diff --git a/app/(admin)/settings/admin-settings-client.tsx b/app/(admin)/settings/admin-settings-client.tsx
index ca122b776..b4a292286 100644
--- a/app/(admin)/settings/admin-settings-client.tsx
+++ b/app/(admin)/settings/admin-settings-client.tsx
@@ -1562,6 +1562,7 @@ export function AdminSettingsClient() {
     "qianwen-video",
     "fal-video",
     "ruxa-video",
+    "runway",
   ].includes(selectedModelProviderType);
 
   // 存储配置状态
@@ -9363,12 +9364,8 @@ export function AdminSettingsClient() {
                     <SelectItem value="auto">不指定首选（完全随机轮换）</SelectItem>
                     {(modelsData?.models || [])
                       .filter((m: AiModel) => {
-                        // 过滤出支持图片生成/编辑的模型
                         const provider = (providersData?.providers || []).find((p: AiProvider) => p.id === m.providerId);
-                        if (!provider) return false;
-                        // 支持阿里生图、Gemini、既梦等图片模型
-                        const imageProviderTypes = ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "midjourney", "gemini-image", "gemini", "volcano-vision"];
-                        return imageProviderTypes.includes(provider.type);
+                        return Boolean(provider && m.supportsImageGen);
                       })
                       .map((m: AiModel) => {
                         const provider = (providersData?.providers || []).find((p: AiProvider) => p.id === m.providerId);
@@ -9382,8 +9379,7 @@ export function AdminSettingsClient() {
                     {/* 如果没有找到生图模型，显示提示 */}
                     {(modelsData?.models || []).filter((m: AiModel) => {
                       const provider = (providersData?.providers || []).find((p: AiProvider) => p.id === m.providerId);
-                      const imageProviderTypes = ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "midjourney", "gemini-image", "gemini", "volcano-vision"];
-                      return provider && imageProviderTypes.includes(provider.type);
+                      return Boolean(provider && m.supportsImageGen);
                     }).length === 0 && (
                       <SelectItem value="no-model" disabled>
                         暂无可用的生图模型，请先配置生图供应商
diff --git a/app/(chat)/api/chat/route.ts b/app/(chat)/api/chat/route.ts
index da9137f25..677611f59 100644
--- a/app/(chat)/api/chat/route.ts
+++ b/app/(chat)/api/chat/route.ts
@@ -98,6 +98,10 @@ import {
   isGridConfirmationReply,
   type PendingGridConfirmation,
 } from "@/lib/ai/conversation-action-resolver";
+import {
+  isImageGenerationModelDescriptor,
+  isVideoModelDescriptor,
+} from "@/lib/ai/model-classification";
 import {
   bootstrapConversationStateProtocol,
   composeConversationStateProtocol,
@@ -457,6 +461,7 @@ function normalizeImageLoadingProviderType(providerType?: string | null): string
     normalized === "fal-image" ||
     normalized === "openai-image" ||
     normalized === "midjourney" ||
+    normalized === "runway" ||
     normalized === "volcano-vision" ||
     normalized === "gemini"
   ) {
@@ -583,21 +588,53 @@ function startStreamKeepalive(writer: StreamWriter, label: string, intervalMs =
 
 type VideoModeType = "text-to-video" | "first-frame" | "first-last-frame" | "reference-images";
 
-const VIDEO_PROVIDER_TYPES = new Set([
-  "volcano-video",
-  "doubao-video",
-  "qianwen-video",
-  "fal-video",
-  "ruxa-video",
-]);
+function isVideoProviderType(
+  providerType?: string | null,
+  model?: {
+    modelId?: string | null;
+    capabilities?: unknown;
+    supportsImageGen?: boolean | null;
+  }
+): boolean {
+  return isVideoModelDescriptor({
+    providerType,
+    modelId: model?.modelId,
+    capabilities: model?.capabilities,
+    supportsImageGen: model?.supportsImageGen,
+  });
+}
 
-function isVideoProviderType(providerType?: string | null): boolean {
-  return Boolean(providerType && VIDEO_PROVIDER_TYPES.has(providerType));
+function isImageGenerationModelType(
+  providerType?: string | null,
+  model?: {
+    modelId?: string | null;
+    capabilities?: unknown;
+    supportsImageGen?: boolean | null;
+  }
+): boolean {
+  return isImageGenerationModelDescriptor({
+    providerType,
+    modelId: model?.modelId,
+    capabilities: model?.capabilities,
+    supportsImageGen: model?.supportsImageGen,
+  });
 }
 
 function inferSupportedVideoModesByModelId(modelId: string): VideoModeType[] {
   const normalized = modelId.toLowerCase();
 
+  if (
+    normalized === "gen4.5" ||
+    normalized === "gen4_turbo" ||
+    normalized === "gen4_aleph" ||
+    normalized === "gen3a_turbo" ||
+    normalized === "veo3" ||
+    normalized === "veo3.1" ||
+    normalized === "veo3.1_fast"
+  ) {
+    return ["text-to-video", "first-frame"];
+  }
+
   if (normalized.includes("lite-t2v")) {
     return ["text-to-video"];
   }
@@ -1341,6 +1378,7 @@ const IMAGE_FAILOVER_PROVIDER_ALLOWLIST = new Set([
   "ruxa-image",
   "fal-image",
   "openai-image",
+  "runway",
   "volcano-vision",
   "gemini",
   "huanke-gemini",
@@ -1980,7 +2018,8 @@ async function convertMessagesWithImages(messages: ChatMessage[], providerType:
     providerType === "ruxa-image" ||
     providerType === "fal-image" ||
     providerType === "openai-image" ||
-    providerType === "midjourney"
+    providerType === "midjourney" ||
+    providerType === "runway"
   ) {
     const convertedMessages = [];
 
@@ -3156,6 +3195,7 @@ export async function POST(request: Request) {
     let requestedModelProviderType: string | undefined;
     let requestedModelIsFixedVisual = false;
     let requestedModelIsConversationMultimodal = false;
+    let requestedModelConfig: Awaited<ReturnType<typeof getAiModelById>> | null = null;
 
     if (!shouldRunAgentConversation && requestedModelId !== "auto") {
       const selectedModelIsUUID =
@@ -3183,33 +3223,21 @@ export async function POST(request: Request) {
         selectedModelProviderType = selectedProvider?.type as string | undefined;
       }
 
+      requestedModelConfig = selectedModelConfig;
       requestedModelProviderType = selectedModelProviderType;
 
       if (selectedModelConfig) {
-        const fixedImageProviderTypes = new Set([
-          "gemini",
-          "huanke-gemini",
-          "alibaba-image",
-          "ruxa-image",
-          "fal-image",
-          "openai-image",
-          "midjourney",
-          "volcano-vision",
-        ]);
         const isFixedVisualModel =
           !selectedModelConfig.isSubModel &&
-          (selectedModelConfig.supportsImageGen ||
-            (selectedModelProviderType
-              ? isVideoProviderType(selectedModelProviderType)
-              : false) ||
-            (selectedModelProviderType
-              ? fixedImageProviderTypes.has(selectedModelProviderType)
-              : false));
+          (
+            isImageGenerationModelType(selectedModelProviderType, selectedModelConfig) ||
+            isVideoProviderType(selectedModelProviderType, selectedModelConfig)
+          );
         const isConversationMultimodalModel =
           !selectedModelConfig.isSubModel &&
           isAiModelMultimodal(selectedModelConfig) &&
-          !selectedModelConfig.supportsImageGen &&
-          !isVideoProviderType(selectedModelProviderType);
+          !isImageGenerationModelType(selectedModelProviderType, selectedModelConfig) &&
+          !isVideoProviderType(selectedModelProviderType, selectedModelConfig);
         requestedModelIsFixedVisual = isFixedVisualModel;
         requestedModelIsConversationMultimodal = isConversationMultimodalModel;
 
@@ -3331,7 +3359,8 @@ export async function POST(request: Request) {
     }
 
     const allowAgentVideoGeneration =
-      Boolean(requestBody.videoParams) || isVideoProviderType(requestedModelProviderType);
+      Boolean(requestBody.videoParams) ||
+      isVideoProviderType(requestedModelProviderType, requestedModelConfig || undefined);
     let smartAutoWorkflowDecision:
       | {
           action?: string;
@@ -3985,7 +4014,9 @@ export async function POST(request: Request) {
           Boolean(smartAutoExecution?.shouldStayConversational);
 
         if (autoGraphResult.route && (!shouldKeepSelectedModel || shouldForceConversationalRouteForVisualModel)) {
-          const routeIsVideo = isVideoProviderType(autoGraphResult.route.providerType);
+          const routeIsVideo = isVideoProviderType(autoGraphResult.route.providerType, {
+            modelId: autoGraphResult.route.modelId,
+          });
           if (routeIsVideo && !allowAgentVideoGeneration) {
             const fallbackIntent: "image_edit" | "image_gen" = hasImage ? "image_edit" : "image_gen";
             smartAutoDetectedIntent = fallbackIntent;
@@ -4021,7 +4052,7 @@ export async function POST(request: Request) {
                 subIntent: autoGraphResult.analysis?.subIntent,
                 hasImage,
               });
-              if (!isVideoProviderType(fallbackRoute.providerType)) {
+              if (!isVideoProviderType(fallbackRoute.providerType, { modelId: fallbackRoute.modelId })) {
                 finalModelId = fallbackRoute.id;
                 console.log(
                   `[Smart Auto Mode] ✅ Fallback routed to: ${fallbackRoute.displayName} (${fallbackRoute.reason})`,
@@ -4281,7 +4312,7 @@ export async function POST(request: Request) {
             (m) =>
               m.model.supportsImageGen &&
               !m.model.isSubModel &&
-              !isVideoProviderType(m.provider?.type),
+              !isVideoProviderType(m.provider?.type, m.model),
           );
           if (isAutoTimeout) {
               smartAutoFriendlyMessage = buildAutoModeFallbackMessage({
@@ -4309,12 +4340,15 @@ export async function POST(request: Request) {
         }
         if (!bestModel) {
           bestModel = allModels.find(
-            (m) => m.model.isDefault && !m.model.isSubModel && !isVideoProviderType(m.provider?.type),
+            (m) =>
+              m.model.isDefault &&
+              !m.model.isSubModel &&
+              !isVideoProviderType(m.provider?.type, m.model),
           );
         }
         if (!bestModel && allModels.length > 0) {
           bestModel = allModels.find(
-            (m) => !m.model.isSubModel && !isVideoProviderType(m.provider?.type),
+            (m) => !m.model.isSubModel && !isVideoProviderType(m.provider?.type, m.model),
           ) || allModels.find((m) => !m.model.isSubModel) || allModels[0];
         }
         
@@ -4335,9 +4369,10 @@ export async function POST(request: Request) {
       const finalModelIsUuid =
         /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(finalModelId);
       let finalProviderType: string | undefined;
+      let finalModelConfig: Awaited<ReturnType<typeof getAiModelById>> | null = null;
 
       if (finalModelIsUuid) {
-        const finalModelConfig = await getAiModelById({ id: finalModelId }).catch(() => null);
+        finalModelConfig = await getAiModelById({ id: finalModelId }).catch(() => null);
         if (finalModelConfig) {
           finalProviderType = (
             await getAiProviderById({ id: finalModelConfig.providerId }).catch(() => null)
@@ -4345,19 +4380,19 @@ export async function POST(request: Request) {
         }
       }
 
-      if (isVideoProviderType(finalProviderType)) {
+      if (isVideoProviderType(finalProviderType, finalModelConfig || undefined)) {
         const allModels = await getEnabledAiModelsWithProvider(false);
         const fallbackModel =
           allModels.find(
             (m) =>
               !m.model.isSubModel &&
               m.model.supportsImageGen &&
-              !isVideoProviderType(m.provider?.type),
+              !isVideoProviderType(m.provider?.type, m.model),
           ) ||
           allModels.find(
             (m) =>
               !m.model.isSubModel &&
-              !isVideoProviderType(m.provider?.type),
+              !isVideoProviderType(m.provider?.type, m.model),
           );
 
         if (fallbackModel) {
@@ -7364,22 +7399,31 @@ export async function POST(request: Request) {
     }
 
     // ========== 图片生成配额检查 ==========
-    const isImageGenerationModel = 
-      (languageModel as any).provider === "gemini-image" || 
-      (languageModel as any).provider === "volcano-vision" ||
-      (languageModel as any).provider === "alibaba-image" ||
-      (languageModel as any).provider === "ruxa-image" ||
-      (languageModel as any).provider === "fal-image" ||
-      (languageModel as any).provider === "openai-image" ||
-      (languageModel as any).provider === "midjourney" ||
-      (languageModel as any).provider === "huanke-gemini";
-
     const isUuidModelId = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(
       finalModelId
     );
     const currentModelConfig = isUuidModelId
       ? await getAiModelById({ id: finalModelId }).catch(() => undefined)
       : undefined;
+    const runtimeProviderType = String(
+      resolvedProviderType || (languageModel as any).provider || ""
+    );
+    const normalizedRuntimeProviderType =
+      runtimeProviderType === "gemini-image"
+        ? "gemini"
+        : runtimeProviderType === "huanke-gemini"
+          ? "huanke-gemini"
+          : runtimeProviderType;
+    const isImageGenerationModel = isImageGenerationModelType(
+      normalizedRuntimeProviderType,
+      currentModelConfig
+        ? {
+            modelId: currentModelConfig.modelId,
+            capabilities: currentModelConfig.capabilities,
+            supportsImageGen: currentModelConfig.supportsImageGen,
+          }
+        : undefined
+    );
     const modelCreditCostPerImage = Number(
       (currentModelConfig as any)?.imageGenConfig?.creditCostPerImage ?? 0
     );
@@ -7438,13 +7482,16 @@ export async function POST(request: Request) {
       }
     }
 
-    const isVideoGenerationModel = [
-      "qianwen-video",
-      "fal-video",
-      "ruxa-video",
-      "volcano-video",
-      "doubao-video",
-    ].includes(String(resolvedProviderType || (languageModel as any).provider || ""));
+    const isVideoGenerationModel = isVideoProviderType(
+      normalizedRuntimeProviderType,
+      currentModelConfig
+        ? {
+            modelId: currentModelConfig.modelId,
+            capabilities: currentModelConfig.capabilities,
+            supportsImageGen: currentModelConfig.supportsImageGen,
+          }
+        : undefined
+    );
 
     if (isVideoGenerationModel) {
       const pointsConfig = await getPointsSystemConfig();
@@ -7925,7 +7972,14 @@ export async function POST(request: Request) {
     // ========== Gemini 处理结束 ==========
 
     // ========== 阿里云千问图片生成独立处理 ==========
-    const isAlibabaLikeImageProvider = ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "midjourney"].includes((languageModel as any).provider);
+    const isAlibabaLikeImageProvider = [
+      "alibaba-image",
+      "ruxa-image",
+      "fal-image",
+      "openai-image",
+      "midjourney",
+      "runway",
+    ].includes((languageModel as any).provider);
     if (isAlibabaLikeImageProvider && typeof (languageModel as any).doGenerate === "function") {
       console.log("[Chat API] Using image generation provider:", (languageModel as any).provider);
       console.log("[Chat API] Image params:", effectiveImageParams);
@@ -8013,6 +8067,8 @@ export async function POST(request: Request) {
                       ? "openai-image"
                       : (languageModel as any).provider === "midjourney"
                         ? "midjourney"
+                      : (languageModel as any).provider === "runway"
+                        ? "runway"
                     : "alibaba-image";
               const modelMessages = await convertMessagesWithImages(messagesForAlibaba, providerForConvert);
               
@@ -8085,8 +8141,9 @@ export async function POST(request: Request) {
                   "ruxa-image": 2,
                   "fal-image": 3,
                   "openai-image": 4,
-                  gemini: 5,
-                  "huanke-gemini": 6,
+                  runway: 5,
+                  gemini: 6,
+                  "huanke-gemini": 7,
                 };
                 const crossProviderCandidates = getImageCrossProviderCandidates(allModels, {
                   currentModelId: finalModelId,
@@ -8115,6 +8172,8 @@ export async function POST(request: Request) {
                             ? "openai-image"
                             : candidate.provider.type === "midjourney"
                               ? "midjourney"
+                            : candidate.provider.type === "runway"
+                              ? "runway"
                             : candidate.provider.type === "gemini"
                               ? "gemini"
                               : candidate.provider.type === "huanke-gemini"
@@ -8473,9 +8532,10 @@ export async function POST(request: Request) {
                   "ruxa-image": 1,
                   "fal-image": 2,
                   "openai-image": 3,
-                  gemini: 4,
-                  "huanke-gemini": 5,
-                  "volcano-vision": 6,
+                  runway: 4,
+                  gemini: 5,
+                  "huanke-gemini": 6,
+                  "volcano-vision": 7,
                 };
                 const crossProviderCandidates = getImageCrossProviderCandidates(allModels, {
                   currentModelId: finalModelId,
@@ -8504,6 +8564,8 @@ export async function POST(request: Request) {
                             ? "openai-image"
                             : candidate.provider.type === "midjourney"
                               ? "midjourney"
+                            : candidate.provider.type === "runway"
+                              ? "runway"
                             : candidate.provider.type === "gemini"
                               ? "gemini"
                               : candidate.provider.type === "huanke-gemini"
@@ -8710,13 +8772,14 @@ export async function POST(request: Request) {
 
     // ========== 千问/FAL/RUXA 视频生成独立处理 ==========
     if (
-      ["qianwen-video", "fal-video", "ruxa-video"].includes((languageModel as any).provider) &&
+      ["qianwen-video", "fal-video", "ruxa-video", "runway"].includes((languageModel as any).provider) &&
       typeof (languageModel as any).doGenerate === "function"
     ) {
       const currentVideoProvider = (languageModel as any).provider as
         | "qianwen-video"
         | "fal-video"
-        | "ruxa-video";
+        | "ruxa-video"
+        | "runway";
       console.log(`[Chat API] Using ${currentVideoProvider} video generation`);
       
       const videoGenerationStream = createUIMessageStream({
@@ -9286,8 +9349,9 @@ export async function POST(request: Request) {
                   "ruxa-image": 2,
                   "fal-image": 3,
                   "openai-image": 4,
-                  gemini: 5,
-                  "huanke-gemini": 6,
+                  runway: 5,
+                  gemini: 6,
+                  "huanke-gemini": 7,
                 };
                 const crossProviderCandidates = getImageCrossProviderCandidates(allModels, {
                   currentModelId: finalModelId,
@@ -9321,6 +9385,8 @@ export async function POST(request: Request) {
                             ? "openai-image"
                             : candidate.provider.type === "midjourney"
                               ? "midjourney"
+                            : candidate.provider.type === "runway"
+                              ? "runway"
                             : candidate.provider.type === "gemini"
                               ? "gemini"
                               : candidate.provider.type === "huanke-gemini"
diff --git a/app/(chat)/page.tsx b/app/(chat)/page.tsx
index dc324994c..c8df373d6 100644
--- a/app/(chat)/page.tsx
+++ b/app/(chat)/page.tsx
@@ -14,14 +14,7 @@ import {
   getSystemSetting,
 } from "@/lib/db/queries";
 import { isAiModelAgentAuto } from "@/lib/db/ai-model-flags";
-
-const VIDEO_PROVIDER_TYPES = new Set([
-  "volcano-video",
-  "doubao-video",
-  "qianwen-video",
-  "fal-video",
-  "ruxa-video",
-]);
+import { isVideoModelDescriptor } from "@/lib/ai/model-classification";
 
 function isNanoBananaProModel(model: { displayName?: string | null; modelId?: string | null }) {
   const displayName = String(model.displayName || "").toLowerCase();
@@ -109,7 +102,12 @@ async function NewChatPage() {
   const preferredImageModels = visibleModels.filter(
     ({ model, provider }) =>
       model.supportsImageGen &&
-      !VIDEO_PROVIDER_TYPES.has(provider.type)
+      !isVideoModelDescriptor({
+        providerType: provider.type,
+        modelId: model.modelId,
+        capabilities: model.capabilities,
+        supportsImageGen: model.supportsImageGen,
+      })
   );
   const nanoBananaModel = preferredImageModels.find(({ model }) =>
     isNanoBananaProModel(model)
diff --git a/lib/ai/dynamic-provider.ts b/lib/ai/dynamic-provider.ts
index 947687631..016cdda72 100644
--- a/lib/ai/dynamic-provider.ts
+++ b/lib/ai/dynamic-provider.ts
@@ -24,6 +24,7 @@ import { createOpenAIImage } from "./providers/openai-image";
 import { createMidjourney } from "./providers/midjourney";
 import { createFalVideo } from "./providers/fal-video";
 import { createRuxaVideo } from "./providers/ruxa-video";
+import { createRunway } from "./providers/runway";
 import { createVolcanoMultimodal } from "./providers/volcano-multimodal";
 import { getProviderDefaultBaseUrl } from "./provider-defaults";
 
@@ -227,6 +228,15 @@ function getOrCreateProvider(provider: Pick<AiProvider, "id" | "type" | "baseUrl
       return midjourneyProvider as any;
     }
 
+    case "runway": {
+      const runwayProvider = createRunway({
+        baseURL: resolvedBaseUrl,
+        apiKey: provider.apiKey,
+      });
+      console.log(`[Provider] Created Runway provider: ${resolvedBaseUrl}`);
+      return runwayProvider as any;
+    }
+
     case "custom":
     default: {
       if (isVolcanoMultimodalBaseUrl(resolvedBaseUrl)) {
@@ -353,6 +363,10 @@ export async function getDynamicModel(
       console.log("[getDynamicModel] Getting Midjourney chat model..."); // 调试日志
       baseModel = providerInstance.chat(resolvedModelId);
       console.log("[getDynamicModel] Midjourney model retrieved:", baseModel ? "success" : "failed"); // 调试日志
+    } else if (provider.type === "runway") {
+      console.log("[getDynamicModel] Getting Runway chat model..."); // 调试日志
+      baseModel = providerInstance.chat(resolvedModelId);
+      console.log("[getDynamicModel] Runway model retrieved:", baseModel ? "success" : "failed"); // 调试日志
     } else if (provider.type === "huanke-gemini") {
       // 环科 Gemini 原生格式使用 chat() 方法
       console.log("[getDynamicModel] Getting Huanke Gemini chat model..."); // 调试日志
diff --git a/lib/ai/model-classification.ts b/lib/ai/model-classification.ts
new file mode 100644
index 000000000..7209fc26f
--- /dev/null
+++ b/lib/ai/model-classification.ts
@@ -0,0 +1,114 @@
+const STATIC_VIDEO_PROVIDER_TYPES = new Set([
+  "doubao-video",
+  "volcano-video",
+  "qianwen-video",
+  "fal-video",
+  "ruxa-video",
+]);
+
+const STATIC_IMAGE_PROVIDER_TYPES = new Set([
+  "gemini",
+  "huanke-gemini",
+  "alibaba-image",
+  "ruxa-image",
+  "fal-image",
+  "openai-image",
+  "midjourney",
+  "volcano-vision",
+]);
+
+const RUNWAY_IMAGE_MODEL_IDS = new Set([
+  "gen4_image",
+  "gen4_image_turbo",
+  "gemini_2.5_flash",
+]);
+
+const RUNWAY_VIDEO_MODEL_PREFIXES = [
+  "gen4.5",
+  "gen4_turbo",
+  "gen4_aleph",
+  "gen3a_turbo",
+  "veo3.1_fast",
+  "veo3.1",
+  "veo3",
+  "act_two",
+  "upscale_v1",
+];
+
+const VIDEO_MODE_KEYS = new Set([
+  "text-to-video",
+  "first-frame",
+  "first-last-frame",
+  "reference-images",
+]);
+
+function normalizeText(value?: string | null) {
+  return String(value || "").trim().toLowerCase();
+}
+
+function hasVideoModes(capabilities: unknown) {
+  if (!capabilities || typeof capabilities !== "object") return false;
+  const supportedModes = (capabilities as { supportedModes?: unknown }).supportedModes;
+  if (!Array.isArray(supportedModes)) return false;
+  return supportedModes.some((mode) => VIDEO_MODE_KEYS.has(String(mode || "").trim()));
+}
+
+export function inferRunwayModelKind(modelId?: string | null): "image" | "video" | null {
+  const normalized = normalizeText(modelId);
+  if (!normalized) return null;
+  if (RUNWAY_IMAGE_MODEL_IDS.has(normalized)) return "image";
+  if (RUNWAY_VIDEO_MODEL_PREFIXES.some((prefix) => normalized === prefix || normalized.startsWith(`${prefix}_`))) {
+    return "video";
+  }
+  return null;
+}
+
+export function isStaticVideoProviderType(providerType?: string | null): boolean {
+  return STATIC_VIDEO_PROVIDER_TYPES.has(normalizeText(providerType));
+}
+
+export function isVideoModelDescriptor(params: {
+  providerType?: string | null;
+  modelId?: string | null;
+  capabilities?: unknown;
+  supportsImageGen?: boolean | null;
+}): boolean {
+  const providerType = normalizeText(params.providerType);
+
+  if (isStaticVideoProviderType(providerType)) {
+    return true;
+  }
+
+  if (hasVideoModes(params.capabilities)) {
+    return true;
+  }
+
+  if (providerType === "runway") {
+    return inferRunwayModelKind(params.modelId) === "video";
+  }
+
+  return false;
+}
+
+export function isImageGenerationModelDescriptor(params: {
+  providerType?: string | null;
+  modelId?: string | null;
+  capabilities?: unknown;
+  supportsImageGen?: boolean | null;
+}): boolean {
+  const providerType = normalizeText(params.providerType);
+
+  if (params.supportsImageGen === true) {
+    return true;
+  }
+
+  if (providerType === "runway") {
+    return inferRunwayModelKind(params.modelId) === "image";
+  }
+
+  if (isVideoModelDescriptor(params)) {
+    return false;
+  }
+
+  return STATIC_IMAGE_PROVIDER_TYPES.has(providerType);
+}
diff --git a/lib/ai/model-router.ts b/lib/ai/model-router.ts
index 09e0896e5..f8eaed78e 100644
--- a/lib/ai/model-router.ts
+++ b/lib/ai/model-router.ts
@@ -6,6 +6,10 @@
  */
 
 import { isAiModelMultimodal } from "@/lib/db/ai-model-flags";
+import {
+  isImageGenerationModelDescriptor,
+  isVideoModelDescriptor,
+} from "@/lib/ai/model-classification";
 
 type EnabledModelWithProvider = Awaited<
   ReturnType<typeof import("@/lib/db/queries")["getEnabledAiModelsWithProvider"]>
@@ -42,14 +46,6 @@ const CHAT_PROVIDER_TYPES = [
   "volcano-multimodal",
 ] as const;
 
-const VIDEO_PROVIDER_TYPES = new Set([
-  "doubao-video",
-  "volcano-video",
-  "qianwen-video",
-  "fal-video",
-  "ruxa-video",
-]);
-
 const IMAGE_ONLY_PROVIDER_TYPES = new Set([
   "alibaba-image",
   "ruxa-image",
@@ -59,8 +55,20 @@ const IMAGE_ONLY_PROVIDER_TYPES = new Set([
   "gemini-image",
 ]);
 
-function isVideoProviderType(providerType: string) {
-  return VIDEO_PROVIDER_TYPES.has(providerType);
+function isVideoProviderType(
+  providerType: string,
+  model?: {
+    modelId?: string | null;
+    capabilities?: unknown;
+    supportsImageGen?: boolean | null;
+  }
+) {
+  return isVideoModelDescriptor({
+    providerType,
+    modelId: model?.modelId,
+    capabilities: model?.capabilities,
+    supportsImageGen: model?.supportsImageGen,
+  });
 }
 
 function isConversationalCandidate(
@@ -69,7 +77,7 @@ function isConversationalCandidate(
 ) {
   const providerType = String(item.provider.type || "");
   if (!item.model.isEnabled || item.model.isSubModel) return false;
-  if (isVideoProviderType(providerType)) return false;
+  if (isVideoProviderType(providerType, item.model)) return false;
 
   const requireVision = Boolean(options?.requireVision);
   const matchesPreferredProvider = CHAT_PROVIDER_TYPES.includes(
@@ -108,13 +116,27 @@ function matchesIntentRequirements(
 
   if (params.intent === "vision") {
     return (
-      !isVideoProviderType(String(item.provider.type || "")) &&
+      !isVideoProviderType(String(item.provider.type || ""), item.model) &&
       hasVisionCapability(item.model)
     );
   }
 
+  if (params.intent === "video_gen") {
+    return isVideoProviderType(String(item.provider.type || ""), item.model);
+  }
+
   if (params.requireVision && !hasVisionCapability(item.model)) return false;
-  if (params.requireImageGen && !item.model.supportsImageGen) return false;
+  if (
+    params.requireImageGen &&
+    !isImageGenerationModelDescriptor({
+      providerType: item.provider.type,
+      modelId: item.model.modelId,
+      capabilities: item.model.capabilities,
+      supportsImageGen: item.model.supportsImageGen,
+    })
+  ) {
+    return false;
+  }
 
   return true;
 }
@@ -132,7 +154,7 @@ const INTENT_MODEL_MAP: Record<
 > = {
   // 图片编辑：优先 alibaba-image（灵魂4.5，支持提示词扩展）
   image_edit: {
-    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "gemini", "huanke-gemini"],
+    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway", "gemini", "huanke-gemini"],
     requireVision: true,
     requireImageGen: true,
     description: "图片编辑",
@@ -147,14 +169,14 @@ const INTENT_MODEL_MAP: Record<
 
   // 图片生成：优先灵魂4.5
   image_gen: {
-    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "gemini", "huanke-gemini", "volcano-vision"],
+    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway", "gemini", "huanke-gemini", "volcano-vision"],
     requireImageGen: true,
     description: "图片生成",
   },
 
   // 视频生成
   video_gen: {
-    providerTypes: ["doubao-video", "volcano-video", "qianwen-video", "fal-video", "ruxa-video"],
+    providerTypes: ["doubao-video", "volcano-video", "qianwen-video", "fal-video", "ruxa-video", "runway"],
     description: "视频生成",
   },
 
@@ -175,25 +197,25 @@ const SUB_INTENT_OVERRIDES: Record<
 > = {
   // 抠图：alibaba-image 效果最好
   remove_background: {
-    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "gemini"],
+    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway", "gemini"],
     reason: "抠图首选灵魂4.5",
   },
 
   // 换背景：alibaba-image 支持场景生成
   replace_background: {
-    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "gemini"],
+    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway", "gemini"],
     reason: "换背景首选灵魂4.5",
   },
 
   // 图片增强
   enhance: {
-    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "gemini", "volcano-vision"],
+    providerTypes: ["alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway", "gemini", "volcano-vision"],
     reason: "增强首选灵魂4.5",
   },
 
   // 风格迁移：gemini 可能更有创意
   style_transfer: {
-    providerTypes: ["gemini", "alibaba-image", "ruxa-image", "fal-image", "openai-image"],
+    providerTypes: ["gemini", "alibaba-image", "ruxa-image", "fal-image", "openai-image", "runway"],
     reason: "风格迁移优选Gemini",
   },
 };
@@ -321,7 +343,7 @@ export async function routeToModel(params: {
     return (
       !item.model.isSubModel &&
       item.model.isEnabled &&
-      !isVideoProviderType(String(item.provider.type || "")) &&
+      !isVideoProviderType(String(item.provider.type || ""), item.model) &&
       (item.model.supportsImageGen || hasVisionCapability(item.model))
     );
   });
@@ -345,7 +367,7 @@ export async function routeToModel(params: {
           (m) =>
             !m.model.isSubModel &&
             m.model.isEnabled &&
-            !isVideoProviderType(String(m.provider.type || "")) &&
+            !isVideoProviderType(String(m.provider.type || ""), m.model) &&
             !m.model.supportsImageGen,
         )
       : undefined) ||
diff --git a/lib/ai/providers/runway.ts b/lib/ai/providers/runway.ts
new file mode 100644
index 000000000..96fd84e8c
--- /dev/null
+++ b/lib/ai/providers/runway.ts
@@ -0,0 +1,581 @@
+import "server-only";
+
+import sharp from "sharp";
+import { uploadFile } from "@/lib/storage/service";
+import { inferRunwayModelKind } from "@/lib/ai/model-classification";
+
+interface RunwayConfig {
+  apiKey: string;
+  baseURL: string;
+}
+
+interface RunwayImageParams {
+  aspectRatio?: string;
+  imageSize?: string;
+  imageCount?: number;
+  seed?: number;
+}
+
+interface RunwayVideoParams {
+  mode?: "text-to-video" | "first-frame" | "first-last-frame" | "reference-images";
+  ratio?: string;
+  duration?: number;
+  seed?: number;
+}
+
+interface UploadedImage {
+  thumbnail: string;
+  original: string;
+}
+
+const RUNWAY_VERSION = "2024-11-06";
+const DEFAULT_BASE_URL = "https://api.dev.runwayml.com";
+const TASK_POLL_INTERVAL_MS = 5000;
+const TASK_TIMEOUT_MS = 15 * 60 * 1000;
+
+const IMAGE_RESOLUTION_MAP: Record<string, Record<string, string>> = {
+  "1K": {
+    "21:9": "1584:672",
+    "16:9": "1360:768",
+    "9:16": "768:1360",
+    "4:3": "1104:832",
+    "3:4": "832:1104",
+    "1:1": "960:960",
+  },
+  "2K": {
+    "21:9": "1584:672",
+    "16:9": "1360:768",
+    "9:16": "768:1360",
+    "4:3": "1104:832",
+    "3:4": "832:1104",
+    "1:1": "960:960",
+  },
+  "4K": {
+    "21:9": "1584:672",
+    "16:9": "1360:768",
+    "9:16": "768:1360",
+    "4:3": "1104:832",
+    "3:4": "832:1104",
+    "1:1": "960:960",
+  },
+};
+
+function normalizeBaseURL(baseURL: string) {
+  const normalized = String(baseURL || "").trim().replace(/\/+$/, "");
+  return normalized || DEFAULT_BASE_URL;
+}
+
+function joinEndpoint(baseURL: string, path: string) {
+  const cleanPath = path.startsWith("/") ? path : `/${path}`;
+  return `${normalizeBaseURL(baseURL)}${cleanPath}`;
+}
+
+function buildHeaders(apiKey: string) {
+  return {
+    Authorization: `Bearer ${String(apiKey || "").trim()}`,
+    "Content-Type": "application/json",
+    "X-Runway-Version": RUNWAY_VERSION,
+  };
+}
+
+function normalizeText(value: string) {
+  return String(value || "").replace(/\u00a0/g, " ").trim().normalize("NFKC");
+}
+
+function normalizeImageSizeLabel(imageSize?: string): "1K" | "2K" | "4K" {
+  const normalized = String(imageSize || "").trim().toUpperCase();
+  if (normalized === "1K" || normalized === "2K" || normalized === "4K") {
+    return normalized;
+  }
+  return "2K";
+}
+
+function normalizeImageResolution(aspectRatio?: string, imageSize?: string) {
+  const normalizedRatio = String(aspectRatio || "").trim() || "16:9";
+  const normalizedSize = normalizeImageSizeLabel(imageSize);
+  return (
+    IMAGE_RESOLUTION_MAP[normalizedSize]?.[normalizedRatio] ||
+    IMAGE_RESOLUTION_MAP[normalizedSize]?.["16:9"] ||
+    "1360:768"
+  );
+}
+
+function normalizeVideoRatio(ratio?: string) {
+  const normalized = String(ratio || "").trim();
+  if (["21:9", "16:9", "4:3", "1:1", "3:4", "9:16"].includes(normalized)) {
+    return normalized;
+  }
+  return "16:9";
+}
+
+function normalizeVideoDuration(duration?: number) {
+  if (!Number.isFinite(duration)) return 5;
+  const normalized = Math.round(Number(duration));
+  return normalized >= 8 ? 10 : 5;
+}
+
+function normalizeSeed(seed?: number) {
+  if (!Number.isFinite(seed)) return undefined;
+  return Math.max(0, Math.min(4294967295, Math.round(Number(seed))));
+}
+
+function detectExtensionFromContentType(contentType?: string | null) {
+  const normalized = String(contentType || "").toLowerCase();
+  if (normalized.includes("jpeg") || normalized.includes("jpg")) return "jpg";
+  if (normalized.includes("webp")) return "webp";
+  if (normalized.includes("gif")) return "gif";
+  if (normalized.includes("png")) return "png";
+  if (normalized.includes("mp4")) return "mp4";
+  if (normalized.includes("quicktime")) return "mov";
+  if (normalized.includes("webm")) return "webm";
+  return "bin";
+}
+
+function extractDataUriParts(dataUri: string) {
+  const match = dataUri.match(/^data:([^;]+);base64,(.+)$/);
+  if (!match) {
+    throw new Error("不支持的 Data URI 格式");
+  }
+  return {
+    mimeType: match[1],
+    buffer: Buffer.from(match[2], "base64"),
+  };
+}
+
+function generateFilename(prefix: string, extension: string) {
+  const timestamp = Date.now();
+  const random = Math.random().toString(36).slice(2, 8);
+  return `${prefix}-${timestamp}-${random}.${extension}`;
+}
+
+async function uploadGeneratedBuffer(
+  buffer: Buffer,
+  extension: string,
+  mimeType: string
+): Promise<UploadedImage> {
+  const thumbnailBuffer = await sharp(buffer)
+    .resize(400, 400, { fit: "inside", withoutEnlargement: true })
+    .jpeg({ quality: 80 })
+    .toBuffer();
+
+  const originalFilename = generateFilename("runway-image", extension);
+  const originalArrayBuffer = buffer.buffer.slice(
+    buffer.byteOffset,
+    buffer.byteOffset + buffer.byteLength
+  ) as ArrayBuffer;
+  const originalUploaded = await uploadFile(originalFilename, originalArrayBuffer);
+
+  const thumbnailFilename = generateFilename("runway-thumb", "jpg");
+  const thumbnailArrayBuffer = thumbnailBuffer.buffer.slice(
+    thumbnailBuffer.byteOffset,
+    thumbnailBuffer.byteOffset + thumbnailBuffer.byteLength
+  ) as ArrayBuffer;
+  const thumbnailUploaded = await uploadFile(thumbnailFilename, thumbnailArrayBuffer);
+
+  console.log("[Runway] Uploaded image to storage:", {
+    original: originalUploaded.url,
+    thumbnail: thumbnailUploaded.url,
+    mimeType,
+  });
+
+  return {
+    thumbnail: thumbnailUploaded.url,
+    original: originalUploaded.url,
+  };
+}
+
+async function uploadDataUriAsset(dataUri: string, prefix: "runway-input" | "runway-video") {
+  const { mimeType, buffer } = extractDataUriParts(dataUri);
+  const extension = detectExtensionFromContentType(mimeType);
+  const filename = generateFilename(prefix, extension);
+  const arrayBuffer = buffer.buffer.slice(
+    buffer.byteOffset,
+    buffer.byteOffset + buffer.byteLength
+  ) as ArrayBuffer;
+  const uploaded = await uploadFile(filename, arrayBuffer);
+  return uploaded.url;
+}
+
+async function normalizeAssetUrl(url?: string | null) {
+  const normalized = String(url || "").trim();
+  if (!normalized) return "";
+  if (/^https:\/\//i.test(normalized) || /^runway:\/\//i.test(normalized)) {
+    return normalized;
+  }
+  if (/^data:(image|video)\//i.test(normalized)) {
+    return uploadDataUriAsset(normalized, /^data:video\//i.test(normalized) ? "runway-video" : "runway-input");
+  }
+  return normalized;
+}
+
+function collectHttpUrls(payload: unknown): string[] {
+  const urls = new Set<string>();
+
+  const visit = (value: unknown) => {
+    if (typeof value === "string") {
+      const normalized = value.trim();
+      if (/^https?:\/\//i.test(normalized)) {
+        urls.add(normalized);
+      }
+      return;
+    }
+
+    if (Array.isArray(value)) {
+      value.forEach(visit);
+      return;
+    }
+
+    if (value && typeof value === "object") {
+      Object.values(value as Record<string, unknown>).forEach(visit);
+    }
+  };
+
+  visit(payload);
+  return Array.from(urls);
+}
+
+async function downloadRemoteAsset(url: string) {
+  const controller = new AbortController();
+  const timeoutId = setTimeout(() => controller.abort(), 120000);
+
+  try {
+    const response = await fetch(url, {
+      cache: "no-store",
+      signal: controller.signal,
+    });
+    if (!response.ok) {
+      throw new Error(`下载 Runway 输出失败: ${response.status}`);
+    }
+    const buffer = Buffer.from(await response.arrayBuffer());
+    return {
+      buffer,
+      contentType: response.headers.get("content-type"),
+    };
+  } finally {
+    clearTimeout(timeoutId);
+  }
+}
+
+async function uploadVideoToStorage(remoteVideoUrl: string) {
+  try {
+    const { buffer, contentType } = await downloadRemoteAsset(remoteVideoUrl);
+    const extension = detectExtensionFromContentType(contentType || "video/mp4");
+    const filename = generateFilename("runway-video", extension === "bin" ? "mp4" : extension);
+    const arrayBuffer = buffer.buffer.slice(
+      buffer.byteOffset,
+      buffer.byteOffset + buffer.byteLength
+    ) as ArrayBuffer;
+    const uploaded = await uploadFile(filename, arrayBuffer);
+    return uploaded.url;
+  } catch (error) {
+    console.warn("[Runway] Failed to archive video output, using temporary URL:", error);
+    return remoteVideoUrl;
+  }
+}
+
+function formatImageMarkdown(images: UploadedImage[], imageSize: string) {
+  return images
+    .map((image) => `![thumb|||${image.original}|||${imageSize}](${image.thumbnail})`)
+    .join("\n\n");
+}
+
+function formatVideoMarkdown(videoUrl: string) {
+  return `<video src="${videoUrl}" controls style="max-width: 100%; border-radius: 8px; margin-bottom: 12px;"></video>\n\n[下载视频](${videoUrl})`;
+}
+
+function extractPromptAndAssets(messages: Array<{ role: string; content: any }>) {
+  const lastUserMessage = messages.filter((message) => message.role === "user").pop();
+  if (!lastUserMessage) {
+    throw new Error("未找到用户输入内容");
+  }
+
+  let prompt = "";
+  const imageUrls: string[] = [];
+
+  const pushImageUrl = (value?: string) => {
+    if (!value) return;
+    const normalized = String(value || "").trim();
+    if (!normalized) return;
+    if (!imageUrls.includes(normalized)) {
+      imageUrls.push(normalized);
+    }
+  };
+
+  if (typeof lastUserMessage.content === "string") {
+    prompt = normalizeText(lastUserMessage.content);
+  } else if (Array.isArray(lastUserMessage.content)) {
+    for (const part of lastUserMessage.content) {
+      if (part?.type === "text" && typeof part.text === "string") {
+        prompt = `${prompt} ${part.text}`.trim();
+      } else if (part?.type === "image_url" && part.image_url?.url) {
+        pushImageUrl(part.image_url.url);
+      } else if (part?.type === "file") {
+        pushImageUrl(part.url || part.data);
+      } else if (part?.type === "image") {
+        pushImageUrl(part.image || part.url);
+      }
+    }
+  }
+
+  return {
+    prompt: normalizeText(prompt) || "生成内容",
+    imageUrls,
+  };
+}
+
+async function createTask(baseURL: string, apiKey: string, path: string, body: Record<string, any>) {
+  const response = await fetch(joinEndpoint(baseURL, path), {
+    method: "POST",
+    headers: buildHeaders(apiKey),
+    body: JSON.stringify(body),
+  });
+
+  const rawText = await response.text();
+  let payload: any = null;
+  try {
+    payload = rawText ? JSON.parse(rawText) : null;
+  } catch {
+    payload = rawText;
+  }
+
+  if (!response.ok) {
+    const message =
+      typeof payload?.error === "string"
+        ? payload.error
+        : typeof payload?.message === "string"
+          ? payload.message
+          : rawText || "Runway 请求失败";
+    throw new Error(`${message} (HTTP ${response.status})`);
+  }
+
+  const taskId = String(payload?.id || "").trim();
+  if (!taskId) {
+    throw new Error("Runway 未返回任务 ID");
+  }
+
+  return taskId;
+}
+
+async function waitForTaskOutput(baseURL: string, apiKey: string, taskId: string) {
+  const deadline = Date.now() + TASK_TIMEOUT_MS;
+
+  while (Date.now() < deadline) {
+    const response = await fetch(joinEndpoint(baseURL, `/v1/tasks/${taskId}`), {
+      method: "GET",
+      headers: buildHeaders(apiKey),
+      cache: "no-store",
+    });
+
+    const rawText = await response.text();
+    let payload: any = null;
+    try {
+      payload = rawText ? JSON.parse(rawText) : null;
+    } catch {
+      payload = rawText;
+    }
+
+    if (!response.ok) {
+      const message =
+        typeof payload?.error === "string"
+          ? payload.error
+          : typeof payload?.message === "string"
+            ? payload.message
+            : rawText || "Runway 轮询失败";
+      throw new Error(`${message} (HTTP ${response.status})`);
+    }
+
+    const status = String(payload?.status || "").trim().toUpperCase();
+    if (status === "SUCCEEDED") {
+      return payload;
+    }
+    if (status === "FAILED" || status === "CANCELLED") {
+      throw new Error(
+        String(payload?.failure || payload?.error || payload?.message || "Runway 任务失败")
+      );
+    }
+
+    await new Promise((resolve) => setTimeout(resolve, TASK_POLL_INTERVAL_MS));
+  }
+
+  throw new Error("Runway 任务超时，请稍后重试");
+}
+
+async function uploadRunwayImages(taskOutput: unknown) {
+  const outputUrls = collectHttpUrls(taskOutput);
+  const uploadedImages: UploadedImage[] = [];
+
+  for (const url of outputUrls) {
+    try {
+      const { buffer, contentType } = await downloadRemoteAsset(url);
+      if (!String(contentType || "").startsWith("image/")) {
+        continue;
+      }
+      const extension = detectExtensionFromContentType(contentType);
+      uploadedImages.push(
+        await uploadGeneratedBuffer(
+          buffer,
+          extension === "bin" ? "png" : extension,
+          contentType || "image/png"
+        )
+      );
+    } catch (error) {
+      console.warn("[Runway] Failed to archive image output:", error);
+    }
+  }
+
+  if (uploadedImages.length === 0) {
+    throw new Error("Runway 未返回可用图片");
+  }
+
+  return uploadedImages;
+}
+
+async function resolveRunwayVideoUrl(taskOutput: unknown) {
+  const outputUrls = collectHttpUrls(taskOutput);
+  if (outputUrls.length === 0) {
+    throw new Error("Runway 未返回可用视频地址");
+  }
+
+  for (const url of outputUrls) {
+    if (/(\.mp4|\.mov|\.webm)(\?|$)/i.test(url) || /video/i.test(url)) {
+      return uploadVideoToStorage(url);
+    }
+  }
+
+  return uploadVideoToStorage(outputUrls[0]!);
+}
+
+function getRunwayVideoEndpoint(modelId: string, hasPromptImage: boolean) {
+  const normalized = normalizeText(modelId);
+  if (hasPromptImage) return "/v1/image_to_video";
+  if (normalized === "gen4.5") return "/v1/image_to_video";
+  return "/v1/text_to_video";
+}
+
+function buildReferenceImages(imageUrls: string[]) {
+  return imageUrls.slice(0, 3).map((uri, index) => ({
+    uri,
+    tag: `ref_${index + 1}`.replace(/[^a-z0-9_]/g, "_"),
+  }));
+}
+
+export function createRunway({ apiKey, baseURL }: RunwayConfig) {
+  return {
+    provider: "runway",
+    chat: (modelId: string) => ({
+      provider: "runway",
+      doGenerate: async (options: {
+        messages: Array<{ role: string; content: any }>;
+        imageParams?: RunwayImageParams;
+        videoParams?: RunwayVideoParams;
+        overrideModelId?: string;
+      }) => {
+        try {
+          const activeModelId = normalizeText(options.overrideModelId || modelId);
+          const { prompt, imageUrls } = extractPromptAndAssets(options.messages);
+          const normalizedImageUrls = (
+            await Promise.all(imageUrls.map((url) => normalizeAssetUrl(url)))
+          ).filter(Boolean);
+          const modelKind =
+            inferRunwayModelKind(activeModelId) ||
+            (options.videoParams ? "video" : options.imageParams ? "image" : null) ||
+            "video";
+
+          if (modelKind === "image") {
+            const imageSize = normalizeImageSizeLabel(options.imageParams?.imageSize);
+            const taskId = await createTask(baseURL, apiKey, "/v1/text_to_image", {
+              model: activeModelId,
+              promptText: prompt,
+              ratio: normalizeImageResolution(options.imageParams?.aspectRatio, imageSize),
+              ...(normalizedImageUrls.length > 0
+                ? { referenceImages: buildReferenceImages(normalizedImageUrls) }
+                : {}),
+              ...(normalizeSeed(options.imageParams?.seed) !== undefined
+                ? { seed: normalizeSeed(options.imageParams?.seed) }
+                : {}),
+            });
+
+            const task = await waitForTaskOutput(baseURL, apiKey, taskId);
+            const uploadedImages = await uploadRunwayImages(task?.output ?? task);
+            const text = formatImageMarkdown(uploadedImages, imageSize);
+
+            return {
+              text,
+              finishReason: "stop" as const,
+              usage: {
+                promptTokens: 0,
+                completionTokens: text.length,
+              },
+            };
+          }
+
+          const videoMode = options.videoParams?.mode || (normalizedImageUrls[0] ? "first-frame" : "text-to-video");
+          if (
+            (videoMode === "first-frame" || videoMode === "first-last-frame" || videoMode === "reference-images") &&
+            !normalizedImageUrls[0]
+          ) {
+            throw new Error("Runway 图生视频至少需要 1 张参考图");
+          }
+
+          const endpoint = getRunwayVideoEndpoint(activeModelId, Boolean(normalizedImageUrls[0]));
+          const taskId = await createTask(baseURL, apiKey, endpoint, {
+            model: activeModelId,
+            promptText: prompt,
+            ratio: normalizeVideoRatio(options.videoParams?.ratio),
+            duration: normalizeVideoDuration(options.videoParams?.duration),
+            ...(normalizedImageUrls[0] ? { promptImage: normalizedImageUrls[0] } : {}),
+            ...(normalizeSeed(options.videoParams?.seed) !== undefined
+              ? { seed: normalizeSeed(options.videoParams?.seed) }
+              : {}),
+          });
+
+          const task = await waitForTaskOutput(baseURL, apiKey, taskId);
+          const videoUrl = await resolveRunwayVideoUrl(task?.output ?? task);
+          const text = formatVideoMarkdown(videoUrl);
+
+          return {
+            text,
+            finishReason: "stop" as const,
+            usage: {
+              promptTokens: 0,
+              completionTokens: text.length,
+            },
+          };
+        } catch (error) {
+          console.error("[Runway] Generation failed:", error);
+          return {
+            text:
+              error instanceof Error
+                ? `${
+                    (inferRunwayModelKind(options.overrideModelId || modelId) ||
+                      (options.videoParams ? "video" : options.imageParams ? "image" : "video")) === "image"
+                      ? "图片"
+                      : "视频"
+                  }生成失败: ${error.message}`
+                : "Runway 生成失败，请稍后重试",
+            finishReason: "error" as const,
+            usage: {
+              promptTokens: 0,
+              completionTokens: 0,
+            },
+          };
+        }
+      },
+      doStream: async function* (options: {
+        messages: Array<{ role: string; content: any }>;
+        imageParams?: RunwayImageParams;
+        videoParams?: RunwayVideoParams;
+        overrideModelId?: string;
+      }) {
+        const result = await this.doGenerate(options);
+        yield {
+          type: "text-delta",
+          textDelta: result.text,
+        };
+      },
+      modelId,
+      specificationVersion: "v1" as const,
+      defaultObjectGenerationMode: "json" as const,
+    }),
+  };
+}
diff --git a/lib/ai/video-tool-capability.ts b/lib/ai/video-tool-capability.ts
index a6686893a..e101aaf59 100644
--- a/lib/ai/video-tool-capability.ts
+++ b/lib/ai/video-tool-capability.ts
@@ -193,6 +193,33 @@ function createProviderDefaultCapability(providerType: string): VideoToolCapabil
     };
   }
 
+  if (providerType === "runway") {
+    return {
+      schemaVersion: "v1",
+      source: "provider-default",
+      providerType,
+      modeOptions: normalizeModeOptions(["text-to-video", "first-frame"]),
+      ratioOptions: [...STANDARD_RATIOS],
+      durationOptions: [5, 10],
+      resolutionOptions: [],
+      ui: {
+        showResolution: false,
+        showFps: false,
+        showCameraMove: false,
+        showGenerateAudio: false,
+        showReturnLastFrame: false,
+        showSeed: true,
+      },
+      defaults: {
+        mode: "text-to-video",
+        ratio: "16:9",
+        duration: 5,
+        generateAudio: false,
+        fps: 24,
+      },
+    };
+  }
+
   return {
     schemaVersion: "v1",
     source: "provider-default",
diff --git a/lib/canvas/generate-image.ts b/lib/canvas/generate-image.ts
index 871d37124..87cfddd22 100644
--- a/lib/canvas/generate-image.ts
+++ b/lib/canvas/generate-image.ts
@@ -7,6 +7,7 @@ import { createRuxaImage } from "@/lib/ai/providers/ruxa-image";
 import { createFalImage } from "@/lib/ai/providers/fal-image";
 import { createOpenAIImage } from "@/lib/ai/providers/openai-image";
 import { createMidjourney } from "@/lib/ai/providers/midjourney";
+import { createRunway } from "@/lib/ai/providers/runway";
 import { filterReachableRemoteUrls } from "@/lib/net/url-access";
 
 export interface CanvasGenerateImageRequest {
@@ -39,6 +40,7 @@ const SUPPORTED_CANVAS_PROVIDER_TYPES = new Set([
   "fal-image",
   "openai-image",
   "midjourney",
+  "runway",
 ]);
 
 function parseImagesFromMarkdown(markdown: string): ParsedImage[] {
@@ -76,6 +78,7 @@ function getProviderBatchLimit(providerType: string) {
     case "ruxa-image":
     case "openai-image":
     case "midjourney":
+    case "runway":
       return 1;
     case "gemini":
       return 2;
@@ -151,7 +154,7 @@ export async function generateCanvasImage(
     }
     if (!SUPPORTED_CANVAS_PROVIDER_TYPES.has(fallbackProvider.type)) {
       throw new Error(
-        "当前图片模型类型暂不支持画布生成，请在后台选择 Gemini/阿里/RUXA/FAL/OpenAI 图片模型。"
+        "当前图片模型类型暂不支持画布生成，请在后台选择 Gemini/阿里/RUXA/FAL/OpenAI/Runway 图片模型。"
       );
     }
     if (fallbackModel.model.id !== selectedModel.model.id) {
@@ -311,9 +314,22 @@ export async function generateCanvasImage(
       });
       return result.text || "";
     };
+  } else if (provider.type === "runway") {
+    const runwayProvider = createRunway({
+      apiKey: provider.apiKey,
+      baseURL: provider.baseUrl,
+    });
+    const modelInstance = runwayProvider.chat(selectedModel.model.modelId);
+    executeGenerateBatch = async (_batchImageCount) => {
+      const result = await modelInstance.doGenerate({
+        messages,
+        imageParams,
+      });
+      return result.text || "";
+    };
   } else {
     throw new Error(
-      `不支持的图片生成模型类型: ${provider.type || "unknown"}，请改用 Gemini/阿里/RUXA/FAL/OpenAI/Midjourney 图片模型`
+      `不支持的图片生成模型类型: ${provider.type || "unknown"}，请改用 Gemini/阿里/RUXA/FAL/OpenAI/Midjourney/Runway 图片模型`
     );
   }
 
diff --git a/lib/db/queries.ts b/lib/db/queries.ts
index d0f1620af..59c47b553 100644
--- a/lib/db/queries.ts
+++ b/lib/db/queries.ts
@@ -91,6 +91,7 @@ import {
 } from "../points/config";
 import { inferOriginalImageUrl } from "../image-url";
 import { getAgentRequiredUploadCount } from "../agents/preset-mode-config";
+import { isVideoModelDescriptor } from "../ai/model-classification";
 
 // Optionally, if not using email/pass login, you can
 // use the Drizzle adapter for Auth.js / NextAuth
@@ -1755,14 +1756,6 @@ type AgentPresetKind = "image" | "video" | "text";
 type AgentPresetScope = "all" | AgentPresetVisibility;
 type AgentCategoryScope = "all" | "public" | "private" | "global";
 type AgentPresetSort = "latest" | "hottest" | "most_used";
-const AGENT_VIDEO_PROVIDER_TYPES = new Set([
-  "volcano-video",
-  "doubao-video",
-  "qianwen-video",
-  "fal-video",
-  "ruxa-video",
-]);
-
 function normalizeAgentCategoryName(value: string) {
   return String(value || "").trim().replace(/\s+/g, " ");
 }
@@ -1804,7 +1797,12 @@ async function validateAgentDefaultModel(params: {
     throw new ChatSDKError("bad_request:database", "Invalid default model");
   }
 
-  const isVideoModel = AGENT_VIDEO_PROVIDER_TYPES.has(String(model.providerType || ""));
+  const isVideoModel = isVideoModelDescriptor({
+    providerType: model.providerType,
+    modelId: model.modelId,
+    capabilities: model.capabilities,
+    supportsImageGen: model.supportsImageGen,
+  });
 
   if (params.presetType === "video") {
     if (!isVideoModel) {