| 1 | import type { DraftGenerationData, DraftGenerationQueueInfo } from '@yikart/aitoearn-queue' |
| 2 | import { HumanMessage } from '@langchain/core/messages' |
| 3 | import { ChatGoogleGenerativeAI } from '@langchain/google-genai' |
| 4 | import { BadRequestException, Injectable, Logger } from '@nestjs/common' |
| 5 | import { QueueService } from '@yikart/aitoearn-queue' |
| 6 | import { AssetsService, VideoMetadataService } from '@yikart/assets' |
| 7 | import { AccountType, AppException, FileUtil, getErrorMessage, poll, ResponseCode, retry, UserType } from '@yikart/common' |
| 8 | import { |
| 9 | AiLogChannel, |
| 10 | AiLogRepository, |
| 11 | AiLogStatus, |
| 12 | AiLogType, |
| 13 | AssetType, |
| 14 | MaterialGroupRepository, |
| 15 | MaterialRepository, |
| 16 | MaterialSource, |
| 17 | MaterialStatus, |
| 18 | MaterialType, |
| 19 | MediaRepository, |
| 20 | MediaType, |
| 21 | } from '@yikart/mongodb' |
| 22 | import { z } from 'zod' |
| 23 | import { TaskStatus } from '../../common' |
| 24 | import { config } from '../../config' |
| 25 | import { AiAvailabilityService } from '../ai-availability' |
| 26 | import { ImageService } from '../ai/image/image.service' |
| 27 | import { VideoService } from '../ai/video/video.service' |
| 28 | import { DraftGenerationMemoryService } from './draft-generation-memory.service' |
| 29 | import { DraftGenerationPlannerService, ImageTextDraftPlanResult, VideoDraftPlanResult } from './draft-generation-planner.service' |
| 30 | import { getCompatibleAccountTypes } from './draft-generation-platforms' |
| 31 | import { |
| 32 | CreateDraftFromVideoUrlDto, |
| 33 | CreateDraftGenerationV2Dto, |
| 34 | CreateImageTextDraftDto, |
| 35 | DraftGenerationMemoryContentType, |
| 36 | DraftType, |
| 37 | ImageTextDraftType, |
| 38 | ListDraftGenerationTasksDto, |
| 39 | QueryDraftGenerationTasksDto, |
| 40 | } from './draft-generation.dto' |
| 41 | import { DraftGenerationPricingVo } from './draft-generation.vo' |
| 42 | |
| 43 | export class DraftGenerationError extends Error { |
| 44 | constructor( |
| 45 | message: string, |
| 46 | cause?: unknown, |
| 47 | ) { |
| 48 | super(message, { cause }) |
| 49 | this.name = 'DraftGenerationError' |
| 50 | } |
| 51 | } |
| 52 | |
| 53 | const VideoDraftMetadataResultSchema = z.object({ |
| 54 | title: z.string().max(200).describe('TikTok video title'), |
| 55 | description: z.string().max(2200).describe('TikTok video description'), |
| 56 | topics: z.array(z.string()).max(5).describe('Hashtag topics without # prefix'), |
| 57 | }) |
| 58 | |
| 59 | interface ImageGenerationPrompt { |
| 60 | prompt: string |
| 61 | promptIndex: number |
| 62 | } |
| 63 | |
| 64 | interface ImageGenerationErrorDetail { |
| 65 | promptIndex: number |
| 66 | errorMessage: string |
| 67 | } |
| 68 | |
| 69 | type ImageGenerationProgressHandler = (imageUrl: string, imageGenerationErrors: ImageGenerationErrorDetail[]) => Promise<void> |
| 70 | |
| 71 | type DraftGenerationTaskWithQueue<T extends { id: string, status: AiLogStatus }> = T & { |
| 72 | queue?: DraftGenerationQueueInfo |
| 73 | } |
| 74 | |
| 75 | enum ImageExecutionReferenceHandling { |
| 76 | None = 'none', |
| 77 | Reference = 'reference', |
| 78 | Edit = 'edit', |
| 79 | Ignored = 'ignored', |
| 80 | } |
| 81 | |
| 82 | const OPENAI_IMAGE_DEFAULT_ASPECT_RATIO = '2:3' |
| 83 | const OPENAI_IMAGE_DEFAULT_RESOLUTION = '1K' |
| 84 | |
| 85 | function getOpenAIImageSizeProfile(imageSize: string): { maxEdge: number, maxPixels: number, squareSize?: string } { |
| 86 | switch (imageSize) { |
| 87 | case '1K': |
| 88 | return { maxEdge: 1536, maxPixels: 1536 * 1024, squareSize: '1024x1024' } |
| 89 | case '2K': |
| 90 | return { maxEdge: 2560, maxPixels: 2560 * 1440 } |
| 91 | case '4K': |
| 92 | return { maxEdge: 3840, maxPixels: 3840 * 2160 } |
| 93 | default: |
| 94 | throw new BadRequestException('imageSize must be one of 1K, 2K, or 4K') |
| 95 | } |
| 96 | } |
| 97 | const OPENAI_IMAGE_SIZE_MULTIPLE = 16 |
| 98 | const OPENAI_IMAGE_MIN_ASPECT_RATIO = 1 / 3 |
| 99 | const OPENAI_IMAGE_MAX_ASPECT_RATIO = 3 |
| 100 | |
| 101 | function getGreatestCommonDivisor(left: number, right: number): number { |
| 102 | let a = Math.abs(left) |
| 103 | let b = Math.abs(right) |
| 104 | |
| 105 | while (b !== 0) { |
| 106 | const next = a % b |
| 107 | a = b |
| 108 | b = next |
| 109 | } |
| 110 | |
| 111 | return a |
| 112 | } |
| 113 | |
| 114 | interface ResolvedImageExecution { |
| 115 | referenceHandling: ImageExecutionReferenceHandling |
| 116 | resolvedSize?: string |
| 117 | mode: 'gemini' | 'openai-generation' | 'openai-edit' |
| 118 | channel: AiLogChannel |
| 119 | } |
| 120 | |
| 121 | /** V2 视频草稿 AiLog.response */ |
| 122 | interface V2DraftResponse { |
| 123 | materialId?: string |
| 124 | mediaId?: string |
| 125 | title?: string |
| 126 | description?: string |
| 127 | topics?: string[] |
| 128 | videoUrl?: string |
| 129 | coverUrl?: string |
| 130 | plan?: VideoDraftPlanResult |
| 131 | } |
| 132 | |
| 133 | /** 图文草稿 AiLog.response */ |
| 134 | interface ImageTextDraftResponse { |
| 135 | materialId?: string |
| 136 | mediaIds?: string[] |
| 137 | title?: string |
| 138 | description?: string |
| 139 | topics?: string[] |
| 140 | coverUrl?: string |
| 141 | imageUrls?: string[] |
| 142 | requestedImageCount?: number |
| 143 | generatedImageCount?: number |
| 144 | imageGenerationErrors?: ImageGenerationErrorDetail[] |
| 145 | plan?: ImageTextDraftPlanResult |
| 146 | } |
| 147 | |
| 148 | @Injectable() |
| 149 | export class DraftGenerationService { |
| 150 | private readonly logger = new Logger(DraftGenerationService.name) |
| 151 | |
| 152 | constructor( |
| 153 | private readonly materialGroupRepository: MaterialGroupRepository, |
| 154 | private readonly materialRepository: MaterialRepository, |
| 155 | private readonly aiLogRepository: AiLogRepository, |
| 156 | private readonly queueService: QueueService, |
| 157 | private readonly videoService: VideoService, |
| 158 | private readonly assetsService: AssetsService, |
| 159 | private readonly videoMetadataService: VideoMetadataService, |
| 160 | private readonly aiAvailability: AiAvailabilityService, |
| 161 | private readonly imageService: ImageService, |
| 162 | private readonly mediaRepository: MediaRepository, |
| 163 | private readonly draftGenerationPlannerService: DraftGenerationPlannerService, |
| 164 | private readonly draftGenerationMemoryService: DraftGenerationMemoryService, |
| 165 | ) { } |
| 166 | |
| 167 | async getTask(taskId: string, userId: string, userType: UserType) { |
| 168 | const aiLog = await this.aiLogRepository.getByIdAndUserId(taskId, userId, userType) |
| 169 | if (!aiLog) { |
| 170 | throw new AppException(ResponseCode.AiLogNotFound) |
| 171 | } |
| 172 | return await this.attachQueueInfo(aiLog) |
| 173 | } |
| 174 | |
| 175 | async listTasks(dto: QueryDraftGenerationTasksDto, userId: string, userType: UserType) { |
| 176 | const tasks = await this.aiLogRepository.listByIdsAndUserId(dto.taskIds, userId, userType) |
| 177 | return await this.attachQueueInfoToTasks(tasks) |
| 178 | } |
| 179 | |
| 180 | async listTasksWithPagination(dto: ListDraftGenerationTasksDto, userId: string, userType: UserType) { |
| 181 | const [tasks, total] = await this.aiLogRepository.listWithPagination({ |
| 182 | ...dto, |
| 183 | userId, |
| 184 | userType, |
| 185 | type: AiLogType.DraftGeneration, |
| 186 | }) |
| 187 | return [await this.attachQueueInfoToTasks(tasks), total] as const |
| 188 | } |
| 189 | |
| 190 | async getStats(userId: string, userType: UserType) { |
| 191 | const generatingCount = await this.aiLogRepository.countByUserIdAndStatus( |
| 192 | userId, |
| 193 | userType, |
| 194 | AiLogType.DraftGeneration, |
| 195 | AiLogStatus.Generating, |
| 196 | ) |
| 197 | return { generatingCount } |
| 198 | } |
| 199 | |
| 200 | private async addDraftGenerationJob(data: DraftGenerationData): Promise<void> { |
| 201 | const options = data.queuePriority == null ? undefined : { priority: data.queuePriority } |
| 202 | if ( |
| 203 | data.queuePriority != null |
| 204 | && data.queuePriority >= config.ai.draftGeneration.queue.lowPriorityMinPriority |
| 205 | ) { |
| 206 | await this.queueService.addLowPriorityDraftGenerationJob(data, options) |
| 207 | return |
| 208 | } |
| 209 | |
| 210 | await this.queueService.addDraftGenerationJob(data, options) |
| 211 | } |
| 212 | |
| 213 | private async attachQueueInfo<T extends { id: string, status: AiLogStatus }>(task: T): Promise<DraftGenerationTaskWithQueue<T>> { |
| 214 | if (task.status !== AiLogStatus.Generating) { |
| 215 | return task |
| 216 | } |
| 217 | |
| 218 | const queue = await this.queueService.getDraftGenerationQueueInfo(task.id) |
| 219 | if (!queue) { |
| 220 | return task |
| 221 | } |
| 222 | |
| 223 | return { ...task, queue } |
| 224 | } |
| 225 | |
| 226 | private async attachQueueInfoToTasks<T extends { id: string, status: AiLogStatus }>(tasks: T[]): Promise<Array<DraftGenerationTaskWithQueue<T>>> { |
| 227 | return await Promise.all(tasks.map(task => this.attachQueueInfo(task))) |
| 228 | } |
| 229 | |
| 230 | private getVideoDraftModelConfig(model: string) { |
| 231 | const modelConfig = config.ai.models.video.generation.find(m => m.name === model) |
| 232 | if (!modelConfig) { |
| 233 | throw new AppException(ResponseCode.InvalidModel) |
| 234 | } |
| 235 | |
| 236 | return modelConfig |
| 237 | } |
| 238 | |
| 239 | private resolveVideoReferenceMode( |
| 240 | modelConfig: ReturnType<DraftGenerationService['getVideoDraftModelConfig']>, |
| 241 | videoUrls?: string[], |
| 242 | audioUrls?: string[], |
| 243 | ): 'multi-ref' | 'video2video' | undefined { |
| 244 | const hasVideoReference = (videoUrls?.length ?? 0) > 0 |
| 245 | const hasAudioReference = (audioUrls?.length ?? 0) > 0 |
| 246 | if (!hasVideoReference && !hasAudioReference) { |
| 247 | return undefined |
| 248 | } |
| 249 | |
| 250 | if (modelConfig.modes.includes('multi-ref')) { |
| 251 | return 'multi-ref' |
| 252 | } |
| 253 | |
| 254 | if (hasVideoReference && !hasAudioReference && modelConfig.modes.includes('video2video')) { |
| 255 | return 'video2video' |
| 256 | } |
| 257 | |
| 258 | throw new AppException(ResponseCode.InvalidModel) |
| 259 | } |
| 260 | |
| 261 | private getImageTextDraftModelConfig(model: string) { |
| 262 | const modelConfig = config.ai.draftGeneration.imageModels.find(m => m.model === model) |
| 263 | if (!modelConfig) { |
| 264 | throw new AppException(ResponseCode.InvalidModel) |
| 265 | } |
| 266 | |
| 267 | return modelConfig |
| 268 | } |
| 269 | |
| 270 | private resolveOpenAIImageSize(aspectRatio?: string, imageSize = OPENAI_IMAGE_DEFAULT_RESOLUTION): string { |
| 271 | const sizeProfile = getOpenAIImageSizeProfile(imageSize) |
| 272 | const parts = (aspectRatio ?? OPENAI_IMAGE_DEFAULT_ASPECT_RATIO).split(':') |
| 273 | if (parts.length !== 2) { |
| 274 | throw new BadRequestException('image aspectRatio must use WIDTH:HEIGHT') |
| 275 | } |
| 276 | |
| 277 | const widthRatio = Number(parts[0]) |
| 278 | const heightRatio = Number(parts[1]) |
| 279 | if (!Number.isInteger(widthRatio) || !Number.isInteger(heightRatio) || widthRatio <= 0 || heightRatio <= 0) { |
| 280 | throw new BadRequestException('image aspectRatio must use positive integer WIDTH:HEIGHT') |
| 281 | } |
| 282 | |
| 283 | const ratio = widthRatio / heightRatio |
| 284 | if (ratio < OPENAI_IMAGE_MIN_ASPECT_RATIO || ratio > OPENAI_IMAGE_MAX_ASPECT_RATIO) { |
| 285 | throw new BadRequestException('image aspectRatio must be between 1:3 and 3:1') |
| 286 | } |
| 287 | |
| 288 | if (widthRatio === heightRatio) { |
| 289 | return sizeProfile.squareSize ?? this.resolveScaledOpenAIImageSize(widthRatio, heightRatio, sizeProfile) |
| 290 | } |
| 291 | |
| 292 | return this.resolveScaledOpenAIImageSize(widthRatio, heightRatio, sizeProfile) |
| 293 | } |
| 294 | |
| 295 | private resolveScaledOpenAIImageSize( |
| 296 | widthRatio: number, |
| 297 | heightRatio: number, |
| 298 | sizeProfile: { maxEdge: number, maxPixels: number }, |
| 299 | ): string { |
| 300 | const divisor = getGreatestCommonDivisor(widthRatio, heightRatio) |
| 301 | const reducedWidth = widthRatio / divisor |
| 302 | const reducedHeight = heightRatio / divisor |
| 303 | const maxScaleByEdge = Math.floor(sizeProfile.maxEdge / Math.max(reducedWidth, reducedHeight)) |
| 304 | const maxScaleByPixels = Math.floor(Math.sqrt(sizeProfile.maxPixels / (reducedWidth * reducedHeight))) |
| 305 | const maxScale = Math.min(maxScaleByEdge, maxScaleByPixels) |
| 306 | const scale = Math.floor(maxScale / OPENAI_IMAGE_SIZE_MULTIPLE) * OPENAI_IMAGE_SIZE_MULTIPLE |
| 307 | |
| 308 | if (scale < OPENAI_IMAGE_SIZE_MULTIPLE) { |
| 309 | throw new BadRequestException('image aspectRatio cannot be converted to a supported size') |
| 310 | } |
| 311 | |
| 312 | return `${reducedWidth * scale}x${reducedHeight * scale}` |
| 313 | } |
| 314 | |
| 315 | private resolveImageExecution( |
| 316 | imageModel: string, |
| 317 | referenceImageUrls: string[], |
| 318 | aspectRatio?: string, |
| 319 | imageSize?: string, |
| 320 | ): ResolvedImageExecution { |
| 321 | this.getImageTextDraftModelConfig(imageModel) |
| 322 | |
| 323 | const geminiImageModel = config.ai.models.chat.find(model => |
| 324 | model.name === imageModel |
| 325 | && model.channel === AiLogChannel.Gemini |
| 326 | && model.outputModalities.includes('image'), |
| 327 | ) |
| 328 | const generationModel = config.ai.models.image.generation.find(model => model.name === imageModel) |
| 329 | const editModel = config.ai.models.image.edit.find(model => model.name === imageModel) |
| 330 | |
| 331 | if (geminiImageModel && (generationModel || editModel)) { |
| 332 | throw new AppException(ResponseCode.InvalidModel) |
| 333 | } |
| 334 | |
| 335 | if (geminiImageModel) { |
| 336 | return { |
| 337 | mode: 'gemini', |
| 338 | channel: geminiImageModel.channel, |
| 339 | referenceHandling: referenceImageUrls.length > 0 ? ImageExecutionReferenceHandling.Reference : ImageExecutionReferenceHandling.None, |
| 340 | } |
| 341 | } |
| 342 | |
| 343 | const resolvedSize = this.resolveOpenAIImageSize(aspectRatio, imageSize) |
| 344 | |
| 345 | if (referenceImageUrls.length > 0 && editModel) { |
| 346 | return { |
| 347 | mode: 'openai-edit', |
| 348 | channel: AiLogChannel.NewApi, |
| 349 | referenceHandling: ImageExecutionReferenceHandling.Edit, |
| 350 | resolvedSize, |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | if (generationModel) { |
| 355 | return { |
| 356 | mode: 'openai-generation', |
| 357 | channel: AiLogChannel.NewApi, |
| 358 | referenceHandling: referenceImageUrls.length > 0 ? ImageExecutionReferenceHandling.Ignored : ImageExecutionReferenceHandling.None, |
| 359 | resolvedSize, |
| 360 | } |
| 361 | } |
| 362 | |
| 363 | throw new AppException(ResponseCode.InvalidModel) |
| 364 | } |
| 365 | |
| 366 | // ==================== V2: 固定管线（无 Agent） ==================== |
| 367 | |
| 368 | /** |
| 369 | * V2: 创建草稿生成任务（投递队列时标记 version=v2） |
| 370 | * 支持选择 modelType、duration、aspectRatio |
| 371 | */ |
| 372 | async createDraftsV2(userId: string, userType: UserType, dto: CreateDraftGenerationV2Dto): Promise<string[]> { |
| 373 | const modelConfig = config.ai.models.video.generation.find(m => m.name === dto.model) |
| 374 | if (!modelConfig) { |
| 375 | throw new AppException(ResponseCode.InvalidModel) |
| 376 | } |
| 377 | |
| 378 | const resolution = dto.resolution ?? modelConfig.defaults?.resolution |
| 379 | const resolvedGroupId = await this.resolveDraftGroupId(userId, dto.groupId) |
| 380 | const draftType = dto.draftType ?? 'draft' |
| 381 | const queuePriority = modelConfig.queuePriority |
| 382 | const plannerModel = draftType === 'draft' ? dto.plannerModel ?? config.ai.draftGeneration.planner.defaultModel : undefined |
| 383 | if (plannerModel && !config.ai.models.chat.some(model => model.name === plannerModel && model.scenes?.includes('draft-generation'))) { |
| 384 | throw new AppException(ResponseCode.InvalidModel) |
| 385 | } |
| 386 | |
| 387 | const quantity = dto.quantity ?? 1 |
| 388 | const aiLogIds: string[] = [] |
| 389 | |
| 390 | for (let i = 0; i < quantity; i++) { |
| 391 | const aiLog = await this.aiLogRepository.create({ |
| 392 | userId, |
| 393 | userType, |
| 394 | type: AiLogType.DraftGeneration, |
| 395 | model: dto.model, |
| 396 | channel: modelConfig.channel as AiLogChannel, |
| 397 | status: AiLogStatus.Generating, |
| 398 | startedAt: new Date(), |
| 399 | request: { |
| 400 | groupId: resolvedGroupId, |
| 401 | version: 'v2', |
| 402 | model: dto.model, |
| 403 | duration: dto.duration, |
| 404 | resolution, |
| 405 | aspectRatio: dto.aspectRatio, |
| 406 | prompt: dto.prompt, |
| 407 | captionPrompt: dto.captionPrompt, |
| 408 | imageUrls: dto.imageUrls, |
| 409 | videoUrls: dto.videoUrls, |
| 410 | audioUrls: dto.audioUrls, |
| 411 | draftType, |
| 412 | plannerModel, |
| 413 | queuePriority, |
| 414 | }, |
| 415 | response: {}, |
| 416 | }) |
| 417 | |
| 418 | await this.addDraftGenerationJob({ |
| 419 | aiLogId: aiLog.id, |
| 420 | userId, |
| 421 | userType, |
| 422 | groupId: resolvedGroupId, |
| 423 | version: 'v2', |
| 424 | prompt: dto.prompt, |
| 425 | captionPrompt: dto.captionPrompt, |
| 426 | imageUrls: dto.imageUrls, |
| 427 | model: dto.model, |
| 428 | duration: dto.duration, |
| 429 | resolution, |
| 430 | aspectRatio: dto.aspectRatio, |
| 431 | videoUrls: dto.videoUrls, |
| 432 | audioUrls: dto.audioUrls, |
| 433 | draftType, |
| 434 | platforms: dto.platforms, |
| 435 | plannerModel, |
| 436 | disableMemory: dto.disableMemory, |
| 437 | queuePriority, |
| 438 | }) |
| 439 | |
| 440 | aiLogIds.push(aiLog.id) |
| 441 | } |
| 442 | |
| 443 | return aiLogIds |
| 444 | } |
| 445 | |
| 446 | /** |
| 447 | * V2: 固定管线执行草稿内容生成（由 Consumer 调用） |
| 448 | * |
| 449 | * 流程： |
| 450 | * 1. 规划模型生成视频 prompt + 元数据 |
| 451 | * 2. 调用视频模型生成视频 |
| 452 | * 3. 截帧生成封面 |
| 453 | * 4. 保存素材 + 更新 AiLog |
| 454 | */ |
| 455 | async generateContentV2( |
| 456 | aiLogId: string, |
| 457 | userId: string, |
| 458 | userType: UserType, |
| 459 | groupId: string, |
| 460 | options?: { |
| 461 | prompt?: string |
| 462 | captionPrompt?: string |
| 463 | imageUrls?: string[] |
| 464 | model?: string |
| 465 | duration?: number |
| 466 | resolution?: string |
| 467 | aspectRatio?: string |
| 468 | videoUrls?: string[] |
| 469 | audioUrls?: string[] |
| 470 | draftType?: DraftType |
| 471 | platforms?: string[] |
| 472 | plannerModel?: string |
| 473 | disableMemory?: boolean |
| 474 | }, |
| 475 | ): Promise<void> { |
| 476 | const startTime = Date.now() |
| 477 | const draftType = options?.draftType ?? 'draft' |
| 478 | |
| 479 | const existingAiLog = await this.aiLogRepository.getById(aiLogId) |
| 480 | const existing = (existingAiLog?.response ?? {}) as V2DraftResponse |
| 481 | |
| 482 | try { |
| 483 | const candidateImageUrls = options?.imageUrls ?? [] |
| 484 | const model = options?.model ?? 'grok-imagine-video' |
| 485 | const duration = options?.duration |
| 486 | const aspectRatio = options?.aspectRatio ?? '9:16' |
| 487 | const resolution = options?.resolution |
| 488 | |
| 489 | let plan: VideoDraftPlanResult | undefined |
| 490 | let videoPrompt = options?.prompt ?? '' |
| 491 | if (draftType === 'draft') { |
| 492 | if (existing.plan?.videoPrompt) { |
| 493 | plan = existing.plan |
| 494 | this.logger.log({ aiLogId }, 'V2: Reusing plan from previous attempt') |
| 495 | } |
| 496 | else { |
| 497 | const memoryItems = options?.disableMemory |
| 498 | ? [] |
| 499 | : (await this.draftGenerationMemoryService.getPlannerMemoryContext(userId, DraftGenerationMemoryContentType.Video)).memoryItems |
| 500 | const planned = await this.draftGenerationPlannerService.planVideo({ |
| 501 | userId, |
| 502 | contentType: DraftGenerationMemoryContentType.Video, |
| 503 | plannerModel: options?.plannerModel, |
| 504 | userPrompt: options?.prompt, |
| 505 | captionPrompt: options?.captionPrompt, |
| 506 | memoryItems, |
| 507 | referenceImageUrls: candidateImageUrls, |
| 508 | referenceVideoUrls: options?.videoUrls, |
| 509 | referenceAudioUrls: options?.audioUrls, |
| 510 | platforms: options?.platforms, |
| 511 | model, |
| 512 | duration, |
| 513 | aspectRatio, |
| 514 | }) |
| 515 | plan = planned.plan |
| 516 | |
| 517 | await this.aiLogRepository.updateById(aiLogId, { |
| 518 | $set: { 'response.plan': plan, 'request.plannerModel': planned.model }, |
| 519 | }) |
| 520 | } |
| 521 | videoPrompt = plan.videoPrompt |
| 522 | } |
| 523 | |
| 524 | // 视频 + 封面：复用已有结果或重新生成（一起保存，不可分开） |
| 525 | let videoUrl: string |
| 526 | let coverUrl: string |
| 527 | if (existing.videoUrl && existing.coverUrl) { |
| 528 | videoUrl = existing.videoUrl |
| 529 | coverUrl = existing.coverUrl |
| 530 | this.logger.log({ aiLogId }, 'V2: Reusing video+cover from previous attempt') |
| 531 | } |
| 532 | else { |
| 533 | const { videoUrl: generatedVideoUrl } = await this.generateVideo( |
| 534 | aiLogId, |
| 535 | userId, |
| 536 | userType, |
| 537 | model, |
| 538 | videoPrompt, |
| 539 | candidateImageUrls.length > 0 ? candidateImageUrls : undefined, |
| 540 | duration, |
| 541 | resolution, |
| 542 | aspectRatio, |
| 543 | options?.videoUrls, |
| 544 | options?.audioUrls, |
| 545 | ) |
| 546 | |
| 547 | const fullVideoUrl = FileUtil.buildUrl(generatedVideoUrl) |
| 548 | const thumbnailBuffer = await this.videoMetadataService.extractThumbnailFromUrl(fullVideoUrl, 2) |
| 549 | const uploadResult = await this.assetsService.uploadFromBuffer(userId, thumbnailBuffer, { |
| 550 | type: AssetType.VideoThumbnail, |
| 551 | mimeType: 'image/png', |
| 552 | filename: 'thumbnail.png', |
| 553 | }) |
| 554 | |
| 555 | videoUrl = generatedVideoUrl |
| 556 | coverUrl = uploadResult.asset.path |
| 557 | |
| 558 | await this.aiLogRepository.updateById(aiLogId, { |
| 559 | $set: { 'response.videoUrl': videoUrl, 'response.coverUrl': coverUrl }, |
| 560 | }) |
| 561 | } |
| 562 | |
| 563 | if (draftType === 'video') { |
| 564 | const mediaId = existing.mediaId ?? (await this.mediaRepository.create({ |
| 565 | userId, |
| 566 | userType, |
| 567 | materialGroupId: groupId, |
| 568 | type: MediaType.VIDEO, |
| 569 | url: videoUrl, |
| 570 | thumbUrl: coverUrl, |
| 571 | })).id |
| 572 | |
| 573 | const response: V2DraftResponse = { |
| 574 | mediaId, |
| 575 | videoUrl, |
| 576 | coverUrl, |
| 577 | } |
| 578 | await this.aiLogRepository.updateById(aiLogId, { |
| 579 | $set: { |
| 580 | status: AiLogStatus.Success, |
| 581 | model, |
| 582 | duration: Date.now() - startTime, |
| 583 | response, |
| 584 | }, |
| 585 | }) |
| 586 | |
| 587 | return |
| 588 | } |
| 589 | |
| 590 | const draftPlan = plan! |
| 591 | |
| 592 | // draft 类型：复用已有 materialId 或创建新素材 |
| 593 | const materialId = existing.materialId ?? (await this.materialRepository.create({ |
| 594 | userId, |
| 595 | userType, |
| 596 | groupId, |
| 597 | type: MaterialType.VIDEO, |
| 598 | source: MaterialSource.PlaceDraft, |
| 599 | status: MaterialStatus.SUCCESS, |
| 600 | title: draftPlan.title, |
| 601 | desc: draftPlan.description, |
| 602 | topics: draftPlan.topics, |
| 603 | coverUrl, |
| 604 | mediaList: [{ url: videoUrl, type: MediaType.VIDEO, thumbUrl: coverUrl }], |
| 605 | useCount: 0, |
| 606 | autoDeleteMedia: false, |
| 607 | model, |
| 608 | generationParams: options, |
| 609 | accountTypes: (options?.platforms as AccountType[]) ?? getCompatibleAccountTypes({ |
| 610 | type: 'video', |
| 611 | title: draftPlan.title, |
| 612 | desc: draftPlan.description, |
| 613 | topics: draftPlan.topics, |
| 614 | duration, |
| 615 | aspectRatio, |
| 616 | }), |
| 617 | })).id |
| 618 | |
| 619 | const response: V2DraftResponse = { |
| 620 | materialId, |
| 621 | title: draftPlan.title, |
| 622 | description: draftPlan.description, |
| 623 | topics: draftPlan.topics, |
| 624 | videoUrl, |
| 625 | coverUrl, |
| 626 | plan: draftPlan, |
| 627 | } |
| 628 | |
| 629 | await this.aiLogRepository.updateById(aiLogId, { |
| 630 | $set: { |
| 631 | status: AiLogStatus.Success, |
| 632 | model, |
| 633 | duration: Date.now() - startTime, |
| 634 | response, |
| 635 | }, |
| 636 | }) |
| 637 | } |
| 638 | catch (error) { |
| 639 | this.logger.error(error, `v2 generateContentV2 failed aiLogId=${aiLogId}, userId=${userId}, groupId=${groupId}`) |
| 640 | throw new DraftGenerationError( |
| 641 | getErrorMessage(error), |
| 642 | error, |
| 643 | ) |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | /** |
| 648 | * V2 辅助方法：使用 Grok 生成视频并轮询结果 |
| 649 | */ |
| 650 | private async generateVideo( |
| 651 | aiLogId: string, |
| 652 | userId: string, |
| 653 | userType: UserType, |
| 654 | model: string, |
| 655 | videoPrompt: string, |
| 656 | imageUrls?: string[], |
| 657 | duration?: number, |
| 658 | resolution?: string, |
| 659 | aspectRatio?: string, |
| 660 | videoUrls?: string[], |
| 661 | audioUrls?: string[], |
| 662 | ): Promise<{ videoUrl: string }> { |
| 663 | let task: { id: string } |
| 664 | try { |
| 665 | const modelConfig = this.getVideoDraftModelConfig(model) |
| 666 | const mode = this.resolveVideoReferenceMode(modelConfig, videoUrls, audioUrls) |
| 667 | task = await this.videoService.userVideoGeneration({ |
| 668 | userId, |
| 669 | userType, |
| 670 | model, |
| 671 | prompt: videoPrompt, |
| 672 | image: imageUrls, |
| 673 | video_url: mode === 'video2video' ? videoUrls?.[0] : undefined, |
| 674 | videos: mode === 'multi-ref' ? videoUrls : undefined, |
| 675 | audios: mode === 'multi-ref' ? audioUrls : undefined, |
| 676 | mode, |
| 677 | duration, |
| 678 | resolution, |
| 679 | ratio: aspectRatio, |
| 680 | metadata: { aspectRatio, resolution }, |
| 681 | }) |
| 682 | } |
| 683 | catch (error) { |
| 684 | const errorMessage = getErrorMessage(error) |
| 685 | await this.aiLogRepository.updateById(aiLogId, { |
| 686 | $set: { |
| 687 | status: AiLogStatus.Failed, |
| 688 | errorMessage, |
| 689 | }, |
| 690 | }) |
| 691 | throw error |
| 692 | } |
| 693 | |
| 694 | this.logger.log({ aiLogId, taskId: task.id, model, duration, aspectRatio }, 'V2: video generation started') |
| 695 | |
| 696 | const url = await poll<string>( |
| 697 | async () => { |
| 698 | const result = await this.videoService.getVideoTaskStatus({ taskId: task.id, userId, userType }) |
| 699 | if (result.status === TaskStatus.Success) { |
| 700 | return { done: true, data: result.videoUrl } |
| 701 | } |
| 702 | if (result.status === TaskStatus.Failure) { |
| 703 | return { done: true, error: result.error?.message } |
| 704 | } |
| 705 | return { done: false } |
| 706 | }, |
| 707 | { |
| 708 | maxPollingMs: 30 * 60 * 1000, |
| 709 | taskName: `Video generation (${model})`, |
| 710 | }, |
| 711 | ) |
| 712 | |
| 713 | return { videoUrl: url } |
| 714 | } |
| 715 | |
| 716 | // ==================== 图文草稿生成 ==================== |
| 717 | |
| 718 | getDraftGenerationPricing(): ReturnType<typeof DraftGenerationPricingVo.create> { |
| 719 | const imageModels = config.ai.draftGeneration.imageModels |
| 720 | |
| 721 | return DraftGenerationPricingVo.create({ imageModels, videoModels: config.ai.models.video.generation }) |
| 722 | } |
| 723 | |
| 724 | /** |
| 725 | * 创建图文草稿生成任务（同步阶段） |
| 726 | */ |
| 727 | async createImageTextDrafts(userId: string, userType: UserType, dto: CreateImageTextDraftDto): Promise<string[]> { |
| 728 | const resolvedGroupId = await this.resolveDraftGroupId(userId, dto.groupId) |
| 729 | const draftType = dto.draftType ?? 'draft' |
| 730 | const plannerModel = draftType === 'draft' ? dto.plannerModel ?? config.ai.draftGeneration.planner.defaultModel : undefined |
| 731 | if (plannerModel && !config.ai.models.chat.some(model => model.name === plannerModel && model.scenes?.includes('draft-generation'))) { |
| 732 | throw new AppException(ResponseCode.InvalidModel) |
| 733 | } |
| 734 | |
| 735 | const imageModelConfig = this.getImageTextDraftModelConfig(dto.imageModel) |
| 736 | const runtimeImageModel = imageModelConfig.runtimeModel ?? dto.imageModel |
| 737 | const queuePriority = imageModelConfig.queuePriority |
| 738 | const imageExecution = this.resolveImageExecution(dto.imageModel, dto.imageUrls ?? [], dto.aspectRatio, dto.imageSize) |
| 739 | const channel = imageExecution.channel |
| 740 | |
| 741 | const quantity = dto.quantity ?? 1 |
| 742 | const aiLogIds: string[] = [] |
| 743 | |
| 744 | for (let i = 0; i < quantity; i++) { |
| 745 | const aiLog = await this.aiLogRepository.create({ |
| 746 | userId, |
| 747 | userType, |
| 748 | type: AiLogType.DraftGeneration, |
| 749 | model: dto.imageModel, |
| 750 | channel, |
| 751 | status: AiLogStatus.Generating, |
| 752 | startedAt: new Date(), |
| 753 | request: { |
| 754 | groupId: resolvedGroupId, |
| 755 | version: 'v2-image-text', |
| 756 | imageModel: dto.imageModel, |
| 757 | imageCount: dto.imageCount, |
| 758 | imageSize: dto.imageSize, |
| 759 | aspectRatio: dto.aspectRatio, |
| 760 | prompt: dto.prompt, |
| 761 | captionPrompt: dto.captionPrompt, |
| 762 | imageUrls: dto.imageUrls, |
| 763 | draftType, |
| 764 | plannerModel, |
| 765 | runtimeImageModel, |
| 766 | queuePriority, |
| 767 | imageExecution, |
| 768 | }, |
| 769 | response: {}, |
| 770 | }) |
| 771 | |
| 772 | await this.addDraftGenerationJob({ |
| 773 | aiLogId: aiLog.id, |
| 774 | userId, |
| 775 | userType, |
| 776 | groupId: resolvedGroupId, |
| 777 | version: 'v2-image-text', |
| 778 | prompt: dto.prompt, |
| 779 | captionPrompt: dto.captionPrompt, |
| 780 | imageUrls: dto.imageUrls, |
| 781 | imageModel: dto.imageModel, |
| 782 | imageCount: dto.imageCount ?? 3, |
| 783 | imageSize: dto.imageSize, |
| 784 | aspectRatio: dto.aspectRatio, |
| 785 | imageTextDraftType: draftType, |
| 786 | platforms: dto.platforms, |
| 787 | plannerModel, |
| 788 | disableMemory: dto.disableMemory, |
| 789 | queuePriority, |
| 790 | }) |
| 791 | |
| 792 | aiLogIds.push(aiLog.id) |
| 793 | } |
| 794 | |
| 795 | return aiLogIds |
| 796 | } |
| 797 | |
| 798 | /** |
| 799 | * 图文草稿内容生成（异步阶段，由 Consumer 调用） |
| 800 | * |
| 801 | * 流程： |
| 802 | * 1. 规划模型生成 title / description / topics / imagePrompts |
| 803 | * 2. 根据 imageModel 批量生成图片（Gemini 或 GPT Image） |
| 804 | * 3. 保存素材（类型 ARTICLE）+ 更新 AiLog |
| 805 | */ |
| 806 | async generateContentImageText( |
| 807 | aiLogId: string, |
| 808 | userId: string, |
| 809 | userType: UserType, |
| 810 | groupId: string, |
| 811 | options: { |
| 812 | prompt: string |
| 813 | captionPrompt?: string |
| 814 | imageUrls?: string[] |
| 815 | imageModel: string |
| 816 | imageCount: number |
| 817 | imageSize?: string |
| 818 | aspectRatio?: string |
| 819 | draftType?: ImageTextDraftType |
| 820 | platforms?: string[] |
| 821 | plannerModel?: string |
| 822 | disableMemory?: boolean |
| 823 | }, |
| 824 | ): Promise<void> { |
| 825 | const existingAiLog = await this.aiLogRepository.getById(aiLogId) |
| 826 | const existing = (existingAiLog?.response ?? {}) as ImageTextDraftResponse |
| 827 | const startTime = Date.now() |
| 828 | const draftType = options.draftType ?? 'draft' |
| 829 | const targetImageCount = options.imageCount |
| 830 | |
| 831 | try { |
| 832 | const referenceImageUrls = options.imageUrls ?? [] |
| 833 | const imageExecution = this.resolveImageExecution(options.imageModel, referenceImageUrls, options.aspectRatio, options.imageSize) |
| 834 | this.logger.log( |
| 835 | { |
| 836 | aiLogId, |
| 837 | imageModel: options.imageModel, |
| 838 | imageCount: options.imageCount, |
| 839 | aspectRatio: options.aspectRatio, |
| 840 | refImageCount: referenceImageUrls.length, |
| 841 | draftType, |
| 842 | referenceHandling: imageExecution.referenceHandling, |
| 843 | resolvedSize: imageExecution.resolvedSize, |
| 844 | }, |
| 845 | 'ImageText: Starting generation', |
| 846 | ) |
| 847 | |
| 848 | let plan: ImageTextDraftPlanResult | undefined |
| 849 | let imagePromptTasks: ImageGenerationPrompt[] |
| 850 | if (draftType === 'image') { |
| 851 | imagePromptTasks = Array.from({ length: targetImageCount }, (_, promptIndex) => ({ |
| 852 | prompt: options.prompt, |
| 853 | promptIndex, |
| 854 | })) |
| 855 | } |
| 856 | else { |
| 857 | const existingPlan = existing.plan |
| 858 | if (existingPlan && existingPlan.imagePrompts.length >= targetImageCount) { |
| 859 | plan = existingPlan |
| 860 | this.logger.log({ aiLogId }, 'ImageText: Reusing plan from previous attempt') |
| 861 | } |
| 862 | else { |
| 863 | const memoryItems = options.disableMemory |
| 864 | ? [] |
| 865 | : (await this.draftGenerationMemoryService.getPlannerMemoryContext(userId, DraftGenerationMemoryContentType.ImageText)).memoryItems |
| 866 | const planned = await this.draftGenerationPlannerService.planImageText({ |
| 867 | userId, |
| 868 | contentType: DraftGenerationMemoryContentType.ImageText, |
| 869 | plannerModel: options.plannerModel, |
| 870 | userPrompt: options.prompt, |
| 871 | captionPrompt: options.captionPrompt, |
| 872 | memoryItems, |
| 873 | referenceImageUrls, |
| 874 | platforms: options.platforms, |
| 875 | imageModel: options.imageModel, |
| 876 | imageCount: targetImageCount, |
| 877 | imageSize: options.imageSize, |
| 878 | aspectRatio: options.aspectRatio, |
| 879 | }) |
| 880 | plan = planned.plan |
| 881 | |
| 882 | await this.aiLogRepository.updateById(aiLogId, { |
| 883 | $set: { 'response.plan': plan, 'request.plannerModel': planned.model }, |
| 884 | }) |
| 885 | |
| 886 | this.logger.log( |
| 887 | { aiLogId, title: plan.title, imagePromptsCount: plan.imagePrompts.length, plannerModel: planned.model }, |
| 888 | 'ImageText: Planning completed', |
| 889 | ) |
| 890 | } |
| 891 | |
| 892 | imagePromptTasks = plan.imagePrompts |
| 893 | .slice(0, targetImageCount) |
| 894 | .map((prompt, promptIndex) => ({ prompt, promptIndex })) |
| 895 | } |
| 896 | let generatedImageUrls = (existing.imageUrls ?? []).slice(0, targetImageCount) |
| 897 | const updateImageProgress = async ( |
| 898 | imageUrls: string[], |
| 899 | imageGenerationErrors: ImageGenerationErrorDetail[] = [], |
| 900 | ) => { |
| 901 | const response: Record<string, unknown> = { |
| 902 | 'response.imageUrls': [...imageUrls], |
| 903 | 'response.requestedImageCount': targetImageCount, |
| 904 | 'response.generatedImageCount': imageUrls.length, |
| 905 | 'response.imageGenerationErrors': [...imageGenerationErrors], |
| 906 | } |
| 907 | if (plan) { |
| 908 | response['response.plan'] = plan |
| 909 | } |
| 910 | await this.aiLogRepository.updateById(aiLogId, { |
| 911 | $set: response, |
| 912 | }) |
| 913 | } |
| 914 | |
| 915 | if (generatedImageUrls.length > 0) { |
| 916 | this.logger.log({ aiLogId, count: generatedImageUrls.length, targetImageCount }, 'ImageText: Reusing images from previous attempt') |
| 917 | } |
| 918 | |
| 919 | if (generatedImageUrls.length < targetImageCount) { |
| 920 | const missingPromptTasks = imagePromptTasks.slice(generatedImageUrls.length, targetImageCount) |
| 921 | const previousGeneratedImageCount = generatedImageUrls.length |
| 922 | const progressImageUrls = [...generatedImageUrls] |
| 923 | const { urls, imageGenerationErrors } = await this.generateImages( |
| 924 | userId, |
| 925 | userType, |
| 926 | options.imageModel, |
| 927 | missingPromptTasks, |
| 928 | imageExecution, |
| 929 | referenceImageUrls, |
| 930 | options.aspectRatio, |
| 931 | options.imageSize, |
| 932 | async (imageUrl, imageGenerationErrors) => { |
| 933 | if (progressImageUrls.length < targetImageCount) { |
| 934 | progressImageUrls.push(imageUrl) |
| 935 | } |
| 936 | await updateImageProgress(progressImageUrls, [ |
| 937 | ...(existing.imageGenerationErrors ?? []), |
| 938 | ...imageGenerationErrors, |
| 939 | ]) |
| 940 | }, |
| 941 | ) |
| 942 | generatedImageUrls = progressImageUrls.length > previousGeneratedImageCount |
| 943 | ? progressImageUrls.slice(0, targetImageCount) |
| 944 | : [...generatedImageUrls, ...urls].slice(0, targetImageCount) |
| 945 | |
| 946 | await updateImageProgress(generatedImageUrls, imageGenerationErrors) |
| 947 | |
| 948 | this.logger.log( |
| 949 | { aiLogId, generatedCount: generatedImageUrls.length, targetImageCount, errorCount: imageGenerationErrors.length }, |
| 950 | 'ImageText: Image generation completed', |
| 951 | ) |
| 952 | |
| 953 | if (generatedImageUrls.length < targetImageCount) { |
| 954 | throw new Error(`ImageText: generated ${generatedImageUrls.length}/${targetImageCount} images`) |
| 955 | } |
| 956 | } |
| 957 | |
| 958 | if (draftType === 'image') { |
| 959 | const mediaIds = (existing.mediaIds ?? []).slice(0, targetImageCount) |
| 960 | for (const imageUrl of generatedImageUrls.slice(mediaIds.length)) { |
| 961 | const media = await this.mediaRepository.create({ |
| 962 | userId, |
| 963 | userType, |
| 964 | materialGroupId: groupId, |
| 965 | type: MediaType.IMG, |
| 966 | url: imageUrl, |
| 967 | }) |
| 968 | mediaIds.push(media.id) |
| 969 | } |
| 970 | |
| 971 | const response: ImageTextDraftResponse = { |
| 972 | mediaIds, |
| 973 | imageUrls: generatedImageUrls, |
| 974 | requestedImageCount: targetImageCount, |
| 975 | generatedImageCount: generatedImageUrls.length, |
| 976 | } |
| 977 | await this.aiLogRepository.updateById(aiLogId, { |
| 978 | $set: { |
| 979 | status: AiLogStatus.Success, |
| 980 | model: options.imageModel, |
| 981 | duration: Date.now() - startTime, |
| 982 | response, |
| 983 | }, |
| 984 | }) |
| 985 | |
| 986 | return |
| 987 | } |
| 988 | |
| 989 | // draft 类型：复用已有 materialId 或创建新素材 |
| 990 | const coverUrl = generatedImageUrls[0] |
| 991 | const draftPlan = plan! |
| 992 | |
| 993 | const materialId = existing.materialId ?? (await this.materialRepository.create({ |
| 994 | userId, |
| 995 | userType, |
| 996 | groupId, |
| 997 | type: MaterialType.ARTICLE, |
| 998 | source: MaterialSource.PlaceDraft, |
| 999 | status: MaterialStatus.SUCCESS, |
| 1000 | title: draftPlan.title, |
| 1001 | desc: draftPlan.description, |
| 1002 | topics: draftPlan.topics, |
| 1003 | coverUrl, |
| 1004 | mediaList: generatedImageUrls.map(url => ({ url, type: MediaType.IMG })), |
| 1005 | useCount: 0, |
| 1006 | autoDeleteMedia: false, |
| 1007 | model: options.imageModel, |
| 1008 | generationParams: { |
| 1009 | model: options.imageModel, |
| 1010 | ...options, |
| 1011 | }, |
| 1012 | accountTypes: (options.platforms as AccountType[]) ?? getCompatibleAccountTypes({ |
| 1013 | type: 'article', |
| 1014 | title: draftPlan.title, |
| 1015 | desc: draftPlan.description, |
| 1016 | topics: draftPlan.topics, |
| 1017 | imageCount: generatedImageUrls.length, |
| 1018 | aspectRatio: options.aspectRatio, |
| 1019 | }), |
| 1020 | })).id |
| 1021 | |
| 1022 | const response: ImageTextDraftResponse = { |
| 1023 | materialId, |
| 1024 | title: draftPlan.title, |
| 1025 | description: draftPlan.description, |
| 1026 | topics: draftPlan.topics, |
| 1027 | coverUrl, |
| 1028 | imageUrls: generatedImageUrls, |
| 1029 | requestedImageCount: targetImageCount, |
| 1030 | generatedImageCount: generatedImageUrls.length, |
| 1031 | plan: draftPlan, |
| 1032 | } |
| 1033 | |
| 1034 | await this.aiLogRepository.updateById(aiLogId, { |
| 1035 | $set: { |
| 1036 | status: AiLogStatus.Success, |
| 1037 | model: options.imageModel, |
| 1038 | duration: Date.now() - startTime, |
| 1039 | response, |
| 1040 | }, |
| 1041 | }) |
| 1042 | } |
| 1043 | catch (error) { |
| 1044 | this.logger.error(error, `v2 generateContentImageText failed aiLogId=${aiLogId}, userId=${userId}, groupId=${groupId}`) |
| 1045 | throw new DraftGenerationError( |
| 1046 | getErrorMessage(error), |
| 1047 | error, |
| 1048 | ) |
| 1049 | } |
| 1050 | } |
| 1051 | |
| 1052 | /** |
| 1053 | * 根据模型类型批量生成图片 |
| 1054 | */ |
| 1055 | private async generateImages( |
| 1056 | userId: string, |
| 1057 | userType: UserType, |
| 1058 | imageModel: string, |
| 1059 | imagePrompts: ImageGenerationPrompt[], |
| 1060 | imageExecution: ResolvedImageExecution, |
| 1061 | referenceImageUrls: string[], |
| 1062 | aspectRatio?: string, |
| 1063 | imageSize?: string, |
| 1064 | onImageGenerated?: ImageGenerationProgressHandler, |
| 1065 | ): Promise<{ urls: string[], imageGenerationErrors: ImageGenerationErrorDetail[] }> { |
| 1066 | if (imageExecution.mode === 'gemini') { |
| 1067 | return this.generateImagesWithGemini(userId, userType, imageModel, imagePrompts, referenceImageUrls, aspectRatio, imageSize, onImageGenerated) |
| 1068 | } |
| 1069 | |
| 1070 | return this.generateImagesWithOpenAI(userId, userType, imageModel, imagePrompts, imageExecution, referenceImageUrls, imageSize, onImageGenerated) |
| 1071 | } |
| 1072 | |
| 1073 | /** |
| 1074 | * 使用 Gemini 模型（nb2/nb-pro）批量生成图片 |
| 1075 | */ |
| 1076 | private async generateImagesWithGemini( |
| 1077 | userId: string, |
| 1078 | userType: UserType, |
| 1079 | model: string, |
| 1080 | imagePrompts: ImageGenerationPrompt[], |
| 1081 | referenceImageUrls: string[], |
| 1082 | aspectRatio?: string, |
| 1083 | imageSize?: string, |
| 1084 | onImageGenerated?: ImageGenerationProgressHandler, |
| 1085 | ): Promise<{ urls: string[], imageGenerationErrors: ImageGenerationErrorDetail[] }> { |
| 1086 | const urls: string[] = [] |
| 1087 | const imageGenerationErrors: ImageGenerationErrorDetail[] = [] |
| 1088 | |
| 1089 | for (const { prompt, promptIndex } of imagePrompts) { |
| 1090 | this.logger.log( |
| 1091 | { model, promptIndex, promptLength: prompt.length, aspectRatio, imageSize }, |
| 1092 | 'ImageText: Generating image with Gemini', |
| 1093 | ) |
| 1094 | |
| 1095 | try { |
| 1096 | const result = await retry( |
| 1097 | () => this.imageService.userGeminiGeneration({ |
| 1098 | userId, |
| 1099 | userType, |
| 1100 | model: model as 'gemini-3.1-flash-image-preview' | 'gemini-3-pro-image-preview', |
| 1101 | prompt, |
| 1102 | imageUrls: referenceImageUrls.length > 0 ? referenceImageUrls : undefined, |
| 1103 | ...(imageSize ? { imageSize: imageSize as '1K' | '2K' | '4K' } : {}), |
| 1104 | ...(aspectRatio ? { aspectRatio: aspectRatio as '1:1' | '2:3' | '3:2' | '3:4' | '4:3' | '4:5' | '5:4' | '9:16' | '16:9' | '21:9' } : {}), |
| 1105 | }), |
| 1106 | { |
| 1107 | maxRetries: 3, |
| 1108 | delayMs: 1000, |
| 1109 | onRetry: (error, attempt) => { |
| 1110 | this.logger.warn( |
| 1111 | error, |
| 1112 | `ImageText: Gemini image generation failed, retrying promptIndex=${promptIndex}, attempt=${attempt}`, |
| 1113 | ) |
| 1114 | }, |
| 1115 | }, |
| 1116 | ) |
| 1117 | |
| 1118 | const remainingSlots = imagePrompts.length - urls.length |
| 1119 | if (remainingSlots <= 0) { |
| 1120 | break |
| 1121 | } |
| 1122 | |
| 1123 | if (result.images.length > 0) { |
| 1124 | const generatedUrls = result.images.slice(0, remainingSlots).map(image => image.url) |
| 1125 | for (const imageUrl of generatedUrls) { |
| 1126 | urls.push(imageUrl) |
| 1127 | await onImageGenerated?.(imageUrl, imageGenerationErrors) |
| 1128 | } |
| 1129 | } |
| 1130 | |
| 1131 | if (urls.length >= imagePrompts.length) { |
| 1132 | break |
| 1133 | } |
| 1134 | } |
| 1135 | catch (error) { |
| 1136 | imageGenerationErrors.push({ |
| 1137 | promptIndex, |
| 1138 | errorMessage: error instanceof Error ? error.message : String(error), |
| 1139 | }) |
| 1140 | this.logger.warn( |
| 1141 | error, |
| 1142 | `ImageText: Failed to generate image after retries, skipping promptIndex=${promptIndex}`, |
| 1143 | ) |
| 1144 | } |
| 1145 | } |
| 1146 | |
| 1147 | return { urls, imageGenerationErrors } |
| 1148 | } |
| 1149 | |
| 1150 | private async generateImagesWithOpenAI( |
| 1151 | userId: string, |
| 1152 | userType: UserType, |
| 1153 | model: string, |
| 1154 | imagePrompts: ImageGenerationPrompt[], |
| 1155 | imageExecution: ResolvedImageExecution, |
| 1156 | referenceImageUrls: string[], |
| 1157 | imageSize?: string, |
| 1158 | onImageGenerated?: ImageGenerationProgressHandler, |
| 1159 | ): Promise<{ urls: string[], imageGenerationErrors: ImageGenerationErrorDetail[] }> { |
| 1160 | const urls: string[] = [] |
| 1161 | const imageGenerationErrors: ImageGenerationErrorDetail[] = [] |
| 1162 | |
| 1163 | for (const { prompt, promptIndex } of imagePrompts) { |
| 1164 | this.logger.log( |
| 1165 | { |
| 1166 | model, |
| 1167 | promptIndex, |
| 1168 | promptLength: prompt.length, |
| 1169 | resolvedSize: imageExecution.resolvedSize, |
| 1170 | referenceHandling: imageExecution.referenceHandling, |
| 1171 | }, |
| 1172 | 'ImageText: Generating image with standard image service', |
| 1173 | ) |
| 1174 | |
| 1175 | try { |
| 1176 | const result = await retry( |
| 1177 | async () => { |
| 1178 | if (imageExecution.mode === 'openai-edit') { |
| 1179 | if (referenceImageUrls.length === 0) { |
| 1180 | throw new Error('ImageText: missing reference images for edit mode') |
| 1181 | } |
| 1182 | |
| 1183 | return this.imageService.userEdit({ |
| 1184 | userId, |
| 1185 | userType, |
| 1186 | model, |
| 1187 | image: referenceImageUrls, |
| 1188 | prompt, |
| 1189 | n: 1, |
| 1190 | size: imageExecution.resolvedSize, |
| 1191 | }) |
| 1192 | } |
| 1193 | |
| 1194 | return this.imageService.userGeneration({ |
| 1195 | userId, |
| 1196 | userType, |
| 1197 | model, |
| 1198 | prompt, |
| 1199 | n: 1, |
| 1200 | size: imageExecution.resolvedSize, |
| 1201 | }) |
| 1202 | }, |
| 1203 | { |
| 1204 | maxRetries: 3, |
| 1205 | delayMs: 1000, |
| 1206 | onRetry: (error, attempt) => { |
| 1207 | this.logger.warn( |
| 1208 | error, |
| 1209 | `ImageText: Standard image generation failed, retrying promptIndex=${promptIndex}, attempt=${attempt}`, |
| 1210 | ) |
| 1211 | }, |
| 1212 | }, |
| 1213 | ) |
| 1214 | |
| 1215 | const remainingSlots = imagePrompts.length - urls.length |
| 1216 | if (remainingSlots <= 0) { |
| 1217 | break |
| 1218 | } |
| 1219 | |
| 1220 | const generatedUrls = result.list |
| 1221 | .map(item => item.url) |
| 1222 | .filter((url): url is string => !!url) |
| 1223 | |
| 1224 | if (generatedUrls.length > 0) { |
| 1225 | for (const imageUrl of generatedUrls.slice(0, remainingSlots)) { |
| 1226 | urls.push(imageUrl) |
| 1227 | await onImageGenerated?.(imageUrl, imageGenerationErrors) |
| 1228 | } |
| 1229 | } |
| 1230 | |
| 1231 | if (urls.length >= imagePrompts.length) { |
| 1232 | break |
| 1233 | } |
| 1234 | } |
| 1235 | catch (error) { |
| 1236 | imageGenerationErrors.push({ |
| 1237 | promptIndex, |
| 1238 | errorMessage: error instanceof Error ? error.message : String(error), |
| 1239 | }) |
| 1240 | this.logger.warn( |
| 1241 | error, |
| 1242 | `ImageText: Failed to generate image after retries, skipping promptIndex=${promptIndex}`, |
| 1243 | ) |
| 1244 | } |
| 1245 | } |
| 1246 | |
| 1247 | return { urls, imageGenerationErrors } |
| 1248 | } |
| 1249 | |
| 1250 | /** |
| 1251 | * 视频 URL 生成草稿：使用 Gemini 多模态模型分析视频内容，生成文案并保存草稿 |
| 1252 | * |
| 1253 | * 流程： |
| 1254 | * 1. 解析素材组（默认草稿箱） |
| 1255 | * 2. Gemini Flash 分析视频 URL → 生成 title / description / topics |
| 1256 | * 3. 从视频截帧生成封面 |
| 1257 | * 4. 保存 Material 草稿 |
| 1258 | */ |
| 1259 | async generateDraftFromVideoUrl( |
| 1260 | userId: string, |
| 1261 | userType: UserType, |
| 1262 | dto: CreateDraftFromVideoUrlDto, |
| 1263 | ): Promise<{ materialId: string }> { |
| 1264 | const resolvedGroupId = await this.resolveDraftGroupId(userId, dto.groupId) |
| 1265 | const modelName = 'gemini-3.5-flash' |
| 1266 | const startedAt = new Date() |
| 1267 | const fullVideoUrl = FileUtil.buildUrl(dto.videoUrl) |
| 1268 | |
| 1269 | const prompt = `You are a TikTok content generation assistant. |
| 1270 | ## Task |
| 1271 | Watch the video and generate TikTok post metadata based on its content. |
| 1272 | |
| 1273 | ## Instructions |
| 1274 | - **title**: Catchy title under 30 characters, in the language that best matches the video content |
| 1275 | - **description**: Engaging description with call-to-action, under 2200 characters, in the language matching the video content |
| 1276 | - **topics**: 3-5 relevant hashtags (without # prefix) |
| 1277 | - **IMPORTANT**: Do NOT generate any content featuring children, minors, or anyone appearing under 18. |
| 1278 | Return the result as JSON.` |
| 1279 | |
| 1280 | const model = new ChatGoogleGenerativeAI({ |
| 1281 | model: modelName, |
| 1282 | apiKey: config.ai.gemini.apiKey, |
| 1283 | baseUrl: config.ai.gemini.baseUrl, |
| 1284 | temperature: 1.0, |
| 1285 | }) |
| 1286 | |
| 1287 | const messageContent: Array< |
| 1288 | { type: 'video', url: string } |
| 1289 | | { type: 'text', text: string } |
| 1290 | > = [ |
| 1291 | { type: 'video', url: fullVideoUrl }, |
| 1292 | { type: 'text', text: prompt }, |
| 1293 | ] |
| 1294 | |
| 1295 | const structuredModel = model.withStructuredOutput(z.toJSONSchema(VideoDraftMetadataResultSchema), { includeRaw: true }) |
| 1296 | const { parsed: structuredResult } = await this.aiAvailability.execute( |
| 1297 | { provider: 'gemini', operation: 'draftGeneration.generateDraftFromVideoUrl', model: modelName }, |
| 1298 | async () => structuredModel.invoke([ |
| 1299 | new HumanMessage({ content: messageContent }), |
| 1300 | ]), |
| 1301 | ) |
| 1302 | |
| 1303 | if (!structuredResult) { |
| 1304 | throw new Error('VideoUrl: No response from Gemini') |
| 1305 | } |
| 1306 | |
| 1307 | const parsed = z.safeParse(VideoDraftMetadataResultSchema, structuredResult) |
| 1308 | if (!parsed.success) { |
| 1309 | throw new Error(`VideoUrl: Invalid plan result: ${z.prettifyError(parsed.error)}`) |
| 1310 | } |
| 1311 | |
| 1312 | const plan = parsed.data |
| 1313 | const duration = Date.now() - startedAt.getTime() |
| 1314 | await this.aiLogRepository.create({ |
| 1315 | userId, |
| 1316 | userType, |
| 1317 | type: AiLogType.Agent, |
| 1318 | model: modelName, |
| 1319 | channel: AiLogChannel.Gemini, |
| 1320 | startedAt, |
| 1321 | duration, |
| 1322 | request: { videoUrl: dto.videoUrl }, |
| 1323 | response: plan, |
| 1324 | status: AiLogStatus.Success, |
| 1325 | }) |
| 1326 | |
| 1327 | this.logger.log({ plan }, 'VideoUrl: Plan generated') |
| 1328 | |
| 1329 | let coverUrl: string | undefined |
| 1330 | try { |
| 1331 | const thumbnailBuffer = await this.videoMetadataService.extractThumbnailFromUrl(fullVideoUrl, 2) |
| 1332 | const uploadResult = await this.assetsService.uploadFromBuffer(userId, thumbnailBuffer, { |
| 1333 | type: AssetType.VideoThumbnail, |
| 1334 | mimeType: 'image/png', |
| 1335 | filename: 'thumbnail.png', |
| 1336 | }) |
| 1337 | coverUrl = uploadResult.asset.path |
| 1338 | } |
| 1339 | catch (error) { |
| 1340 | this.logger.warn(error, `VideoUrl: Failed to extract video thumbnail, proceeding without cover videoUrl=${dto.videoUrl}`) |
| 1341 | } |
| 1342 | |
| 1343 | const material = await this.materialRepository.create({ |
| 1344 | userId, |
| 1345 | userType, |
| 1346 | groupId: resolvedGroupId, |
| 1347 | type: MaterialType.VIDEO, |
| 1348 | source: MaterialSource.PlaceDraft, |
| 1349 | status: MaterialStatus.SUCCESS, |
| 1350 | title: plan.title, |
| 1351 | desc: plan.description, |
| 1352 | topics: plan.topics, |
| 1353 | coverUrl, |
| 1354 | mediaList: [{ url: dto.videoUrl, type: MediaType.VIDEO, thumbUrl: coverUrl }], |
| 1355 | useCount: 0, |
| 1356 | autoDeleteMedia: false, |
| 1357 | model: modelName, |
| 1358 | generationParams: { |
| 1359 | model: modelName, |
| 1360 | videoUrl: dto.videoUrl, |
| 1361 | }, |
| 1362 | accountTypes: dto.platforms ?? getCompatibleAccountTypes({ |
| 1363 | type: 'video', |
| 1364 | title: plan.title, |
| 1365 | desc: plan.description, |
| 1366 | topics: plan.topics, |
| 1367 | }), |
| 1368 | }) |
| 1369 | |
| 1370 | return { materialId: material.id } |
| 1371 | } |
| 1372 | |
| 1373 | private async resolveDraftGroupId(userId: string, groupId?: string): Promise<string> { |
| 1374 | if (groupId) { |
| 1375 | const group = await this.materialGroupRepository.getInfo(groupId) |
| 1376 | if (!group || group.userId !== userId) { |
| 1377 | throw new AppException(ResponseCode.MaterialGroupNotFound) |
| 1378 | } |
| 1379 | return group.id |
| 1380 | } |
| 1381 | |
| 1382 | const defaultGroup = await this.materialGroupRepository.getDefaultGroup(userId) |
| 1383 | if (!defaultGroup) { |
| 1384 | throw new AppException(ResponseCode.MaterialGroupNotFound) |
| 1385 | } |
| 1386 | |
| 1387 | return defaultGroup.id |
| 1388 | } |
| 1389 | } |
| 1390 |