feat: enhance model caching and output modalities handling

- Updated `refresh_models_cache` to include output modalities in the models cache.
- Added `get_model_output_modalities` function to retrieve output modalities for a specific model.
- Modified tests to cover new functionality for output modalities.
- Updated OpenRouter video generation functions to support audio generation and improved error handling.
- Enhanced dashboard to display generated images and videos.
- Refactored frontend templates to accommodate new data structures for generated content.
- Adjusted tests to validate changes in model handling and dashboard rendering.

Co-authored-by: Copilot <copilot@github.com>

backend/app/services/openrouter.py

@@ -95,8 +95,9 @@ async def generate_video(
     duration_seconds: int | None = None,
     aspect_ratio: str = "16:9",
     resolution: str | None = None,
+    generate_audio: bool | None = None,
 ) -> dict[str, Any]:
-    """Request text-to-video generation via OpenRouter."""
+    """Request text-to-video generation via OpenRouter POST /videos."""
     base_url = os.getenv("OPENROUTER_BASE_URL", OPENROUTER_BASE_URL)
     payload: dict[str, Any] = {
         "model": model,
@@ -104,9 +105,12 @@ async def generate_video(
         "aspect_ratio": aspect_ratio,
     }
     if duration_seconds is not None:
-        payload["duration_seconds"] = duration_seconds
+        # API uses 'duration' not 'duration_seconds'
+        payload["duration"] = duration_seconds
     if resolution is not None:
         payload["resolution"] = resolution
+    if generate_audio is not None:
+        payload["generate_audio"] = generate_audio
     async with httpx.AsyncClient(timeout=120) as client:
         resp = client.build_request(
             "POST", f"{base_url}/videos", headers=_headers(), json=payload
@@ -123,19 +127,31 @@ async def generate_video_from_image(
     duration_seconds: int | None = None,
     aspect_ratio: str = "16:9",
     resolution: str | None = None,
+    generate_audio: bool | None = None,
 ) -> dict[str, Any]:
-    """Request image-to-video generation via OpenRouter."""
+    """Request image-to-video generation via OpenRouter POST /videos.
+
+    Uses frame_images array with first_frame as per OpenRouter API spec.
+    """
     base_url = os.getenv("OPENROUTER_BASE_URL", OPENROUTER_BASE_URL)
     payload: dict[str, Any] = {
         "model": model,
-        "image_url": image_url,
         "prompt": prompt,
         "aspect_ratio": aspect_ratio,
+        "frame_images": [
+            {
+                "type": "image_url",
+                "image_url": {"url": image_url},
+                "frame_type": "first_frame",
+            }
+        ],
     }
     if duration_seconds is not None:
-        payload["duration_seconds"] = duration_seconds
+        payload["duration"] = duration_seconds
     if resolution is not None:
         payload["resolution"] = resolution
+    if generate_audio is not None:
+        payload["generate_audio"] = generate_audio
     async with httpx.AsyncClient(timeout=120) as client:
         resp = client.build_request(
             "POST", f"{base_url}/videos", headers=_headers(), json=payload
@@ -154,6 +170,18 @@ async def poll_video_status(polling_url: str) -> dict[str, Any]:
         return response.json()
 
 
+async def list_video_models() -> list[dict[str, Any]]:
+    """Return video generation models from the dedicated /videos/models endpoint."""
+    base_url = os.getenv("OPENROUTER_BASE_URL", OPENROUTER_BASE_URL)
+    async with httpx.AsyncClient(timeout=15) as client:
+        resp = client.build_request(
+            "GET", f"{base_url}/videos/models", headers=_headers()
+        )
+        response = await client.send(resp)
+        response.raise_for_status()
+        return response.json().get("data", [])
+
+
 async def generate_image_chat(
     model: str,
     prompt: str,