diff --git a/backend/internal/app/model_capability.go b/backend/internal/app/model_capability.go
index 5eb25973b..a9b0612ab 100644
--- a/backend/internal/app/model_capability.go
+++ b/backend/internal/app/model_capability.go
@@ -105,6 +105,11 @@ type VideoReferenceConfig struct {
 	MaxAudioDuration int   `json:"maxAudioDurationSeconds"`
 }
 
+// DefaultVideoPromptMaxChars 是普通视频模型提示词字符数的默认上限。
+// 视频提示词由输入框文本、连线内容和技能上下文合成，远长于用户手输内容；
+// 默认值过小会把画布工作流正常可用的提示词拦在本地。管理员仍可按模型覆盖。
+const DefaultVideoPromptMaxChars = 8000
+
 type VideoDurationConfig struct {
 	Selection string `json:"selection"`
 	Min       int    `json:"min,omitempty"`
@@ -206,7 +211,7 @@ func DefaultModelCapabilityConfigForModel(protocol string, modelName string) *Mo
 	streaming := true
 	text := &TextCapabilityConfig{Streaming: &streaming, References: TextReferenceConfig{PromptMaxChars: 32000}}
 	video := &VideoCapabilityConfig{
-		References:        VideoReferenceConfig{PromptMaxChars: 1000, MinImages: 0, MaxImages: 9, MaxImageBytes: 30 * 1024 * 1024, MaxVideos: 0, MaxVideoBytes: 0, MaxVideoDuration: 0, MaxAudios: 0, MaxAudioBytes: 0, MaxAudioDuration: 0},
+		References:        VideoReferenceConfig{PromptMaxChars: DefaultVideoPromptMaxChars, MinImages: 0, MaxImages: 9, MaxImageBytes: 30 * 1024 * 1024, MaxVideos: 0, MaxVideoBytes: 0, MaxVideoDuration: 0, MaxAudios: 0, MaxAudioBytes: 0, MaxAudioDuration: 0},
 		Duration:          VideoDurationConfig{Selection: "range", Min: 1, Max: 15, Step: 1, Default: 6},
 		Ratios:            []string{"16:9", "9:16", "1:1", "4:3", "3:4", "21:9"},
 		DefaultRatio:      "16:9",
diff --git a/backend/internal/app/model_capability_test.go b/backend/internal/app/model_capability_test.go
index 3ee40d7bf..2028dc406 100644
--- a/backend/internal/app/model_capability_test.go
+++ b/backend/internal/app/model_capability_test.go
@@ -52,6 +52,34 @@ func TestValidateVideoTaskRejectsPromptAboveCapabilityLimit(t *testing.T) {
 	}
 }
 
+// 视频提示词由输入框文本、连线内容和技能上下文合成，默认上限必须留出足够余量，
+// 否则画布工作流会连同线内容一起被拦在本地。这里锁定默认值与边界行为。
+func TestDefaultVideoPromptMaxCharsAllowsComposedCanvasPrompt(t *testing.T) {
+	profile := DefaultModelCapabilityConfigForModel("seedance-videos-compatible", "sd-2.5").Video
+	if profile.References.PromptMaxChars != 8000 {
+		t.Fatalf("视频默认提示词上限 = %d, want 8000", profile.References.PromptMaxChars)
+	}
+
+	// 画布实际合成出的提示词量级（输入框 + 连线 + 技能上下文）必须通过。
+	composed := canvasGenerationInput{Mode: "video", Prompt: strings.Repeat("镜", 2399), Config: providerConfig{Model: "sd-2.5", VideoSeconds: "6", Size: "16:9", VQuality: "720p"}}
+	if err := validateVideoTask(profile, composed); err != nil {
+		t.Fatalf("合成提示词 2399 字符被拒绝: %v", err)
+	}
+	// 恰好等于上限仍然放行，与前端 `actualChars <= maxChars` 的判定保持一致。
+	atLimit := composed
+	atLimit.Prompt = strings.Repeat("镜", 8000)
+	if err := validateVideoTask(profile, atLimit); err != nil {
+		t.Fatalf("提示词 8000 字符被拒绝: %v", err)
+	}
+	// 超出上限必须明确失败，并保持“不自动截断”的语义。
+	overLimit := composed
+	overLimit.Prompt = strings.Repeat("镜", 8001)
+	err := validateVideoTask(profile, overLimit)
+	if err == nil || !strings.Contains(err.Error(), "最多 8000 个字符") || !strings.Contains(err.Error(), "8001 个字符") {
+		t.Fatalf("validateVideoTask() error = %v", err)
+	}
+}
+
 func TestValidateTaskCapabilityRejectsWorkflowPromptAboveConfiguredLimit(t *testing.T) {
 	profile := DefaultModelCapabilityConfigForModel("runninghub-workflow-video", "workflow-video")
 	profile.Video.References.PromptMaxChars = 10_000
diff --git a/web/src/lib/model-capabilities.ts b/web/src/lib/model-capabilities.ts
index 60958bba3..3fa86df56 100644
--- a/web/src/lib/model-capabilities.ts
+++ b/web/src/lib/model-capabilities.ts
@@ -24,6 +24,15 @@ export type TextCapabilityConfig = {
 
 export type ImageSizeParameter = "none" | "size" | "aspect_ratio";
 
+/**
+ * 普通视频模型提示词字符数的默认上限。
+ *
+ * 视频提示词由「输入框文本 + 连线内容 + 技能上下文」合成，远长于用户手输内容，
+ * 因此不能沿用偏小的默认值把画布工作流拦在本地。必须与后端
+ * `DefaultVideoPromptMaxChars` 保持一致，否则前端放行后端拒绝（或反之）。
+ */
+export const DEFAULT_VIDEO_PROMPT_MAX_CHARS = 8000;
+
 export type ImageCapabilityConfig = {
     references: {
         promptMaxChars: number;
@@ -262,7 +271,7 @@ export function defaultModelCapabilityConfig(protocol?: ModelProtocol, model = "
     };
     const video: VideoCapabilityConfig = {
         references: {
-            promptMaxChars: 1000,
+            promptMaxChars: DEFAULT_VIDEO_PROMPT_MAX_CHARS,
             minImages: 0,
             maxImages: 9,
             maxImageBytes: 30 * 1024 * 1024,