diff --git a/web/src/components/canvas/canvas-config-node-panel.tsx b/web/src/components/canvas/canvas-config-node-panel.tsx
index e73f2489a..21477784a 100644
--- a/web/src/components/canvas/canvas-config-node-panel.tsx
+++ b/web/src/components/canvas/canvas-config-node-panel.tsx
@@ -7,8 +7,8 @@ import { configuredModelMatchesCapability, defaultConfig, modelOptionName, resol
 import { CreditSymbol, requestCreditCost } from "@/constant/credits";
 import { canvasThemes } from "@/lib/canvas-theme";
 import { normalizeVideoDuration, normalizeVideoResolution } from "@/lib/video-generation-options";
-import { modelCapabilityConfigFor, normalizeImageValue, normalizeVideoValue } from "@/lib/model-capabilities";
-import { modelCompatibilityError, modelRequestOptions, resolveCompatibleModel, defaultImageParamsForModel, type ModelRequirements } from "@/lib/model-selection";
+import { modelCapabilityConfigFor } from "@/lib/model-capabilities";
+import { modelCompatibilityError, modelRequestOptions, resolveCompatibleModel, resolveModelGenerationDefaults, defaultImageParamsForModel, type ModelRequirements } from "@/lib/model-selection";
 import { navigateToSettings } from "@/lib/settings-navigation";
 import { useThemeStore } from "@/stores/use-theme-store";
 import { useUserStore } from "@/stores/use-user-store";
@@ -50,8 +50,20 @@ export function CanvasConfigNodePanel({ node, isRunning, inputSummary, onConfigC
         capability: mode,
         input: inputSummary,
         videoOperation: node.metadata?.videoEditOperation,
-        videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds,
-		options: modelRequestOptions({ ...globalConfig, size: node.metadata?.size || globalConfig.size, quality: node.metadata?.quality || globalConfig.quality, count: String(node.metadata?.count || globalConfig.count), videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds, vquality: node.metadata?.vquality || globalConfig.vquality, videoGenerateAudio: node.metadata?.generateAudio || globalConfig.videoGenerateAudio, videoWatermark: node.metadata?.watermark || globalConfig.videoWatermark, audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice, audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat, audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed }, mode),
+        videoSeconds: mode === "video" ? node.metadata?.seconds || globalConfig.videoSeconds : undefined,
+        options: modelRequestOptions({
+            ...globalConfig,
+            size: node.metadata?.size || globalConfig.size,
+            quality: node.metadata?.quality || globalConfig.quality,
+            count: String(node.metadata?.count || globalConfig.count),
+            videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds,
+            vquality: node.metadata?.vquality || globalConfig.vquality,
+            videoGenerateAudio: node.metadata?.generateAudio || globalConfig.videoGenerateAudio,
+            videoWatermark: node.metadata?.watermark || globalConfig.videoWatermark,
+            audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice,
+            audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat,
+            audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed,
+        }, mode),
     };
     const config = buildNodeConfig(globalConfig, node, mode, requirements);
     const videoProfile = mode === "video" ? modelCapabilityConfigFor(config, config.model).video! : undefined;
@@ -229,25 +241,51 @@ function buildNodeConfig(globalConfig: AiConfig, node: CanvasNodeData, mode: Can
     const storedModel = node.metadata?.model;
     const preferredModel = storedModel && configuredModelMatchesCapability(globalConfig, storedModel, mode) ? storedModel : defaultModel && configuredModelMatchesCapability(globalConfig, defaultModel, mode) ? defaultModel : fallbackModel;
     const model = resolveCompatibleModel(globalConfig, preferredModel, mode === "image" ? { ...requirements, imageSize: node.metadata?.size || globalConfig.size || defaultConfig.size } : requirements) || preferredModel;
-    const imageProfile = mode === "image" ? modelCapabilityConfigFor(globalConfig, model).image! : undefined;
-    const normalizedImage = imageProfile ? normalizeImageValue(imageProfile, { size: node.metadata?.size || globalConfig.size || defaultConfig.size, quality: node.metadata?.quality || globalConfig.quality || defaultConfig.quality, transparentBackground: node.metadata?.transparentBackground || globalConfig.transparentBackground, count: String(node.metadata?.count || globalConfig.canvasImageCount || globalConfig.count || defaultConfig.count) }) : undefined;
+    const generationDefaults = resolveModelGenerationDefaults(
+        globalConfig,
+        model,
+        mode === "image" ? "image" : mode === "video" ? "video" : undefined,
+        mode === "image"
+            ? {
+                  size: node.metadata?.size,
+                  quality: node.metadata?.quality,
+                  transparentBackground: node.metadata?.transparentBackground,
+                  count: String(node.metadata?.count || globalConfig.canvasImageCount || globalConfig.count || defaultConfig.count),
+              }
+            : {
+                  size: node.metadata?.size,
+                  videoSeconds: node.metadata?.seconds,
+                  vquality: node.metadata?.vquality,
+                  videoGenerateAudio: node.metadata?.generateAudio,
+                  videoWatermark: node.metadata?.watermark,
+              },
+        {
+            size: globalConfig.size || defaultConfig.size,
+            quality: globalConfig.quality || defaultConfig.quality,
+            transparentBackground: globalConfig.transparentBackground || defaultConfig.transparentBackground,
+            count: String(globalConfig.canvasImageCount || globalConfig.count || defaultConfig.count),
+            videoSeconds: globalConfig.videoSeconds || defaultConfig.videoSeconds,
+            vquality: globalConfig.vquality || defaultConfig.vquality,
+            videoGenerateAudio: globalConfig.videoGenerateAudio || defaultConfig.videoGenerateAudio,
+            videoWatermark: globalConfig.videoWatermark || defaultConfig.videoWatermark,
+        },
+    );
     const videoProfile = mode === "video" ? modelCapabilityConfigFor(globalConfig, model).video! : undefined;
-    const normalizedVideo = videoProfile ? normalizeVideoValue(videoProfile, { seconds: node.metadata?.seconds || globalConfig.videoSeconds || defaultConfig.videoSeconds, ratio: node.metadata?.size || globalConfig.size || defaultConfig.size, resolution: node.metadata?.vquality || globalConfig.vquality || defaultConfig.vquality }) : undefined;
     return {
         ...globalConfig,
         model,
-        quality: normalizedImage?.quality || node.metadata?.quality || globalConfig.quality || defaultConfig.quality,
-        size: normalizedImage?.size || normalizedVideo?.ratio || node.metadata?.size || globalConfig.size || defaultConfig.size,
-        transparentBackground: normalizedImage?.transparentBackground || ((node.metadata?.transparentBackground || globalConfig.transparentBackground) === "true" ? "true" : "false"),
-        videoSeconds: normalizedVideo?.seconds || normalizeVideoDuration(node.metadata?.seconds || globalConfig.videoSeconds || defaultConfig.videoSeconds),
-        vquality: normalizedVideo?.resolution.replace(/p$/i, "") || normalizeVideoResolution(node.metadata?.vquality || globalConfig.vquality || defaultConfig.vquality),
-        videoGenerateAudio: videoProfile?.generateAudio.supported ? node.metadata?.generateAudio || globalConfig.videoGenerateAudio || String(videoProfile.generateAudio.default) : "false",
-        videoWatermark: videoProfile?.watermark.supported ? node.metadata?.watermark || globalConfig.videoWatermark || String(videoProfile.watermark.default) : "false",
+        quality: generationDefaults.quality || globalConfig.quality || defaultConfig.quality,
+        size: generationDefaults.size || globalConfig.size || defaultConfig.size,
+        transparentBackground: generationDefaults.transparentBackground || "false",
+        videoSeconds: generationDefaults.videoSeconds || normalizeVideoDuration(globalConfig.videoSeconds || defaultConfig.videoSeconds),
+        vquality: generationDefaults.vquality || normalizeVideoResolution(globalConfig.vquality || defaultConfig.vquality),
+        videoGenerateAudio: videoProfile?.generateAudio.supported ? generationDefaults.videoGenerateAudio || String(videoProfile.generateAudio.default) : "false",
+        videoWatermark: videoProfile?.watermark.supported ? generationDefaults.videoWatermark || String(videoProfile.watermark.default) : "false",
         audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice || defaultConfig.audioVoice,
         audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat || defaultConfig.audioFormat,
         audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed || defaultConfig.audioSpeed,
         audioInstructions: node.metadata?.audioInstructions || globalConfig.audioInstructions || defaultConfig.audioInstructions,
-        count: normalizedImage?.count || String(node.metadata?.count || (mode === "image" ? globalConfig.canvasImageCount || globalConfig.count : globalConfig.count) || defaultConfig.count),
+        count: generationDefaults.count || String(node.metadata?.count || (mode === "image" ? globalConfig.canvasImageCount || globalConfig.count : globalConfig.count) || defaultConfig.count),
     };
 }
 
diff --git a/web/src/components/canvas/canvas-node-prompt-panel.tsx b/web/src/components/canvas/canvas-node-prompt-panel.tsx
index 602f4e491..ad9606720 100644
--- a/web/src/components/canvas/canvas-node-prompt-panel.tsx
+++ b/web/src/components/canvas/canvas-node-prompt-panel.tsx
@@ -8,7 +8,7 @@ import { resolveCanvasGenerationModel } from "@/lib/canvas/canvas-project-genera
 import { CreditSymbol, requestCreditCost } from "@/constant/credits";
 import { canvasThemes } from "@/lib/canvas-theme";
 import { normalizeVideoDuration, normalizeVideoResolution } from "@/lib/video-generation-options";
-import { modelRequestOptions, resolveCompatibleModel, defaultImageParamsForModel, type ModelRequirements } from "@/lib/model-selection";
+import { modelRequestOptions, resolveCompatibleModel, resolveModelGenerationDefaults, defaultImageParamsForModel, type ModelRequirements } from "@/lib/model-selection";
 import { navigateToSettings } from "@/lib/settings-navigation";
 import { useThemeStore } from "@/stores/use-theme-store";
 import { useUserStore } from "@/stores/use-user-store";
@@ -77,8 +77,20 @@ export function CanvasNodePromptPanel({ node, isRunning, onPromptChange, onConfi
             characterCount: activeReferences.filter((item) => item.kind === "character").length,
         },
         videoOperation: node.metadata?.videoEditOperation,
-        videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds,
-		options: modelRequestOptions({ ...globalConfig, size: node.metadata?.size || globalConfig.size, quality: node.metadata?.quality || globalConfig.quality, count: String(node.metadata?.count || globalConfig.count), videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds, vquality: node.metadata?.vquality || globalConfig.vquality, videoGenerateAudio: node.metadata?.generateAudio || globalConfig.videoGenerateAudio, videoWatermark: node.metadata?.watermark || globalConfig.videoWatermark, audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice, audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat, audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed }, mode),
+        videoSeconds: mode === "video" ? node.metadata?.seconds || globalConfig.videoSeconds : undefined,
+        options: modelRequestOptions({
+            ...globalConfig,
+            size: node.metadata?.size || globalConfig.size,
+            quality: node.metadata?.quality || globalConfig.quality,
+            count: String(node.metadata?.count || globalConfig.count),
+            videoSeconds: node.metadata?.seconds || globalConfig.videoSeconds,
+            vquality: node.metadata?.vquality || globalConfig.vquality,
+            videoGenerateAudio: node.metadata?.generateAudio || globalConfig.videoGenerateAudio,
+            videoWatermark: node.metadata?.watermark || globalConfig.videoWatermark,
+            audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice,
+            audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat,
+            audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed,
+        }, mode),
     };
     const config = buildNodeConfig(globalConfig, node, mode, requirements);
     const generationCount = Math.max(1, Math.min(15, Math.floor(Math.abs(Number(config.count)) || 1)));
@@ -548,21 +560,50 @@ function buildNodeConfig(globalConfig: AiConfig, node: CanvasNodeData, mode: Can
     const fallbackModel = mode === "image" ? defaultConfig.imageModel : mode === "video" ? defaultConfig.videoModel : mode === "audio" ? defaultConfig.audioModel : defaultConfig.textModel;
     const preferredModel = resolveCanvasGenerationModel(globalConfig, node.metadata?.model, mode) || resolveCanvasGenerationModel(globalConfig, defaultModel, mode) || fallbackModel;
     const model = resolveCompatibleModel(globalConfig, preferredModel, mode === "image" ? { ...requirements, imageSize: node.metadata?.size || globalConfig.size || defaultConfig.size } : requirements) || preferredModel;
+    const defaults = resolveModelGenerationDefaults(
+        globalConfig,
+        model,
+        mode === "image" ? "image" : mode === "video" ? "video" : undefined,
+        mode === "image"
+            ? {
+                  size: node.metadata?.size,
+                  quality: node.metadata?.quality,
+                  transparentBackground: node.metadata?.transparentBackground,
+                  count: String(node.metadata?.count || globalConfig.canvasImageCount || globalConfig.count || defaultConfig.count),
+              }
+            : {
+                  size: node.metadata?.size,
+                  videoSeconds: node.metadata?.seconds,
+                  vquality: node.metadata?.vquality,
+                  videoGenerateAudio: node.metadata?.generateAudio,
+                  videoWatermark: node.metadata?.watermark,
+              },
+        {
+            size: globalConfig.size || defaultConfig.size,
+            quality: globalConfig.quality || defaultConfig.quality,
+            transparentBackground: globalConfig.transparentBackground || defaultConfig.transparentBackground,
+            count: String(globalConfig.canvasImageCount || globalConfig.count || defaultConfig.count),
+            videoSeconds: globalConfig.videoSeconds || defaultConfig.videoSeconds,
+            vquality: globalConfig.vquality || defaultConfig.vquality,
+            videoGenerateAudio: globalConfig.videoGenerateAudio || defaultConfig.videoGenerateAudio,
+            videoWatermark: globalConfig.videoWatermark || defaultConfig.videoWatermark,
+        },
+    );
     return {
         ...globalConfig,
         model,
-        quality: node.metadata?.quality || globalConfig.quality || defaultConfig.quality,
-        size: node.metadata?.size || globalConfig.size || defaultConfig.size,
-        transparentBackground: (node.metadata?.transparentBackground || globalConfig.transparentBackground) === "true" ? "true" : "false",
-        videoSeconds: normalizeVideoDuration(node.metadata?.seconds || globalConfig.videoSeconds || defaultConfig.videoSeconds),
-        vquality: normalizeVideoResolution(node.metadata?.vquality || globalConfig.vquality || defaultConfig.vquality),
-        videoGenerateAudio: node.metadata?.generateAudio || globalConfig.videoGenerateAudio || defaultConfig.videoGenerateAudio,
-        videoWatermark: node.metadata?.watermark || globalConfig.videoWatermark || defaultConfig.videoWatermark,
+        quality: defaults.quality || globalConfig.quality || defaultConfig.quality,
+        size: defaults.size || globalConfig.size || defaultConfig.size,
+        transparentBackground: defaults.transparentBackground || "false",
+        videoSeconds: defaults.videoSeconds || normalizeVideoDuration(globalConfig.videoSeconds || defaultConfig.videoSeconds),
+        vquality: defaults.vquality || normalizeVideoResolution(globalConfig.vquality || defaultConfig.vquality),
+        videoGenerateAudio: defaults.videoGenerateAudio || globalConfig.videoGenerateAudio || defaultConfig.videoGenerateAudio,
+        videoWatermark: defaults.videoWatermark || globalConfig.videoWatermark || defaultConfig.videoWatermark,
         audioVoice: node.metadata?.audioVoice || globalConfig.audioVoice || defaultConfig.audioVoice,
         audioFormat: node.metadata?.audioFormat || globalConfig.audioFormat || defaultConfig.audioFormat,
         audioSpeed: node.metadata?.audioSpeed || globalConfig.audioSpeed || defaultConfig.audioSpeed,
         audioInstructions: node.metadata?.audioInstructions || globalConfig.audioInstructions || defaultConfig.audioInstructions,
-        count: String(node.metadata?.count || (mode === "image" ? globalConfig.canvasImageCount || globalConfig.count : globalConfig.count) || defaultConfig.count),
+        count: defaults.count || String(node.metadata?.count || (mode === "image" ? globalConfig.canvasImageCount || globalConfig.count : globalConfig.count) || defaultConfig.count),
     };
 }
 
diff --git a/web/src/lib/canvas/canvas-project-generation.ts b/web/src/lib/canvas/canvas-project-generation.ts
index f4db0c10b..cc95561c1 100644
--- a/web/src/lib/canvas/canvas-project-generation.ts
+++ b/web/src/lib/canvas/canvas-project-generation.ts
@@ -7,8 +7,8 @@ import { resourceIdFromStorageKey } from "@/services/api/resources";
 import { NODE_DEFAULT_SIZE } from "@/constant/canvas";
 import { normalizeVideoDuration, normalizeVideoResolution } from "@/lib/video-generation-options";
 import { isSeedanceVideoConfig } from "@/lib/seedance-video";
-import { modelCapabilityConfigFor, normalizeImageValue } from "@/lib/model-capabilities";
-import { modelRequestOptions, resolveCompatibleModel, resolveVideoOperation, type ModelRequirements } from "@/lib/model-selection";
+import { modelCapabilityConfigFor } from "@/lib/model-capabilities";
+import { modelRequestOptions, resolveCompatibleModel, resolveModelGenerationDefaults, resolveVideoOperation, type ModelRequirements } from "@/lib/model-selection";
 import { imageMetadata } from "@/lib/canvas/canvas-generation-task-sync";
 import { ensureMediaNodeMinimumSize } from "@/lib/canvas/canvas-node-size";
 import type { CanvasNodeGenerationMode } from "@/components/canvas/canvas-node-prompt-panel";
@@ -399,22 +399,46 @@ export function buildGenerationConfig(config: AiConfig, node: CanvasNodeData | u
 		? { ...requirements, options: { ...liveOptions, ...(requirements.options || {}) } }
 		: { capability: mode, options: liveOptions };
     const model = resolveCompatibleModel(config, preferredModel, imageSize ? { ...baseRequirements, imageSize } : baseRequirements) || preferredModel;
-    const imageProfile = mode === "image" ? modelCapabilityConfigFor(config, model).image! : undefined;
-    const normalizedImage = imageProfile
-        ? normalizeImageValue(imageProfile, {
-              quality: requestedConfig.quality,
-              size: requestedConfig.size,
-              transparentBackground: requestedConfig.transparentBackground,
-              count: requestedConfig.count,
-          })
-        : undefined;
+    const generationDefaults = resolveModelGenerationDefaults(
+        config,
+        model,
+        mode === "image" ? "image" : mode === "video" ? "video" : undefined,
+        mode === "image"
+            ? {
+                  size: node?.metadata?.size,
+                  quality: node?.metadata?.quality,
+                  transparentBackground: node?.metadata?.transparentBackground,
+                  count: requestedConfig.count,
+              }
+            : {
+                  size: node?.metadata?.size,
+                  videoSeconds: node?.metadata?.seconds,
+                  vquality: node?.metadata?.vquality,
+                  videoGenerateAudio: node?.metadata?.generateAudio,
+                  videoWatermark: node?.metadata?.watermark,
+              },
+        {
+            size: requestedConfig.size,
+            quality: requestedConfig.quality,
+            transparentBackground: requestedConfig.transparentBackground,
+            count: requestedConfig.count,
+            videoSeconds: requestedConfig.videoSeconds,
+            vquality: requestedConfig.vquality,
+            videoGenerateAudio: requestedConfig.videoGenerateAudio,
+            videoWatermark: requestedConfig.videoWatermark,
+        },
+    );
     return {
         ...requestedConfig,
         model,
-        quality: normalizedImage?.quality || requestedConfig.quality,
-        size: normalizedImage?.size || requestedConfig.size,
-        transparentBackground: normalizedImage?.transparentBackground || (requestedConfig.transparentBackground === "true" ? "true" : "false"),
-        count: normalizedImage?.count || requestedConfig.count,
+        quality: generationDefaults.quality || requestedConfig.quality,
+        size: generationDefaults.size || requestedConfig.size,
+        transparentBackground: generationDefaults.transparentBackground || (requestedConfig.transparentBackground === "true" ? "true" : "false"),
+        videoSeconds: generationDefaults.videoSeconds || requestedConfig.videoSeconds,
+        vquality: generationDefaults.vquality || requestedConfig.vquality,
+        videoGenerateAudio: generationDefaults.videoGenerateAudio || requestedConfig.videoGenerateAudio,
+        videoWatermark: generationDefaults.videoWatermark || requestedConfig.videoWatermark,
+        count: generationDefaults.count || requestedConfig.count,
     };
 }
 
diff --git a/web/src/lib/model-selection.ts b/web/src/lib/model-selection.ts
index dbc6c1aec..b0d0a7c91 100644
--- a/web/src/lib/model-selection.ts
+++ b/web/src/lib/model-selection.ts
@@ -1,4 +1,4 @@
-import { defaultImageCapabilityConfig, modelCapabilityConfigFor, STANDARD_IMAGE_SIZE_VALUES, videoDurationAllowed, type ImageCapabilityConfig } from "@/lib/model-capabilities";
+import { defaultImageCapabilityConfig, modelCapabilityConfigFor, normalizeImageValue, normalizeVideoValue, STANDARD_IMAGE_SIZE_VALUES, videoDurationAllowed, type ImageCapabilityConfig } from "@/lib/model-capabilities";
 import { modelOptionName, resolveModelChannel, selectableModelsByCapability, type AiConfig, type ModelCapability } from "@/stores/use-config-store";
 
 export type ModelInputSummary = {
@@ -129,7 +129,13 @@ function logicalModelCompatibilityError(spec: NonNullable<NonNullable<AiConfig["
     }
     const operation = requirements.capability === "video" && input ? resolveVideoOperation(input, requirements.videoOperation) : requirements.videoOperation;
     if (operation && spec.operations?.length && !spec.operations.includes(operation)) return "不支持当前生成模式";
-    const options = { ...requirements.options, ...(requirements.videoSeconds ? { videoSeconds: requirements.videoSeconds } : {}), ...(requirements.imageSize ? { size: requirements.imageSize } : {}) };
+    // 图片创作状态也会携带全局默认视频时长；这个字段只对视频模型有意义，
+    // 不能把它拼进图片逻辑模型的能力匹配，否则图片模型会被误判为“不支持当前时长”。
+    const options = {
+        ...requirements.options,
+        ...(requirements.capability === "video" && requirements.videoSeconds ? { videoSeconds: requirements.videoSeconds } : {}),
+        ...(requirements.capability === "image" && requirements.imageSize ? { size: requirements.imageSize } : {}),
+    };
     for (const [name, value] of Object.entries(options)) {
         if (value === undefined || value === null || value === "") continue;
         const constraint = spec.options?.[name];
@@ -223,6 +229,51 @@ export function defaultImageParamsForModel(config: AiConfig, model: string): Pic
     };
 }
 
+
+export type ModelGenerationDefaults = Pick<AiConfig, "size" | "quality" | "transparentBackground" | "count" | "videoSeconds" | "vquality" | "videoGenerateAudio" | "videoWatermark">;
+
+export function resolveModelGenerationDefaults(
+    config: AiConfig,
+    model: string,
+    capability: "image" | "video" | undefined,
+    explicit: Partial<ModelGenerationDefaults> = {},
+    fallback: Partial<ModelGenerationDefaults> = {},
+): Partial<ModelGenerationDefaults> {
+    if (!capability) return {};
+    const channel = resolveModelChannel(config, model);
+    const cost = channel.modelCosts?.find((item) => item.model === modelOptionName(model));
+    const isManagedModel = Boolean(cost?.logicalModelId || cost?.logicalCapabilitySpec);
+    const source = (key: keyof ModelGenerationDefaults) => explicit[key] ?? (isManagedModel ? undefined : fallback[key]);
+    const profile = modelCapabilityConfigFor(config, model);
+
+    if (capability === "image" && profile.image) {
+        const normalized = normalizeImageValue(profile.image, {
+            size: source("size"),
+            quality: source("quality"),
+            count: source("count"),
+            transparentBackground: source("transparentBackground"),
+        });
+        return normalized;
+    }
+
+    if (capability === "video" && profile.video) {
+        const normalized = normalizeVideoValue(profile.video, {
+            seconds: source("videoSeconds"),
+            ratio: source("size"),
+            resolution: source("vquality"),
+        });
+        return {
+            videoSeconds: normalized.seconds,
+            size: normalized.ratio,
+            vquality: normalized.resolution.replace(/p$/i, ""),
+            videoGenerateAudio: source("videoGenerateAudio") ?? String(profile.video.generateAudio.default),
+            videoWatermark: source("videoWatermark") ?? String(profile.video.watermark.default),
+        };
+    }
+
+    return {};
+}
+
 export function maxModelInputCapacity(config: AiConfig, capability: "image" | "video", kind: "image" | "video" | "audio") {
     const options = selectableModelsByCapability(config, capability);
     if (!options.length) return null;
diff --git a/web/src/pages/create/index.tsx b/web/src/pages/create/index.tsx
index 27dddce97..f5b2eb579 100644
--- a/web/src/pages/create/index.tsx
+++ b/web/src/pages/create/index.tsx
@@ -175,7 +175,7 @@ export default function CreatePage() {
             audioCount: attachments.filter((attachment) => creationAttachmentKind(attachment) === "audio").length,
             characterCount: 0,
         },
-        videoSeconds: seconds,
+        videoSeconds: mode === "video" ? seconds : undefined,
         imageSize: mode === "image" ? ratio : undefined,
 		options: mode === "image"
 			? { size: ratio, quality, count: Number(count), transparentBackground: config.transparentBackground === "true" }
@@ -497,7 +497,20 @@ export default function CreatePage() {
         const controller = new AbortController();
         const requestLifecycle = beginGenerationConsumer(controller.signal);
         abortRef.current = controller;
-        const requestConfig = { ...config, model: selectedModel, imageModel: selectedModel, videoModel: selectedModel, textModel: selectedModel, size: ratio, videoSeconds: seconds, quality, vquality: videoQuality, count };
+        const normalizedImage = mode === "image" ? normalizeImageValue(imageProfile, { size: ratio, quality, count }) : undefined;
+        const normalizedVideo = mode === "video" ? normalizeVideoValue(videoProfile, { seconds, ratio, resolution: videoQuality }) : undefined;
+        const requestConfig = {
+            ...config,
+            model: selectedModel,
+            imageModel: selectedModel,
+            videoModel: selectedModel,
+            textModel: selectedModel,
+            ...(mode === "image"
+                ? { size: normalizedImage?.size || ratio, quality: normalizedImage?.quality || quality, count: normalizedImage?.count || count, videoSeconds: config.videoSeconds }
+                : mode === "video"
+                  ? { size: normalizedVideo?.ratio || ratio, videoSeconds: normalizedVideo?.seconds || seconds, vquality: (normalizedVideo?.resolution || videoQuality).replace(/p$/i, "") }
+                  : {}),
+        };
         try {
             if (mode === "text") {
 				if (logicalModelIDForConfig(requestConfig)) {
diff --git a/web/test/canvas-model-policy.test.ts b/web/test/canvas-model-policy.test.ts
index 8faa96862..6bf208a27 100644
--- a/web/test/canvas-model-policy.test.ts
+++ b/web/test/canvas-model-policy.test.ts
@@ -3,7 +3,7 @@ import { describe, expect, test } from "bun:test";
 import { canvasConnectionError } from "../src/lib/canvas/canvas-connection-policy";
 import { assertCanvasImageReferenceLimit, buildGenerationConfig, canvasImageReferenceLimitError, resolveCanvasGenerationModel } from "../src/lib/canvas/canvas-project-generation";
 import { defaultModelCapabilityConfig } from "../src/lib/model-capabilities";
-import { groupModelsByDisplayName, modelCompatibilityError, modelGroupReferenceLimits, resolveCompatibleModel } from "../src/lib/model-selection";
+import { groupModelsByDisplayName, modelCompatibilityError, modelGroupReferenceLimits, resolveCompatibleModel, resolveModelGenerationDefaults } from "../src/lib/model-selection";
 import { defaultConfig, type AiConfig, type ModelChannel } from "../src/stores/use-config-store";
 import { CanvasNodeType, type CanvasConnection, type CanvasNodeData } from "../src/types/canvas";
 
@@ -53,6 +53,82 @@ function node(id: string, type: CanvasNodeType, generationMode?: "image" | "vide
 }
 
 describe("逻辑模型选择", () => {
+    test("图片逻辑模型忽略全局视频时长，不应显示为不支持当前时长", () => {
+        const model = "platform::gpt-image-2";
+        const channel: ModelChannel = {
+            id: "platform",
+            name: "平台模型",
+            baseUrl: "/api",
+            apiKey: "system",
+            apiFormat: "openai",
+            scope: "system",
+            enabled: true,
+            models: ["gpt-image-2"],
+            modelCosts: [{
+                model: "gpt-image-2",
+                displayName: "GPT Image 2",
+                capability: "image",
+                billingMode: "fixed_request",
+                unitPriceMicrocredits: 1,
+                logicalCapabilitySpec: {
+                    version: 1,
+                    capability: "image",
+                    options: { size: { values: ["1:1", "3:2"] } },
+                },
+            }],
+        };
+        const config: AiConfig = {
+            ...defaultConfig,
+            channels: [channel],
+            models: [model],
+            imageModels: [model],
+            imageModel: model,
+            model,
+        };
+
+        expect(modelCompatibilityError(config, model, {
+            capability: "image",
+            input: { textCount: 1, imageCount: 0, videoCount: 0, audioCount: 0, characterCount: 0 },
+            videoSeconds: "6",
+            imageSize: "3:2",
+            options: { size: "3:2" },
+        })).toBe("");
+    });
+
+    test("后台图片默认比例优先于旧的全局 1:1 配置", () => {
+        const model = "platform::managed-image";
+        const profile = defaultModelCapabilityConfig(undefined, "managed-image");
+        profile.image!.size = { parameter: "size", values: ["1:1", "16:9"], default: "16:9", allowCustom: false };
+        const config: AiConfig = {
+            ...defaultConfig,
+            size: "1:1",
+            channels: [{ id: "platform", name: "平台模型", baseUrl: "/api", apiKey: "system", apiFormat: "openai", scope: "system", models: ["managed-image"], modelCosts: [{ model: "managed-image", capability: "image", billingMode: "fixed_request", unitPriceMicrocredits: 1, logicalModelId: "managed-image", logicalCapabilitySpec: { version: 1, capability: "image" }, capabilityConfig: profile }] }],
+            models: [model],
+            imageModels: [model],
+            imageModel: model,
+            model,
+        };
+
+        expect(resolveModelGenerationDefaults(config, model, "image", {}, { size: "1:1" }).size).toBe("16:9");
+    });
+
+    test("后台视频默认时长优先于旧的全局 6 秒配置", () => {
+        const model = "platform::managed-video";
+        const profile = defaultModelCapabilityConfig(undefined, "managed-video");
+        profile.video!.duration = { selection: "enum", values: [6, 15], default: 15 };
+        const config: AiConfig = {
+            ...defaultConfig,
+            videoSeconds: "6",
+            channels: [{ id: "platform", name: "平台模型", baseUrl: "/api", apiKey: "system", apiFormat: "openai", scope: "system", models: ["managed-video"], modelCosts: [{ model: "managed-video", capability: "video", billingMode: "per_second", unitPriceMicrocredits: 1, logicalModelId: "managed-video", logicalCapabilitySpec: { version: 1, capability: "video" }, capabilityConfig: profile }] }],
+            models: [model],
+            videoModels: [model],
+            videoModel: model,
+            model,
+        };
+
+        expect(resolveModelGenerationDefaults(config, model, "video", {}, { videoSeconds: "6" }).videoSeconds).toBe("15");
+    });
+
     test("后台标注的视频模型不因内部标识缺少视频关键词而回退", () => {
         const config = policyConfig();
         const selectedModel = config.videoModels[0]!;