feat(video): support Ark full-modal references (#300)

backend/internal/service/model_capability.go

@@ -206,6 +206,7 @@ func DefaultModelCapabilityConfigForModel(protocol string, modelName string) *Mo
 		video.Duration = VideoDurationConfig{Selection: "enum", Values: []int{4, 6, 8}, Default: 6}
 		video.Resolutions = []string{"720p", "1080p"}
 	case model.ChannelInterfaceVolcengineArkVideo:
+		video.Operations = append(video.Operations, "reference_to_video", "audio_to_video")
 		video.References.MaxVideos, video.References.MaxAudios = 3, 3
 		video.References.MaxVideoBytes, video.References.MaxAudioBytes = 200*1024*1024, 15*1024*1024
 		video.References.MaxVideoDuration, video.References.MaxAudioDuration = 15, 15
@@ -601,6 +602,9 @@ func validateVideoTask(profile *VideoCapabilityConfig, input canvasGenerationInp
 	if len(input.ReferenceImages) > profile.References.MaxImages || len(input.ReferenceVideos) > profile.References.MaxVideos || len(input.ReferenceAudios) > profile.References.MaxAudios {
 		return BadAuthRequest("参考素材数量超过当前模型限制")
 	}
+	if input.Config.InterfaceType == string(model.ChannelInterfaceVolcengineArkVideo) && len(input.ReferenceAudios) > 0 && len(input.ReferenceImages) == 0 && len(input.ReferenceVideos) == 0 {
+		return BadAuthRequest("火山方舟全模态参考不支持纯音频或文本+音频，请同时添加参考图片或参考视频")
+	}
 	if len(input.ReferenceImages) < profile.References.MinImages {
 		return BadAuthRequest(fmt.Sprintf("当前视频模型至少需要 %d 张参考图", profile.References.MinImages))
 	}