feat(*): 画布创作 - 支持文本多模态生成并优化媒体节点、连接线与视频预览

Fixes #59

backend/internal/service/provider.go

@@ -593,7 +593,7 @@ func runChatCompletionsTextTask(ctx context.Context, input canvasGenerationInput
 
 func textResponseInput(input canvasGenerationInput) (interface{}, error) {
 	systemPrompt := strings.TrimSpace(input.Config.SystemPrompt)
-	if len(input.ReferenceImages) == 0 {
+	if len(input.ReferenceImages) == 0 && len(input.ReferenceVideos) == 0 {
 		return withSystemPrompt(input.Config, input.Prompt), nil
 	}
 	messages := make([]map[string]interface{}, 0, 2)
@@ -617,11 +617,18 @@ func textResponseContent(input canvasGenerationInput) ([]map[string]interface{},
 		}
 		content = append(content, map[string]interface{}{"type": "input_image", "image_url": url})
 	}
+	for _, video := range input.ReferenceVideos {
+		url, err := openAIVideoInputURL(video)
+		if err != nil {
+			return nil, err
+		}
+		content = append(content, map[string]interface{}{"type": "input_video", "video_url": url})
+	}
 	return content, nil
 }
 
 func textChatContent(input canvasGenerationInput) (interface{}, error) {
-	if len(input.ReferenceImages) == 0 {
+	if len(input.ReferenceImages) == 0 && len(input.ReferenceVideos) == 0 {
 		return input.Prompt, nil
 	}
 	content := []map[string]interface{}{{"type": "text", "text": input.Prompt}}
@@ -632,6 +639,13 @@ func textChatContent(input canvasGenerationInput) (interface{}, error) {
 		}
 		content = append(content, map[string]interface{}{"type": "image_url", "image_url": map[string]interface{}{"url": url}})
 	}
+	for _, video := range input.ReferenceVideos {
+		url, err := openAIVideoInputURL(video)
+		if err != nil {
+			return nil, err
+		}
+		content = append(content, map[string]interface{}{"type": "video_url", "video_url": map[string]interface{}{"url": url}})
+	}
 	return content, nil
 }
 
@@ -653,6 +667,24 @@ func openAIImageInputURL(media providerMedia) (string, error) {
 	return "", errors.New("OpenAI 文本多模态参考图片需要公网 URL 或 base64 data URL")
 }
 
+func openAIVideoInputURL(media providerMedia) (string, error) {
+	value := strings.TrimSpace(media.DataURL)
+	if strings.HasPrefix(value, "data:video/") {
+		return value, nil
+	}
+	if strings.HasPrefix(value, "data:") {
+		return "", errors.New("参考视频 MIME 类型无效，请重新读取或上传视频")
+	}
+	value = strings.TrimSpace(media.URL)
+	if strings.HasPrefix(value, "data:video/") || isPublicMediaURL(value) {
+		return value, nil
+	}
+	if strings.HasPrefix(value, "data:") {
+		return "", errors.New("参考视频 MIME 类型无效，请重新读取或上传视频")
+	}
+	return "", errors.New("文本多模态参考视频需要公网 URL 或 base64 data URL")
+}
+
 func shouldFallbackTextToChat(err error) bool {
 	var httpErr providerHTTPError
 	if !errors.As(err, &httpErr) {

backend/internal/service/provider_test.go

@@ -117,13 +117,16 @@ func TestDoBinaryRejectsOversizedProviderResponse(t *testing.T) {
 	}
 }
 
-func TestTextResponseInputIncludesReferenceImages(t *testing.T) {
+func TestTextResponseInputIncludesReferenceMedia(t *testing.T) {
 	input := canvasGenerationInput{
 		Prompt: "describe this image",
 		Config: providerConfig{SystemPrompt: "answer in Chinese"},
 		ReferenceImages: []providerMedia{
 			{ID: "image-1", Name: "image.png", Type: "image/png", DataURL: testReferenceImageDataURL},
 		},
+		ReferenceVideos: []providerMedia{
+			{ID: "video-1", Name: "video.mp4", Type: "video/mp4", URL: "https://example.com/reference.mp4"},
+		},
 	}
 
 	value, err := textResponseInput(input)
@@ -144,8 +147,8 @@ func TestTextResponseInputIncludesReferenceImages(t *testing.T) {
 	if !ok {
 		t.Fatalf("user content = %T, want []map[string]interface{}", messages[1]["content"])
 	}
-	if len(content) != 2 {
-		t.Fatalf("len(content) = %d, want 2", len(content))
+	if len(content) != 3 {
+		t.Fatalf("len(content) = %d, want 3", len(content))
 	}
 	if content[0]["type"] != "input_text" || content[0]["text"] != "describe this image" {
 		t.Fatalf("text content = %#v", content[0])
@@ -153,14 +156,20 @@ func TestTextResponseInputIncludesReferenceImages(t *testing.T) {
 	if content[1]["type"] != "input_image" || content[1]["image_url"] != testReferenceImageDataURL {
 		t.Fatalf("image content = %#v", content[1])
 	}
+	if content[2]["type"] != "input_video" || content[2]["video_url"] != "https://example.com/reference.mp4" {
+		t.Fatalf("video content = %#v", content[2])
+	}
 }
 
-func TestTextChatContentIncludesReferenceImages(t *testing.T) {
+func TestTextChatContentIncludesReferenceMedia(t *testing.T) {
 	input := canvasGenerationInput{
 		Prompt: "describe this image",
 		ReferenceImages: []providerMedia{
 			{ID: "image-1", Name: "image.png", Type: "image/png", DataURL: testReferenceImageDataURL},
 		},
+		ReferenceVideos: []providerMedia{
+			{ID: "video-1", Name: "video.mp4", Type: "video/mp4", URL: "https://example.com/reference.mp4"},
+		},
 	}
 
 	value, err := textChatContent(input)
@@ -171,8 +180,8 @@ func TestTextChatContentIncludesReferenceImages(t *testing.T) {
 	if !ok {
 		t.Fatalf("textChatContent() = %T, want []map[string]interface{}", value)
 	}
-	if len(content) != 2 {
-		t.Fatalf("len(content) = %d, want 2", len(content))
+	if len(content) != 3 {
+		t.Fatalf("len(content) = %d, want 3", len(content))
 	}
 	if content[0]["type"] != "text" || content[0]["text"] != "describe this image" {
 		t.Fatalf("text content = %#v", content[0])
@@ -184,6 +193,13 @@ func TestTextChatContentIncludesReferenceImages(t *testing.T) {
 	if content[1]["type"] != "image_url" || imageURL["url"] != testReferenceImageDataURL {
 		t.Fatalf("image content = %#v", content[1])
 	}
+	videoURL, ok := content[2]["video_url"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("video_url = %T, want map[string]interface{}", content[2]["video_url"])
+	}
+	if content[2]["type"] != "video_url" || videoURL["url"] != "https://example.com/reference.mp4" {
+		t.Fatalf("video content = %#v", content[2])
+	}
 }
 
 func TestTextReferenceImageRejectsInternalAssetURL(t *testing.T) {