fix(fal_ai): handle seconds=auto and oversized sizes for minimax h3 videos (#42504)

* test(fal_ai): e2e for minimax h3 auto duration and oversized size

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

* fix(fal_ai): keep auto duration literal and make h3 tier lookup total

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

* fix(fal_ai): omit duration for h3 when seconds is auto

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

* test(fal_ai): move the minimax h3 auto duration and oversized size repro to tests/integration

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

---------

Co-authored-by: kerry <kerry@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/llms/fal_ai/videos/transformation.py

@@ -104,7 +104,10 @@ def _profile_for_model(model: str) -> _ModelProfile:
 
 
 def _resolution_for_short_side(short_side: int, profile: _ModelProfile) -> str:
-    return next(resolution for threshold, resolution in profile.resolution_tiers if short_side <= threshold)
+    return next(
+        (resolution for threshold, resolution in profile.resolution_tiers if short_side <= threshold),
+        profile.resolution_tiers[-1][1],
+    )
 
 
 def _model_path_from_request_url(raw_response: httpx.Response) -> str | None:
@@ -351,6 +354,8 @@ class FalAIVideoConfig(BaseVideoConfig):
         duration: Final[str | None] = _duration_value(seconds)
         if duration is None:
             raise ValueError("fal.ai seconds must be a numeric value")
+        if duration == "auto":
+            return MappingProxyType({}) if profile.integer_duration else MappingProxyType({"duration": duration})
         return MappingProxyType({"duration": int(duration) if profile.integer_duration else duration})
 
     def validate_environment(

tests/integration/contracts.json

@@ -169,6 +169,12 @@
     "tests/integration/providers/test_fal_ai_video_wire.py::test_fal_video_failed_result_reports_failed_status_and_fal_error": [
       "other.provider_wire.fal_ai.video_failed_result_surfaces_fal_error"
     ],
+    "tests/integration/providers/test_fal_ai_video_wire.py::test_fal_h3_auto_duration_omits_duration_and_queues": [
+      "other.provider_wire.fal_ai.h3_auto_duration_omits_duration_and_queues"
+    ],
+    "tests/integration/providers/test_fal_ai_video_wire.py::test_fal_h3_oversized_size_uses_top_resolution_tier_and_queues": [
+      "other.provider_wire.fal_ai.h3_oversized_size_uses_top_resolution_tier"
+    ],
     "tests/integration/providers/test_fal_ai_image_wire.py::test_fal_gpt_image_25_generation_sends_quality_and_size_and_charges_keyed_row": [
       "other.provider_wire.fal_ai.gpt_image_generation_quality_size_wire_and_keyed_pricing"
     ],

tests/integration/providers/test_fal_ai_video_wire.py

@@ -9,6 +9,11 @@ from integration._support.wire import Reply, Request, wire_server
 _MODEL: Final = "bytedance/seedance-2.5/text-to-video"
 _H3_MODEL: Final = "minimax/h3/text-to-video"
 _MP4: Final = b"\x00\x00\x00\x18ftypmp42" + uuid.uuid4().bytes * 4
+_OVERSIZED_SIDE: Final = "9" * 30
+
+
+def _h3_queue_reply(request_id: str) -> Reply:
+    return Reply(body=json.dumps({"status": "IN_QUEUE", "request_id": request_id, "queue_position": 0}).encode())
 
 
 @pytest.mark.covers("other.provider_wire.fal_ai.video_queue_create_status_and_content_download")
@@ -177,3 +182,61 @@ def test_fal_video_failed_result_reports_failed_status_and_fal_error(gateway: Ga
         content: Final = gateway.request("GET", f"/v1/videos/{video_id}/content")
         assert content.status_code == 422, content.text
         assert "Failed to download the file" in content.text
+
+
+@pytest.mark.covers("other.provider_wire.fal_ai.h3_auto_duration_omits_duration_and_queues")
+def test_fal_h3_auto_duration_omits_duration_and_queues(gateway: Gateway) -> None:
+    request_id: Final = "fal-h3-auto-" + uuid.uuid4().hex
+
+    def respond(request: Request) -> Reply:
+        assert request.method == "POST"
+        assert request.target == f"/{_H3_MODEL}"
+        assert json.loads(request.body) == {
+            "prompt": "a cat playing volleyball on a beach",
+            "resolution": "768P",
+            "aspect_ratio": "16:9",
+        }
+        return _h3_queue_reply(request_id)
+
+    with wire_server(respond) as wire, gateway.scenario() as scenario:
+        model: Final = scenario.model(model=f"fal_ai/{_H3_MODEL}", api_base=wire.url, api_key="synthetic-fal-key")
+        response: Final = gateway.request(
+            "POST",
+            "/v1/videos",
+            {"model": model, "prompt": "a cat playing volleyball on a beach", "seconds": "auto", "size": "1280x720"},
+        )
+        assert response.status_code == 200, response.text
+        assert response.json()["status"] == "queued"
+        assert [(request.method, request.target) for request in wire.drain()] == [("POST", f"/{_H3_MODEL}")]
+
+
+@pytest.mark.covers("other.provider_wire.fal_ai.h3_oversized_size_uses_top_resolution_tier")
+def test_fal_h3_oversized_size_uses_top_resolution_tier_and_queues(gateway: Gateway) -> None:
+    request_id: Final = "fal-h3-oversized-" + uuid.uuid4().hex
+
+    def respond(request: Request) -> Reply:
+        assert request.method == "POST"
+        assert request.target == f"/{_H3_MODEL}"
+        assert json.loads(request.body) == {
+            "prompt": "a cat playing volleyball on a beach",
+            "duration": 5,
+            "resolution": "4K",
+            "aspect_ratio": "1:1",
+        }
+        return _h3_queue_reply(request_id)
+
+    with wire_server(respond) as wire, gateway.scenario() as scenario:
+        model: Final = scenario.model(model=f"fal_ai/{_H3_MODEL}", api_base=wire.url, api_key="synthetic-fal-key")
+        response: Final = gateway.request(
+            "POST",
+            "/v1/videos",
+            {
+                "model": model,
+                "prompt": "a cat playing volleyball on a beach",
+                "seconds": "5",
+                "size": f"{_OVERSIZED_SIDE}x{_OVERSIZED_SIDE}",
+            },
+        )
+        assert response.status_code == 200, response.text
+        assert response.json()["status"] == "queued"
+        assert [(request.method, request.target) for request in wire.drain()] == [("POST", f"/{_H3_MODEL}")]

tests/test_litellm/llms/fal_ai/videos/test_fal_ai_video_transformation.py

@@ -1,3 +1,4 @@
+import sys
 from typing import Final
 from unittest.mock import AsyncMock, Mock
 
@@ -86,6 +87,17 @@ class TestFalAIVideoTransformation:
         assert mapped["reference_image_urls"] == [url]
         assert "image_url" not in mapped
 
+    def test_map_openai_params_h3_omits_auto_duration(self):
+        assert self.config.map_openai_params({"seconds": "auto"}, H3_TEXT_MODEL, False) == {}
+        assert self.config.map_openai_params({"seconds": "auto"}, MODEL, False) == {"duration": "auto"}
+
+    def test_map_openai_params_h3_size_beyond_tiers_uses_top_resolution(self):
+        side = str(sys.maxsize + 1)
+        assert self.config.map_openai_params({"size": f"{side}x{side}"}, H3_TEXT_MODEL, False) == {
+            "resolution": "4K",
+            "aspect_ratio": "1:1",
+        }
+
     def test_transform_video_create_request(self):
         body, files, url = self.config.transform_video_create_request(
             model=MODEL,