBase):
     omni: bool
     stage_configs_path: Optional[str] = None
 
-    # Video diffusion output
-    video_output_dir: str = "/tmp/dynamo_videos"  # noqa: S108
+    # Video encoding
     default_video_fps: int = 16
 
     # Diffusion engine-level parameters (passed to AsyncOmni constructor)

components/src/dynamo/vllm/main.py

@@ -17,6 +17,7 @@ from vllm.v1.metrics.prometheus import setup_multiprocess_prometheus
 
 from dynamo import prometheus_names
 from dynamo.common.config_dump import dump_config
+from dynamo.common.storage import get_fs
 from dynamo.common.utils.endpoint_types import parse_endpoint_types
 from dynamo.common.utils.output_modalities import get_output_modalities
 from dynamo.common.utils.prometheus import (
@@ -907,7 +908,6 @@ async def init_omni(
     Supports text-to-text, text-to-image, and text-to-video generation
     through a single unified OmniHandler.
     """
-    # Lazy import to avoid loading vllm-omni unless explicitly needed
     from dynamo.vllm.omni import OmniHandler
 
     generate_endpoint = runtime.endpoint(
@@ -915,6 +915,11 @@ async def init_omni(
     )
     component = generate_endpoint.component()
 
+    # Initialize media filesystem for storing generated images/videos
+    media_fs = (
+        get_fs(config.media_output_fs_url) if config.media_output_fs_url else None
+    )
+
     # Initialize unified OmniHandler
     handler = OmniHandler(
         runtime=runtime,
@@ -922,6 +927,8 @@ async def init_omni(
         config=config,
         default_sampling_params={},
         shutdown_event=shutdown_event,
+        media_output_fs=media_fs,
+        media_output_http_url=config.media_output_http_url,
     )
 
     logger.info(f"Omni worker initialized for model: {config.model}")

components/src/dynamo/vllm/omni/base_handler.py

@@ -131,7 +131,6 @@ class BaseOmniHandler(BaseWorkerHandler):
     ) -> AsyncGenerator[Dict, None]:
         """Generate outputs using AsyncOmni orchestrator with OpenAI-compatible format.
 
-        Routes to OpenAI mode (detokenized text) or token mode based on config.
         Subclasses should override ``_generate_openai_mode`` for custom output handling.
         """
         request_id = context.id()

components/src/dynamo/vllm/omni/omni_handler.py

@@ -3,14 +3,15 @@
 import asyncio
 import base64
 import logging
-import os
+import tempfile
 import time
 import uuid
 from dataclasses import dataclass
 from io import BytesIO
-from typing import Any, AsyncGenerator, Dict, Union
+from typing import Any, AsyncGenerator, Dict, Optional, Union
 
 from diffusers.utils import export_to_video
+from fsspec.implementations.dirfs import DirFileSystem
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams, OmniTextPrompt
 
 from dynamo.common.protocols.image_protocol import (
@@ -23,6 +24,7 @@ from dynamo.common.protocols.video_protocol import (
     NvVideosResponse,
     VideoData,
 )
+from dynamo.common.storage import upload_to_fs
 from dynamo.common.utils.output_modalities import RequestType, parse_request_type
 from dynamo.common.utils.video_utils import (
     compute_num_frames,
@@ -33,9 +35,7 @@ from dynamo.vllm.omni.base_handler import BaseOmniHandler
 
 logger = logging.getLogger(__name__)
 
-# TODO: Migrate to fs_url based approach in another PR
 DEFAULT_VIDEO_FPS = 16
-DEFAULT_VIDEO_OUTPUT_DIR = "/tmp/dynamo_videos"  # noqa: S108
 
 
 @dataclass
@@ -59,40 +59,6 @@ class EngineInputs:
     response_format: str | None = None
 
 
-def prepare_image_output(images: list, response_format: str | None = None):
-    """Prepare image output for response.
-
-    Args:
-        images: List of PIL Image objects.
-        response_format: Response format.
-
-    Returns:
-        List of image URLs or base64 strings.
-    """
-    ## This is a temporary function to prepare image output for response.
-    ## Right now, there are different utilities across components that uploads image/video outputs to urls or b64_json.
-    ## (ayushag) TODO: follow up, move all the utilities to common
-    outlist = []
-
-    for img in images:
-        if response_format == "url":
-            output_dir = "/tmp/dynamo_images"  # noqa: S108
-            os.makedirs(output_dir, exist_ok=True)
-            img_path = os.path.join(output_dir, f"{uuid.uuid4()}.png")
-            img.save(img_path)
-            outlist.append(img_path)
-        elif response_format == "b64_json" or response_format is None:
-            # convert image to base64
-            buffer = BytesIO()
-            img.save(buffer, format="PNG")
-            img_base64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
-            data_url = f"data:image/png;base64,{img_base64}"
-            outlist.append(data_url)
-        else:
-            raise ValueError(f"Invalid response format: {response_format}")
-    return outlist
-
-
 class OmniHandler(BaseOmniHandler):
     """Unified handler for multi-stage pipelines using vLLM-Omni.
 
@@ -106,6 +72,8 @@ class OmniHandler(BaseOmniHandler):
         config,
         default_sampling_params: Dict[str, Any],
         shutdown_event: asyncio.Event | None = None,
+        media_output_fs: Optional[DirFileSystem] = None,
+        media_output_http_url: Optional[str] = None,
     ):
         """Initialize the unified Omni handler.
 
@@ -115,6 +83,8 @@ class OmniHandler(BaseOmniHandler):
             config: Parsed Config object from args.py.
             default_sampling_params: Default sampling parameters dict.
             shutdown_event: Optional asyncio event for graceful shutdown.
+            media_output_fs: Filesystem for storing generated images/videos.
+            media_output_http_url: Base URL for rewriting media paths in responses.
         """
         super().__init__(
             runtime=runtime,
@@ -123,6 +93,8 @@ class OmniHandler(BaseOmniHandler):
             default_sampling_params=default_sampling_params,
             shutdown_event=shutdown_event,
         )
+        self.media_output_fs = media_output_fs
+        self.media_output_http_url = media_output_http_url
 
     async def generate(
         self, request: Dict[str, Any], context
@@ -194,7 +166,7 @@ class OmniHandler(BaseOmniHandler):
                                 fps=inputs.fps,
                             )
                         else:
-                            chunk = self._format_image_chunk(
+                            chunk = await self._format_image_chunk(
                                 stage_output.images,
                                 request_id,
                                 response_format=inputs.response_format,
@@ -339,14 +311,51 @@ class OmniHandler(BaseOmniHandler):
             fps=fps,
         )
 
-    def _format_image_chunk(
+    async def _prepare_image_output(
+        self, images: list, request_id: str, response_format: str | None = None
+    ) -> list:
+        """Prepare image output for response.
+
+        Args:
+            images: List of PIL Image objects.
+            request_id: Unique request identifier.
+            response_format: Response format ("url" or "b64_json").
+
+        Returns:
+            List of image URLs or base64 data-URL strings.
+        """
+        outlist = []
+
+        for img in images:
+            buffer = BytesIO()
+            img.save(buffer, format="PNG")
+            image_bytes = buffer.getvalue()
+
+            if response_format == "url":
+                storage_path = f"images/{request_id}/{uuid.uuid4()}.png"
+                url = await upload_to_fs(
+                    self.media_output_fs,
+                    storage_path,
+                    image_bytes,
+                    self.media_output_http_url,
+                )
+                outlist.append(url)
+            elif response_format == "b64_json" or response_format is None:
+                img_base64 = base64.b64encode(image_bytes).decode("utf-8")
+                data_url = f"data:image/png;base64,{img_base64}"
+                outlist.append(data_url)
+            else:
+                raise ValueError(f"Invalid response format: {response_format}")
+        return outlist
+
+    async def _format_image_chunk(
         self,
         images: list,
         request_id: str,
         response_format: str | None = None,
         request_type: RequestType = RequestType.IMAGE_GENERATION,
     ) -> Dict[str, Any] | None:
-        """Format image output as OpenAI chat completion chunk with base64 data URLs.
+        """Format image output for the appropriate endpoint response.
 
         Args:
             images: List of PIL Image objects generated by AsyncOmni engine.
@@ -355,17 +364,16 @@ class OmniHandler(BaseOmniHandler):
             request_type: Request type (chat completion, image generation).
 
         Returns:
-            Dict[str, Any] | None: Formatted chunk, or None if no images generated.
+            Formatted response dict, or None if no images generated.
         """
-
         if not images:
             return self._error_chunk(request_id, "No images generated")
 
-        data_urls = prepare_image_output(images, response_format)
+        data_urls = await self._prepare_image_output(
+            images, request_id, response_format
+        )
 
         if request_type == RequestType.CHAT_COMPLETION:
-            # This branch is used when user send request via /v1/chat/completions endpoint.
-            # We need to return chat completion chunk with image_url content part.
             chunk = {
                 "id": request_id,
                 "created": int(time.time()),
@@ -387,14 +395,11 @@ class OmniHandler(BaseOmniHandler):
             }
             return chunk
         elif request_type == RequestType.IMAGE_GENERATION:
-            # This branch is used when user send request via /v1/images/generations endpoint.
-            # This will return NvImagesResponse with list of ImageData objects.
             image_data_list = []
             for data_url in data_urls:
                 if response_format == "url":
                     image_data_list.append(ImageData(url=data_url))
                 elif response_format == "b64_json" or response_format is None:
-                    # strip explicit prefix if present
                     if data_url.startswith("data:image"):
                         _, b64_part = data_url.split(",", 1)
                         image_data_list.append(ImageData(b64_json=b64_part))
@@ -435,19 +440,21 @@ class OmniHandler(BaseOmniHandler):
                 f"(fps={fps})"
             )
 
-            os.makedirs(DEFAULT_VIDEO_OUTPUT_DIR, exist_ok=True)
-            video_path = os.path.join(DEFAULT_VIDEO_OUTPUT_DIR, f"{request_id}.mp4")
-
-            loop = asyncio.get_running_loop()
-            await loop.run_in_executor(
-                None,
-                export_to_video,
-                frame_list,
-                video_path,
-                fps,
+            # Encode frames to MP4 via temp file, then read bytes for upload
+            with tempfile.NamedTemporaryFile(suffix=".mp4", delete=True) as tmp:
+                await asyncio.to_thread(export_to_video, frame_list, tmp.name, fps)
+                video_bytes = tmp.read()
+
+            # Upload via filesystem
+            storage_path = f"videos/{request_id}.mp4"
+            video_url = await upload_to_fs(
+                self.media_output_fs,
+                storage_path,
+                video_bytes,
+                self.media_output_http_url,
             )
 
-            logger.info(f"Video saved to {video_path} for request {request_id}")
+            logger.info(f"Video uploaded to {video_url} for request {request_id}")
 
             inference_time = time.time() - start_time
 
@@ -458,7 +465,7 @@ class OmniHandler(BaseOmniHandler):
                 status="completed",
                 progress=100,
                 created=int(time.time()),
-                data=[VideoData(url=video_path)],
+                data=[VideoData(url=video_url)],
                 inference_time_s=inference_time,
             )
             return response.model_dump()

components/src/dynamo/vllm/tests/test_vllm_omni_handler.py

@@ -9,7 +9,6 @@ from dynamo.common.protocols.image_protocol import NvCreateImageRequest
 from dynamo.common.protocols.video_protocol import NvCreateVideoRequest
 from dynamo.common.utils.output_modalities import RequestType
 
-# TODO: Install vLLM omni dependencies in CI container so this skip is no longer needed.
 try:
     from dynamo.vllm.omni.omni_handler import (
         EngineInputs,