Merge pull request 'fix(kling): 补齐全模态基础视频能力' (#16) from codex/omni-default-video-capabilities into main

apps/api/internal/httpapi/kling_compat_integration_test.go

@@ -29,6 +29,23 @@ func TestKlingCompatibilitySimulationFlow(t *testing.T) {
 	}
 	defer db.Close()
 
+	var upgradedBaseModels int
+	if err := db.Pool().QueryRow(ctx, `
+SELECT count(*)
+FROM base_model_catalog
+WHERE provider_key = 'keling'
+  AND provider_model_name IN ('kling-video-o1', 'kling-v3-omni')
+  AND model_type @> '["video_generate","image_to_video","omni_video"]'::jsonb
+  AND capabilities ? 'video_generate'
+  AND capabilities ? 'image_to_video'
+  AND capabilities ? 'omni_video'
+  AND metadata->'rawModel'->'types' @> '["video_generate","image_to_video","omni_video"]'::jsonb`).Scan(&upgradedBaseModels); err != nil {
+		t.Fatalf("read upgraded Kling Omni base model capabilities: %v", err)
+	}
+	if upgradedBaseModels != 2 {
+		t.Fatalf("expected both Kling Omni base models to expose base video capabilities, got %d", upgradedBaseModels)
+	}
+
 	serverCtx, cancelServer := context.WithCancel(ctx)
 	defer cancelServer()
 	server := httptest.NewServer(NewServerWithContext(serverCtx, config.Config{
@@ -89,6 +106,55 @@ func TestKlingCompatibilitySimulationFlow(t *testing.T) {
 			"displayName":       model,
 		}, http.StatusCreated, nil)
 	}
+	var upgradedPlatformModels int
+	if err := db.Pool().QueryRow(ctx, `
+SELECT count(*)
+FROM platform_models
+WHERE platform_id = $1::uuid
+  AND model_type @> '["video_generate","image_to_video","omni_video"]'::jsonb
+  AND capabilities ? 'video_generate'
+  AND capabilities ? 'image_to_video'
+  AND capabilities ? 'omni_video'`, platform.ID).Scan(&upgradedPlatformModels); err != nil {
+		t.Fatalf("read upgraded Kling Omni platform model capabilities: %v", err)
+	}
+	if upgradedPlatformModels != 2 {
+		t.Fatalf("expected both Kling Omni platform models to expose base video capabilities, got %d", upgradedPlatformModels)
+	}
+
+	assertGenericVideoGeneration := func(name string, model string, image string, expectedModelType string) {
+		t.Helper()
+		t.Run(name, func(t *testing.T) {
+			request := map[string]any{
+				"model":                model,
+				"prompt":               "通用视频接口模拟任务",
+				"duration":             5,
+				"resolution":           "720p",
+				"runMode":              "simulation",
+				"simulation":           true,
+				"simulationDurationMs": 5,
+			}
+			if image != "" {
+				request["image"] = image
+			}
+			var response struct {
+				Task struct {
+					Status        string `json:"status"`
+					ModelType     string `json:"modelType"`
+					ResolvedModel string `json:"resolvedModel"`
+				} `json:"task"`
+			}
+			doJSON(t, server.URL, http.MethodPost, "/api/v1/videos/generations", apiKeyResponse.Secret, request, http.StatusAccepted, &response)
+			resolvedModel, resolved := klingV2ProviderModel(response.Task.ResolvedModel)
+			if response.Task.Status != "succeeded" || response.Task.ModelType != expectedModelType || !resolved || resolvedModel != model {
+				t.Fatalf("generic video request without modelType should use inferred capability: %+v", response.Task)
+			}
+		})
+	}
+
+	for _, model := range []string{klingO1Model, klingV3OmniModel} {
+		assertGenericVideoGeneration(model+"-text-to-video", model, "", "video_generate")
+		assertGenericVideoGeneration(model+"-image-to-video", model, "https://example.com/first.png", "image_to_video")
+	}
 
 	createV1 := func(model string, duration int, externalID string) string {
 		t.Helper()

apps/api/internal/store/base_models.go

@@ -488,6 +488,8 @@ func modelTypeAliases(value string) []string {
 		return []string{"image_edit"}
 	case "video", "videos.generations":
 		return []string{"video_generate"}
+	case "omni_video":
+		return []string{"video_generate", "image_to_video", "omni_video"}
 	case "song", "music", "song.generations", "music.generations", "music_generate":
 		return []string{"audio_generate"}
 	case "speech", "speech.generations", "tts":

apps/api/internal/store/candidates_test.go

@@ -9,6 +9,19 @@ func TestNormalizeModelMatchKeyRemovesWhitespace(t *testing.T) {
 	}
 }
 
+func TestNormalizeModelTypeListExpandsOmniVideoBaseCapabilities(t *testing.T) {
+	got := normalizeModelTypeList([]string{"omni_video"})
+	want := StringList{"video_generate", "image_to_video", "omni_video"}
+	if len(got) != len(want) {
+		t.Fatalf("omni_video should include text-to-video and image-to-video capabilities: got=%v want=%v", got, want)
+	}
+	for index := range want {
+		if got[index] != want[index] {
+			t.Fatalf("omni_video capability mismatch at %d: got=%v want=%v", index, got, want)
+		}
+	}
+}
+
 func TestTaskBillingModelIdentityKeepsRequestedModelPrimary(t *testing.T) {
 	identity := taskBillingModelIdentity(GatewayTask{
 		Model:          "doubao-5.0 图像编辑",

apps/api/migrations/0074_keling_omni_base_video_capabilities.sql

@@ -0,0 +1,182 @@
+WITH keling_omni_models(provider_model_name, model_type, capabilities) AS (
+  VALUES
+    (
+      'kling-video-o1',
+      '["video_generate","image_to_video","omni_video"]'::jsonb,
+      '{
+        "video_generate": {
+          "output_resolutions": ["720p", "1080p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10],
+          "output_audio": false,
+          "input_audio": false,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "image_to_video": {
+          "output_resolutions": ["720p", "1080p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10],
+          "input_first_frame": true,
+          "input_last_frame": false,
+          "input_first_last_frame": true,
+          "input_reference_generate_single": true,
+          "input_reference_generate_multiple": true,
+          "max_images": 7,
+          "max_images_for_last_frame": 2,
+          "support_video_effect_template": false,
+          "output_audio": false,
+          "input_audio": false,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "omni_video": {
+          "supported_modes": ["text_to_video", "image_reference", "element_reference", "first_last_frame", "video_reference", "video_edit"],
+          "output_resolutions": ["720p", "1080p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10],
+          "output_audio": false,
+          "input_audio": false,
+          "max_videos": 1,
+          "max_audios": 0,
+          "max_images": 7,
+          "max_elements": 7,
+          "max_images_and_elements": 7,
+          "limits_with_video": {"max_images_and_elements": 4},
+          "max_images_for_last_frame": 2,
+          "support_instruction_edit": true,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "originalTypes": ["video_generate", "image_to_video", "omni_video"]
+      }'::jsonb
+    ),
+    (
+      'kling-v3-omni',
+      '["video_generate","image_to_video","omni_video"]'::jsonb,
+      '{
+        "video_generate": {
+          "output_resolutions": ["720p", "1080p", "2160p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
+          "output_audio": true,
+          "input_audio": false,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "image_to_video": {
+          "output_resolutions": ["720p", "1080p", "2160p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
+          "input_first_frame": true,
+          "input_last_frame": false,
+          "input_first_last_frame": true,
+          "input_reference_generate_single": true,
+          "input_reference_generate_multiple": true,
+          "max_images": 7,
+          "max_images_for_last_frame": 2,
+          "support_video_effect_template": false,
+          "output_audio": true,
+          "input_audio": false,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "omni_video": {
+          "supported_modes": ["text_to_video", "image_reference", "element_reference", "first_last_frame", "video_reference", "video_edit", "multi_shot"],
+          "output_resolutions": ["720p", "1080p", "2160p"],
+          "aspect_ratio_allowed": ["16:9", "1:1", "9:16"],
+          "duration_options": [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
+          "output_audio": true,
+          "input_audio": false,
+          "max_videos": 1,
+          "max_audios": 0,
+          "max_images": 7,
+          "max_elements": 7,
+          "max_images_and_elements": 7,
+          "limits_with_video": {
+            "max_images_and_elements": 4,
+            "duration_options": [3, 4, 5, 6, 7, 8, 9, 10]
+          },
+          "max_images_for_last_frame": 2,
+          "support_instruction_edit": true,
+          "prompt_length_limit": {
+            "max": 2500,
+            "count_mode": "non_ascii_weighted",
+            "label": "可灵口径"
+          }
+        },
+        "originalTypes": ["video_generate", "image_to_video", "omni_video"]
+      }'::jsonb
+    )
+)
+UPDATE base_model_catalog model
+SET model_type = defs.model_type,
+    capabilities = defs.capabilities,
+    metadata = COALESCE(model.metadata, '{}'::jsonb)
+      || jsonb_build_object(
+        'originalTypes', defs.model_type,
+        'rawModel', COALESCE(model.metadata->'rawModel', '{}'::jsonb)
+          || jsonb_build_object(
+            'types', defs.model_type,
+            'capabilities', defs.capabilities
+          )
+      ),
+    default_snapshot = CASE
+      WHEN COALESCE(model.default_snapshot, '{}'::jsonb) = '{}'::jsonb THEN model.default_snapshot
+      ELSE jsonb_set(
+        jsonb_set(
+          jsonb_set(model.default_snapshot, '{modelType}', defs.model_type, true),
+          '{capabilities}',
+          defs.capabilities,
+          true
+        ),
+        '{metadata}',
+        COALESCE(model.default_snapshot->'metadata', '{}'::jsonb)
+          || jsonb_build_object(
+            'originalTypes', defs.model_type,
+            'rawModel', COALESCE(model.default_snapshot->'metadata'->'rawModel', '{}'::jsonb)
+              || jsonb_build_object(
+                'types', defs.model_type,
+                'capabilities', defs.capabilities
+              )
+          ),
+        true
+      )
+    END,
+    updated_at = now()
+FROM keling_omni_models defs
+WHERE model.provider_model_name = defs.provider_model_name
+  AND model.model_type @> '["omni_video"]'::jsonb;
+
+WITH keling_omni_models(provider_model_name, model_type, capabilities) AS (
+  SELECT DISTINCT ON (model.provider_model_name)
+    model.provider_model_name,
+    model.model_type,
+    model.capabilities
+  FROM base_model_catalog model
+  WHERE model.provider_model_name IN ('kling-video-o1', 'kling-v3-omni')
+    AND model.model_type @> '["omni_video"]'::jsonb
+  ORDER BY model.provider_model_name, (model.provider_key = 'keling') DESC, model.created_at ASC
+)
+UPDATE platform_models model
+SET model_type = defs.model_type,
+    capabilities = defs.capabilities,
+    updated_at = now()
+FROM keling_omni_models defs
+WHERE COALESCE(NULLIF(model.provider_model_name, ''), model.model_name) = defs.provider_model_name
+  AND model.model_type @> '["omni_video"]'::jsonb;