diff --git a/backend/internal/protocol/builtin.go b/backend/internal/protocol/builtin.go
index 19777a361..12359b584 100644
--- a/backend/internal/protocol/builtin.go
+++ b/backend/internal/protocol/builtin.go
@@ -367,10 +367,22 @@ func arkVideosAdapter() Adapter {
 	info := metadata("volcengine-ark-video", "火山方舟视频", "Volcengine Ark", CapabilityVideo, "POST /api/v3/contents/generations/tasks", "GET /api/v3/contents/generations/tasks/{task_id}", "application/json")
 	info.Parameters = videoParams()
 	return videoAdapter(info, func(r GenerationRequest) (RequestSpec, error) {
+		if len(r.Images) > 9 || len(r.Videos) > 3 || len(r.Audios) > 3 {
+			return RequestSpec{}, fmt.Errorf("火山方舟全模态参考最多支持 9 张图片、3 个视频和 3 个音频")
+		}
+		if len(r.Audios) > 0 && len(r.Images) == 0 && len(r.Videos) == 0 {
+			return RequestSpec{}, fmt.Errorf("火山方舟全模态参考不支持纯音频或文本+音频，请同时添加参考图片或参考视频")
+		}
 		content := []any{map[string]any{"type": "text", "text": r.Prompt}}
 		for _, image := range r.Images {
 			content = append(content, map[string]any{"type": "image_url", "image_url": map[string]any{"url": mediaValue(image)}, "role": "reference_image"})
 		}
+		for _, video := range r.Videos {
+			content = append(content, map[string]any{"type": "video_url", "video_url": map[string]any{"url": mediaValue(video)}, "role": "reference_video"})
+		}
+		for _, audio := range r.Audios {
+			content = append(content, map[string]any{"type": "audio_url", "audio_url": map[string]any{"url": mediaValue(audio)}, "role": "reference_audio"})
+		}
 		body := map[string]any{"model": r.Model, "content": content, "ratio": defaultValue(r.AspectRatio, "16:9"), "resolution": defaultValue(r.Resolution, "720p"), "duration": defaultInt(r.Duration, 5)}
 		if r.GenerateAudio {
 			body["generate_audio"] = true
diff --git a/backend/internal/protocol/builtin_test.go b/backend/internal/protocol/builtin_test.go
index 3200a76e5..7a1663424 100644
--- a/backend/internal/protocol/builtin_test.go
+++ b/backend/internal/protocol/builtin_test.go
@@ -150,6 +150,41 @@ func TestImageAndVideoAdaptersMapProviderShapes(t *testing.T) {
 	}
 }
 
+func TestArkVideoAdapterMapsFullModalReferences(t *testing.T) {
+	adapter, ok := Builtins().Get("volcengine-ark-video")
+	if !ok {
+		t.Fatal("adapter missing")
+	}
+	spec, err := adapter.BuildCreate(context.Background(), RequestContext{Request: GenerationRequest{
+		Model:  "doubao-seedance-2-0-260128",
+		Prompt: "follow all references",
+		Images: []MediaReference{{URL: "https://example.com/subject.png"}},
+		Videos: []MediaReference{{URL: "https://example.com/motion.mp4"}},
+		Audios: []MediaReference{{URL: "https://example.com/music.mp3"}},
+	}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	body := spec.Body.(map[string]any)
+	content := body["content"].([]any)
+	want := []string{"text", "image_url", "video_url", "audio_url"}
+	if len(content) != len(want) {
+		t.Fatalf("content = %#v", content)
+	}
+	for index, item := range content {
+		if item.(map[string]any)["type"] != want[index] {
+			t.Fatalf("content[%d] = %#v, want type %q", index, item, want[index])
+		}
+	}
+
+	_, err = adapter.BuildCreate(context.Background(), RequestContext{Request: GenerationRequest{
+		Model: "doubao-seedance-2-0-260128", Prompt: "follow the soundtrack", Audios: []MediaReference{{URL: "https://example.com/music.mp3"}},
+	}})
+	if err == nil || !strings.Contains(err.Error(), "文本+音频") {
+		t.Fatalf("audio-only combination error = %v", err)
+	}
+}
+
 func TestAsyncVideoResponseNormalizesStatusAndResult(t *testing.T) {
 	adapter, _ := Builtins().Get("minimax-video")
 	created, err := adapter.ParseCreate(context.Background(), []byte(`{"task":{"id":"mm-1","status":"processing"}}`))
diff --git a/backend/internal/protocol/docs/volcengine-ark-video.md b/backend/internal/protocol/docs/volcengine-ark-video.md
index 15d0b5754..35e89298e 100644
--- a/backend/internal/protocol/docs/volcengine-ark-video.md
+++ b/backend/internal/protocol/docs/volcengine-ark-video.md
@@ -1,6 +1,6 @@
 # 火山方舟视频任务
 
-本插件使用 Ark Contents Generations Tasks 异步接口。提示词和参考图组成 `content[]`；比例、分辨率、时长、音频和水印位于顶层任务参数。
+本插件使用 Ark Contents Generations Tasks 异步接口。提示词、参考图、参考视频和参考音频组成 `content[]`；比例、分辨率、时长、生成音频和水印位于顶层任务参数。
 
 ## 接口与鉴权
 
@@ -21,7 +21,13 @@ Content-Type: application/json
 
 {{PARAMETERS}}
 
-每张图片构造为 `{"type":"image_url","image_url":{"url":"..."},"role":"reference_image"}`。当前不发送参考视频与参考音频；`generate_audio` 和 `watermark` 只在 true 时发送。
+全模态参考的素材映射如下：
+
+- 图片：`{"type":"image_url","image_url":{"url":"..."},"role":"reference_image"}`
+- 视频：`{"type":"video_url","video_url":{"url":"..."},"role":"reference_video"}`
+- 音频：`{"type":"audio_url","audio_url":{"url":"..."},"role":"reference_audio"}`
+
+单次最多发送 9 张图片、3 个视频和 3 个音频。参考素材可以组合使用，但上游不支持纯音频或“文本 + 音频”；使用音频时必须同时提供至少一张参考图或一个参考视频。`generate_audio` 和 `watermark` 只在 true 时发送。
 
 ## 创建任务示例
 
@@ -32,8 +38,10 @@ curl "{channel_base_url}/api/v3/contents/generations/tasks" \
   -d '{
     "model":"YOUR_ARK_VIDEO_MODEL",
     "content":[
-      {"type":"text","text":"保持人物一致，镜头缓慢环绕"},
-      {"type":"image_url","image_url":{"url":"https://example.com/subject.png"},"role":"reference_image"}
+      {"type":"text","text":"沿用图片 1 的主体、视频 1 的运镜和音频 1 的音乐风格"},
+      {"type":"image_url","image_url":{"url":"https://example.com/subject.png"},"role":"reference_image"},
+      {"type":"video_url","video_url":{"url":"https://example.com/motion.mp4"},"role":"reference_video"},
+      {"type":"audio_url","audio_url":{"url":"https://example.com/music.mp3"},"role":"reference_audio"}
     ],
     "ratio":"16:9",
     "resolution":"720p",
@@ -53,5 +61,6 @@ curl "{channel_base_url}/api/v3/contents/generations/tasks" \
 
 - [火山方舟视频生成 API](https://www.volcengine.com/docs/82379)
 - [火山方舟内容生成任务文档](https://www.volcengine.com/docs/82379/1520757)
+- [Doubao Seedance 2.0 全模态参考教程](https://console.volcengine.com/ark/region:cn-beijing/docs/82379/2291680?lang=zh#50e1b4ea)
 
 {{CONTRACT}}
diff --git a/backend/internal/service/model_capability.go b/backend/internal/service/model_capability.go
index 578e4b2f7..3163dfc06 100644
--- a/backend/internal/service/model_capability.go
+++ b/backend/internal/service/model_capability.go
@@ -206,6 +206,7 @@ func DefaultModelCapabilityConfigForModel(protocol string, modelName string) *Mo
 		video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{4, 6, 8}, Default: 6}
 		video.Resolutions = []string{"720p", "1080p"}
 	case model.ChannelInterfaceVolcengineArkVideo:
+		video.Operations = append(video.Operations, "reference_to_video", "audio_to_video")
 		video.References.MaxVideos, video.References.MaxAudios = 3, 3
 		video.References.MaxVideoBytes, video.References.MaxAudioBytes = 200*1024*1024, 15*1024*1024
 		video.References.MaxVideoDuration, video.References.MaxAudioDuration = 15, 15
@@ -601,6 +602,9 @@ func validateVideoTask(profile *VideoCapabilityConfig, input canvasGenerationInp
 	if len(input.ReferenceImages) > profile.References.MaxImages || len(input.ReferenceVideos) > profile.References.MaxVideos || len(input.ReferenceAudios) > profile.References.MaxAudios {
 		return BadAuthRequest("参考素材数量超过当前模型限制")
 	}
+	if input.Config.InterfaceType == string(model.ChannelInterfaceVolcengineArkVideo) && len(input.ReferenceAudios) > 0 && len(input.ReferenceImages) == 0 && len(input.ReferenceVideos) == 0 {
+		return BadAuthRequest("火山方舟全模态参考不支持纯音频或文本+音频，请同时添加参考图片或参考视频")
+	}
 	if len(input.ReferenceImages) < profile.References.MinImages {
 		return BadAuthRequest(fmt.Sprintf("当前视频模型至少需要 %d 张参考图", profile.References.MinImages))
 	}
diff --git a/backend/internal/service/model_capability_test.go b/backend/internal/service/model_capability_test.go
index 831201a39..2ce2991c1 100644
--- a/backend/internal/service/model_capability_test.go
+++ b/backend/internal/service/model_capability_test.go
@@ -93,6 +93,40 @@ func TestDefaultMiniMaxVideoCapabilitySupportsReferenceGeneration(t *testing.T)
 	}
 }
 
+func TestDefaultVolcengineArkVideoCapabilitySupportsFullModalReference(t *testing.T) {
+	profile := DefaultModelCapabilityConfigForModel("volcengine-ark-video", "doubao-seedance-2-0-260128")
+	if profile == nil || profile.Video == nil {
+		t.Fatal("Volcengine Ark video profile = nil")
+	}
+	for _, operation := range []string{"reference_to_video", "audio_to_video"} {
+		if !containsCapabilityString(profile.Video.Operations, operation) {
+			t.Fatalf("operations = %v, want %s", profile.Video.Operations, operation)
+		}
+	}
+	if profile.Video.References.MaxImages != 9 || profile.Video.References.MaxVideos != 3 || profile.Video.References.MaxAudios != 3 {
+		t.Fatalf("reference limits = %#v", profile.Video.References)
+	}
+}
+
+func TestValidateVolcengineArkFullModalReferenceRejectsTextAndAudioOnly(t *testing.T) {
+	profile := DefaultModelCapabilityConfigForModel("volcengine-ark-video", "doubao-seedance-2-0-260128").Video
+	input := canvasGenerationInput{
+		Prompt:          "follow the soundtrack",
+		Config:          providerConfig{InterfaceType: "volcengine-ark-video", VideoSeconds: "6", Size: "16:9", VQuality: "720p"},
+		ReferenceAudios: []providerMedia{{URL: "https://example.com/music.mp3"}},
+		Metadata:        map[string]interface{}{"videoEditOperation": "audio_to_video"},
+	}
+	err := validateVideoTask(profile, input)
+	if err == nil || !strings.Contains(err.Error(), "文本+音频") {
+		t.Fatalf("validateVideoTask() error = %v", err)
+	}
+
+	input.ReferenceImages = []providerMedia{{URL: "https://example.com/subject.png"}}
+	if err := validateVideoTask(profile, input); err != nil {
+		t.Fatalf("validateVideoTask(full modal) error = %v", err)
+	}
+}
+
 func TestCapabilitySpecFromModelCapabilityConfigRestoresLegacyWildcardImageSizes(t *testing.T) {
 	config := &ModelCapabilityConfig{
 		Version: 1,
diff --git a/backend/internal/service/provider_test.go b/backend/internal/service/provider_test.go
index bd1b7b58c..9e0916d63 100644
--- a/backend/internal/service/provider_test.go
+++ b/backend/internal/service/provider_test.go
@@ -1446,12 +1446,19 @@ func TestVolcengineArkVideoProtocolUsesContentTaskAndDownloadsResult(t *testing.
 				t.Fatalf("decode request: %v", err)
 			}
 			content, _ := body["content"].([]interface{})
-			if len(content) != 2 {
+			if len(content) != 4 {
 				t.Errorf("body = %#v", body)
 				return
 			}
-			imageContent, _ := content[1].(map[string]interface{})
-			if body["model"] != "doubao-seedance-test" || imageContent["role"] != "reference_image" {
+			wantTypes := []string{"text", "image_url", "video_url", "audio_url"}
+			wantRoles := []string{"", "reference_image", "reference_video", "reference_audio"}
+			for index, item := range content {
+				entry, _ := item.(map[string]interface{})
+				if entry["type"] != wantTypes[index] || (wantRoles[index] != "" && entry["role"] != wantRoles[index]) {
+					t.Errorf("content[%d] = %#v", index, entry)
+				}
+			}
+			if body["model"] != "doubao-seedance-test" {
 				t.Errorf("body = %#v", body)
 			}
 			_, _ = w.Write([]byte(`{"id":"ark-task-1","status":"running"}`))
@@ -1470,6 +1477,8 @@ func TestVolcengineArkVideoProtocolUsesContentTaskAndDownloadsResult(t *testing.
 		Prompt:          "make it move",
 		Config:          providerConfig{BaseURL: server.URL + "/api/v3", APIKey: "test-key", Model: "doubao-seedance-test", InterfaceType: "volcengine-ark-video"},
 		ReferenceImages: []providerMedia{{ID: "start", URL: server.URL + "/reference.png"}},
+		ReferenceVideos: []providerMedia{{ID: "motion", URL: server.URL + "/reference.mp4"}},
+		ReferenceAudios: []providerMedia{{ID: "music", URL: server.URL + "/reference.mp3"}},
 		Metadata:        map[string]interface{}{"videoStartFrameNodeId": "start"},
 	})
 	if err != nil {
diff --git a/web/src/components/canvas/canvas-config-node-panel.tsx b/web/src/components/canvas/canvas-config-node-panel.tsx
index faf74e151..9c30b51ec 100644
--- a/web/src/components/canvas/canvas-config-node-panel.tsx
+++ b/web/src/components/canvas/canvas-config-node-panel.tsx
@@ -30,6 +30,7 @@ type CanvasConfigNodePanelProps = {
 const videoOperationOptions: Array<{ label: string; value: CanvasVideoEditOperation }> = [
     { label: "文生视频", value: "text_to_video" },
     { label: "图生视频", value: "image_to_video" },
+    { label: "全模态参考", value: "reference_to_video" },
     { label: "视频续写", value: "extend" },
     { label: "局部修改", value: "inpaint" },
     { label: "元素替换", value: "replace_element" },
diff --git a/web/src/components/canvas/canvas-video-segment-dialog.tsx b/web/src/components/canvas/canvas-video-segment-dialog.tsx
index 2f3688071..dc3a4896f 100644
--- a/web/src/components/canvas/canvas-video-segment-dialog.tsx
+++ b/web/src/components/canvas/canvas-video-segment-dialog.tsx
@@ -413,6 +413,7 @@ function videoOperationLabel(operation: CanvasVideoEditOperation) {
     const labels: Record<CanvasVideoEditOperation, string> = {
         text_to_video: "文生视频",
         image_to_video: "图生视频",
+        reference_to_video: "全模态参考",
         extend: "视频续写",
         inpaint: "局部修改",
         replace_element: "元素替换",
diff --git a/web/src/components/model-capability-editor.tsx b/web/src/components/model-capability-editor.tsx
index 23583ba07..aab079d30 100644
--- a/web/src/components/model-capability-editor.tsx
+++ b/web/src/components/model-capability-editor.tsx
@@ -9,6 +9,7 @@ const ratioOptions = ["16:9", "9:16", "1:1", "4:3", "3:4", "21:9"];
 const operationOptions = [
     { label: "文生视频", value: "text_to_video" },
     { label: "图生视频", value: "image_to_video" },
+    { label: "全模态参考", value: "reference_to_video" },
     { label: "视频续写", value: "extend" },
     { label: "局部修改", value: "inpaint" },
     { label: "元素替换", value: "replace_element" },
diff --git a/web/src/components/model-picker.tsx b/web/src/components/model-picker.tsx
index ab709abc7..7e34d0567 100644
--- a/web/src/components/model-picker.tsx
+++ b/web/src/components/model-picker.tsx
@@ -294,6 +294,7 @@ function logicalCapabilitySummary(spec: NonNullable<NonNullable<AiConfig["channe
     const operationLabels: Record<string, string> = {
         text_to_video: "文生视频",
         image_to_video: "图生视频",
+        reference_to_video: "全模态参考",
         audio_to_video: "音频生视频",
         extend: "视频续写",
         inpaint: "局部修改",
diff --git a/web/src/lib/generation-task-display.ts b/web/src/lib/generation-task-display.ts
index 6b59806aa..75903f978 100644
--- a/web/src/lib/generation-task-display.ts
+++ b/web/src/lib/generation-task-display.ts
@@ -43,6 +43,7 @@ export const operationOptions = [
     { label: "Agent 会话：拆解影视工作流", value: "agent_session" },
     { label: "文生视频", value: "text_to_video" },
     { label: "图生视频", value: "image_to_video" },
+    { label: "全模态参考", value: "reference_to_video" },
     { label: "视频续写", value: "extend" },
     { label: "视频局部修改", value: "inpaint" },
     { label: "元素替换", value: "replace_element" },
diff --git a/web/src/lib/model-capabilities.ts b/web/src/lib/model-capabilities.ts
index abb984841..14848c7c8 100644
--- a/web/src/lib/model-capabilities.ts
+++ b/web/src/lib/model-capabilities.ts
@@ -292,7 +292,10 @@ export function defaultModelCapabilityConfig(protocol?: ModelProtocol, model = "
         video.generateAudio = { supported: true, default: true };
     }
     if (protocol === "volcengine-ark-video" || protocol === "newapi-channel-1") video.resolutions = ["480p", "720p", "1080p"];
-    if (protocol === "volcengine-ark-video") video.watermark = { supported: true, default: false };
+    if (protocol === "volcengine-ark-video") {
+        video.watermark = { supported: true, default: false };
+        video.operations.push("reference_to_video", "audio_to_video");
+    }
     if (protocol === "novita-video") {
         video.references.maxImages = 1;
         video.references.maxImageBytes = 10 * 1024 * 1024;
diff --git a/web/src/lib/model-selection.ts b/web/src/lib/model-selection.ts
index 35d539cf6..dbcef2281 100644
--- a/web/src/lib/model-selection.ts
+++ b/web/src/lib/model-selection.ts
@@ -336,7 +336,7 @@ function videoOperationLabel(operation: string) {
     if (operation === "text_to_video") return "文生视频";
     if (operation === "image_to_video") return "图生视频";
     if (operation === "audio_to_video") return "音频生视频";
-    if (operation === "reference_to_video") return "参考素材生视频";
+    if (operation === "reference_to_video") return "全模态参考";
     if (operation === "extend") return "视频续写";
     return "当前生成模式";
 }
diff --git a/web/src/pages/admin/logical-models/model-routing-capabilities.tsx b/web/src/pages/admin/logical-models/model-routing-capabilities.tsx
index bb950660b..077b6f11d 100644
--- a/web/src/pages/admin/logical-models/model-routing-capabilities.tsx
+++ b/web/src/pages/admin/logical-models/model-routing-capabilities.tsx
@@ -637,6 +637,7 @@ export function operationLabel(value: string) {
         {
             text_to_video: "文生视频",
             image_to_video: "图生视频",
+            reference_to_video: "全模态参考",
             extend: "视频续写",
             inpaint: "局部修改",
             replace_element: "元素替换",
diff --git a/web/src/services/api/video-provider-seedance.ts b/web/src/services/api/video-provider-seedance.ts
index bc92776bd..a82e69d09 100644
--- a/web/src/services/api/video-provider-seedance.ts
+++ b/web/src/services/api/video-provider-seedance.ts
@@ -81,8 +81,8 @@ function seedanceApiUrl(config: ResolvedAiConfig, taskId?: string) {
 }
 
 async function buildSeedanceAgentPlanPayload(config: ResolvedAiConfig, model: string, prompt: string, references: ReferenceImage[], videoReferences: ReferenceVideo[], audioReferences: ReferenceAudio[], deps: VideoProviderDeps) {
-    if (config.interfaceType !== "volcengine-ark-video" && audioReferences.length && !references.length && !videoReferences.length) {
-        throw new Error("Seedance 参考音频不能单独使用，请同时添加参考图或参考视频");
+    if (audioReferences.length && !references.length && !videoReferences.length) {
+        throw new Error(config.interfaceType === "volcengine-ark-video" ? "火山方舟全模态参考不支持纯音频或文本+音频，请同时添加参考图片或参考视频" : "Seedance 参考音频不能单独使用，请同时添加参考图或参考视频");
     }
     const content = config.interfaceType === "volcengine-ark-video"
         ? await buildVolcengineArkContent(prompt, references, videoReferences, audioReferences)
diff --git a/web/src/types/canvas.ts b/web/src/types/canvas.ts
index 721fe9449..c137af395 100644
--- a/web/src/types/canvas.ts
+++ b/web/src/types/canvas.ts
@@ -50,7 +50,7 @@ export type CanvasGenerationBatchStatus = "queued" | "running" | "partial_failed
 export type CanvasGenerationBatchItemStatus = "waiting" | "submitting" | "queued" | "running" | "succeeded" | "failed" | "cancelled";
 export type CanvasImageGenerationType = "generation" | "edit";
 export type CanvasWorkflowKind = "free" | "script" | "story_input" | "character" | "scene" | "storyboard" | "shot" | "final" | "styleboard" | "reference_set" | "reference_video" | "action_board";
-export type CanvasVideoEditOperation = "text_to_video" | "image_to_video" | "extend" | "inpaint" | "replace_element" | "camera_motion" | "style_transfer" | "audio_to_video" | "compare_versions" | "concat";
+export type CanvasVideoEditOperation = "text_to_video" | "image_to_video" | "reference_to_video" | "extend" | "inpaint" | "replace_element" | "camera_motion" | "style_transfer" | "audio_to_video" | "compare_versions" | "concat";
 export type CanvasSkillCategory = "writing" | "storyboard" | "image" | "video" | "utility";
 export type CanvasSkillOutputMode = "text" | "json" | "image_prompt" | "workflow";
 export type StoryboardColumn =
diff --git a/web/test/video-api-contracts.test.ts b/web/test/video-api-contracts.test.ts
index aee69819f..5a8017087 100644
--- a/web/test/video-api-contracts.test.ts
+++ b/web/test/video-api-contracts.test.ts
@@ -1,6 +1,9 @@
 import { describe, expect, test } from "bun:test";
 
-import { unwrapEnvelope, videoTaskId } from "../src/services/api/video-response";
+import { defaultConfig } from "../src/stores/use-config-store";
+import { createSeedanceTask } from "../src/services/api/video-provider-seedance";
+import type { VideoProviderDeps } from "../src/services/api/video-provider-deps";
+import { unwrapEnvelope, videoResponseTools, videoTaskId } from "../src/services/api/video-response";
 import { isPublicMediaUrl, normalizeVideoResolution, normalizeVideoSeconds, normalizeVideoSize } from "../src/services/api/video-validation";
 
 describe("video API response contracts", () => {
@@ -34,3 +37,66 @@ describe("video request normalization", () => {
         expect(isPublicMediaUrl("/resources/video.mp4")).toBe(false);
     });
 });
+
+describe("Volcengine Ark full-modal references", () => {
+    const model = "seedance-2-0-250824";
+    const config = {
+        ...defaultConfig,
+        baseUrl: "https://ark.cn-beijing.volces.com/api/v3",
+        apiKey: "test-key",
+        interfaceType: "volcengine-ark-video",
+        channels: [
+            {
+                id: "ark",
+                name: "Ark",
+                baseUrl: "https://ark.cn-beijing.volces.com/api/v3",
+                apiKey: "test-key",
+                secretKey: "",
+                headers: [],
+                apiFormat: "openai",
+                interfaceType: "volcengine-ark-video",
+                models: [model],
+                scope: "user",
+                enabled: true,
+            },
+        ],
+    };
+
+    test("映射图片、视频和音频参考素材", async () => {
+        let requestUrl = "";
+        let requestBody: unknown;
+        const deps = {
+            transport: {
+                post: async (url: string, body: unknown) => {
+                    requestUrl = url;
+                    requestBody = body;
+                    return { id: "task-1", status: "queued" };
+                },
+            },
+            response: videoResponseTools,
+        } as unknown as VideoProviderDeps;
+
+        await createSeedanceTask(
+            deps,
+            config as never,
+            model,
+            "保持主体一致",
+            [{ id: "image-1", name: "image.png", type: "image/png", dataUrl: "", url: "https://cdn.example.com/image.png" }],
+            [{ id: "video-1", name: "video.mp4", type: "video/mp4", url: "https://cdn.example.com/video.mp4" }],
+            [{ id: "audio-1", name: "audio.mp3", type: "audio/mpeg", url: "https://cdn.example.com/audio.mp3" }],
+        );
+
+        expect(requestUrl).toBe("https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks");
+        expect((requestBody as { content: unknown[] }).content).toEqual([
+            { type: "text", text: "保持主体一致" },
+            { type: "image_url", image_url: { url: "https://cdn.example.com/image.png" }, role: "reference_image" },
+            { type: "video_url", video_url: { url: "https://cdn.example.com/video.mp4" }, role: "reference_video" },
+            { type: "audio_url", audio_url: { url: "https://cdn.example.com/audio.mp3" }, role: "reference_audio" },
+        ]);
+    });
+
+    test("拒绝纯音频和文本加音频", async () => {
+        const deps = { transport: {}, response: videoResponseTools } as unknown as VideoProviderDeps;
+        await expect(createSeedanceTask(deps, config as never, model, "跟随节奏", [], [], [{ id: "audio-1", name: "audio.mp3", type: "audio/mpeg", url: "https://cdn.example.com/audio.mp3" }])).rejects.toThrow("不支持纯音频或文本+音频");
+    });
+});
diff --git a/web/test/video-generation-options.test.ts b/web/test/video-generation-options.test.ts
index c76564d07..a385e7d62 100644
--- a/web/test/video-generation-options.test.ts
+++ b/web/test/video-generation-options.test.ts
@@ -22,4 +22,8 @@ describe("video generation resolution options", () => {
         expect(defaultModelCapabilityConfig("volcengine-jimeng-video").video?.resolutions).toEqual(["720p"]);
         expect(defaultModelCapabilityConfig("gemini-veo").video?.resolutions).toEqual(["720p", "1080p"]);
     });
+
+    test("火山方舟默认开启全模态参考模式", () => {
+        expect(defaultModelCapabilityConfig("volcengine-ark-video").video?.operations).toEqual(expect.arrayContaining(["reference_to_video", "audio_to_video"]));
+    });
 });