705 lines
24 KiB
TypeScript
705 lines
24 KiB
TypeScript
import { logInfo as _ulogInfo, logWarn as _ulogWarn, logError as _ulogError } from '@/lib/logging/core'
|
||
/**
|
||
* 分镜生成多阶段处理器
|
||
* 将分镜生成拆分为3个独立阶段，每阶段控制在Vercel时间限制内
|
||
*
|
||
* 每个阶段失败后重试一次
|
||
*/
|
||
|
||
import { executeAiTextStep } from '@/lib/ai-runtime'
|
||
import { logAIAnalysis } from '@/lib/logging/semantic'
|
||
import { buildCharactersIntroduction } from '@/lib/constants'
|
||
import type { Locale } from '@/i18n/routing'
|
||
import { getPromptTemplate, PROMPT_IDS } from '@/lib/prompt-i18n'
|
||
import {
|
||
buildPromptAssetContext,
|
||
compileAssetPromptFragments,
|
||
} from '@/lib/assets/services/asset-prompt-context'
|
||
|
||
// 阶段类型
|
||
export type StoryboardPhase = 1 | '2-cinematography' | '2-acting' | 3
|
||
|
||
type JsonRecord = Record<string, unknown>
|
||
|
||
export type ClipCharacterRef = string | { name?: string | null }
|
||
|
||
type CharacterAppearance = {
|
||
changeReason?: string | null
|
||
descriptions?: string | null
|
||
selectedIndex?: number | null
|
||
description?: string | null
|
||
}
|
||
|
||
export type CharacterAsset = {
|
||
name: string
|
||
appearances?: CharacterAppearance[]
|
||
}
|
||
|
||
export type LocationAsset = {
|
||
name: string
|
||
images?: Array<{
|
||
isSelected?: boolean
|
||
description?: string | null
|
||
}>
|
||
}
|
||
|
||
export type PropAsset = {
|
||
name: string
|
||
summary?: string | null
|
||
}
|
||
|
||
type ClipAsset = {
|
||
id?: string
|
||
start?: string | number | null
|
||
end?: string | number | null
|
||
startText?: string | null
|
||
endText?: string | null
|
||
characters?: string | null
|
||
location?: string | null
|
||
props?: string | null
|
||
content?: string | null
|
||
screenplay?: string | null
|
||
}
|
||
|
||
type SessionAsset = {
|
||
user: {
|
||
id: string
|
||
name: string
|
||
}
|
||
}
|
||
|
||
type NovelPromotionAssetData = {
|
||
analysisModel: string
|
||
characters: CharacterAsset[]
|
||
locations: LocationAsset[]
|
||
props?: PropAsset[]
|
||
}
|
||
|
||
export type StoryboardPanel = JsonRecord & {
|
||
panel_number?: number
|
||
description?: string
|
||
location?: string
|
||
source_text?: string
|
||
characters?: unknown
|
||
props?: unknown
|
||
srt_range?: unknown[]
|
||
scene_type?: string
|
||
shot_type?: string
|
||
camera_move?: string
|
||
video_prompt?: string
|
||
duration?: number
|
||
photographyPlan?: JsonRecord
|
||
actingNotes?: unknown
|
||
}
|
||
|
||
export type PhotographyRule = JsonRecord & {
|
||
panel_number?: number
|
||
composition?: string
|
||
lighting?: string
|
||
color_palette?: string
|
||
atmosphere?: string
|
||
technical_notes?: string
|
||
}
|
||
|
||
export type ActingDirection = JsonRecord & {
|
||
panel_number?: number
|
||
characters?: unknown
|
||
}
|
||
|
||
function isJsonRecord(value: unknown): value is JsonRecord {
|
||
return typeof value === 'object' && value !== null
|
||
}
|
||
|
||
function parseClipCharacters(raw: string | null | undefined): ClipCharacterRef[] {
|
||
if (!raw) return []
|
||
try {
|
||
const parsed = JSON.parse(raw)
|
||
return Array.isArray(parsed) ? (parsed as ClipCharacterRef[]) : []
|
||
} catch {
|
||
return []
|
||
}
|
||
}
|
||
|
||
function parseScreenplay(raw: string | null | undefined): unknown {
|
||
if (!raw) return null
|
||
try {
|
||
return JSON.parse(raw)
|
||
} catch {
|
||
return null
|
||
}
|
||
}
|
||
|
||
// 阶段进度映射
|
||
export const PHASE_PROGRESS: Record<string, { start: number, end: number, label: string, labelKey: string }> = {
|
||
'1': { start: 10, end: 40, label: '规划分镜', labelKey: 'phases.planning' },
|
||
'2-cinematography': { start: 40, end: 55, label: '设计摄影', labelKey: 'phases.cinematography' },
|
||
'2-acting': { start: 55, end: 70, label: '设计演技', labelKey: 'phases.acting' },
|
||
'3': { start: 70, end: 100, label: '补充细节', labelKey: 'phases.detail' }
|
||
}
|
||
|
||
// 中间结果存储接口
|
||
export interface PhaseResult {
|
||
clipId: string
|
||
planPanels?: StoryboardPanel[]
|
||
photographyRules?: PhotographyRule[]
|
||
actingDirections?: ActingDirection[] // 演技指导数据
|
||
finalPanels?: StoryboardPanel[]
|
||
}
|
||
|
||
// ========== 辅助函数 ==========
|
||
|
||
// 根据 clip.characters 筛选角色形象列表
|
||
export function getFilteredAppearanceList(characters: CharacterAsset[], clipCharacters: ClipCharacterRef[]): string {
|
||
return compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters,
|
||
locations: [],
|
||
props: [],
|
||
clipCharacters,
|
||
clipLocation: null,
|
||
clipProps: [],
|
||
})).appearanceListText
|
||
}
|
||
|
||
// 根据 clip.characters 筛选角色完整描述
|
||
export function getFilteredFullDescription(characters: CharacterAsset[], clipCharacters: ClipCharacterRef[]): string {
|
||
return compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters,
|
||
locations: [],
|
||
props: [],
|
||
clipCharacters,
|
||
clipLocation: null,
|
||
clipProps: [],
|
||
})).fullDescriptionText
|
||
}
|
||
|
||
// 根据 clip.location 筛选场景描述
|
||
export function getFilteredLocationsDescription(
|
||
locations: LocationAsset[],
|
||
clipLocation: string | null,
|
||
locale: Locale = 'zh',
|
||
): string {
|
||
return compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters: [],
|
||
locations,
|
||
props: [],
|
||
clipCharacters: [],
|
||
clipLocation,
|
||
clipProps: [],
|
||
locale,
|
||
})).locationDescriptionText
|
||
}
|
||
|
||
function parseClipProps(raw: string | null | undefined): string[] {
|
||
if (!raw) return []
|
||
try {
|
||
const parsed = JSON.parse(raw)
|
||
if (!Array.isArray(parsed)) return []
|
||
return parsed.map((item) => (typeof item === 'string' ? item.trim() : '')).filter(Boolean)
|
||
} catch {
|
||
return []
|
||
}
|
||
}
|
||
|
||
// 格式化Clip标识（支持SRT模式和Agent模式）
|
||
export function formatClipId(clip: ClipAsset): string {
|
||
// SRT 模式
|
||
if (clip.start !== undefined && clip.start !== null) {
|
||
return `${clip.start}-${clip.end}`
|
||
}
|
||
// Agent 模式
|
||
if (clip.startText && clip.endText) {
|
||
return `${clip.startText.substring(0, 10)}...~...${clip.endText.substring(0, 10)}`
|
||
}
|
||
// 回退
|
||
return clip.id?.substring(0, 8) || 'unknown'
|
||
}
|
||
|
||
// 解析JSON响应
|
||
function parseJsonResponse<T extends JsonRecord>(responseText: string, clipId: string, phase: number): T[] {
|
||
let jsonText = responseText.trim()
|
||
jsonText = jsonText.replace(/^```json\s*/i, '').replace(/^```\s*/, '').replace(/\s*```$/, '')
|
||
|
||
const firstBracket = jsonText.indexOf('[')
|
||
const lastBracket = jsonText.lastIndexOf(']')
|
||
|
||
if (firstBracket === -1 || lastBracket === -1 || lastBracket <= firstBracket) {
|
||
throw new Error(`Phase ${phase}: JSON格式错误 clip ${clipId}`)
|
||
}
|
||
|
||
jsonText = jsonText.substring(firstBracket, lastBracket + 1)
|
||
const result = JSON.parse(jsonText)
|
||
|
||
if (!Array.isArray(result) || result.length === 0) {
|
||
throw new Error(`Phase ${phase}: 返回空数据 clip ${clipId}`)
|
||
}
|
||
|
||
const normalized = result.filter(isJsonRecord) as T[]
|
||
if (normalized.length === 0) {
|
||
throw new Error(`Phase ${phase}: 数据结构错误 clip ${clipId}`)
|
||
}
|
||
|
||
return normalized
|
||
}
|
||
|
||
// ========== Phase 1: 基础分镜规划 ==========
|
||
export async function executePhase1(
|
||
clip: ClipAsset,
|
||
novelPromotionData: NovelPromotionAssetData,
|
||
session: SessionAsset,
|
||
projectId: string,
|
||
projectName: string,
|
||
locale: Locale,
|
||
taskId?: string
|
||
): Promise<PhaseResult> {
|
||
const clipId = formatClipId(clip)
|
||
void taskId
|
||
_ulogInfo(`[Phase 1] Clip ${clipId}: 开始基础分镜规划...`)
|
||
|
||
// 读取提示词模板
|
||
const planPromptTemplate = getPromptTemplate(PROMPT_IDS.NP_AGENT_STORYBOARD_PLAN, locale)
|
||
|
||
// 解析clip数据
|
||
const clipCharacters = parseClipCharacters(clip.characters)
|
||
const clipLocation = clip.location || null
|
||
const clipProps = parseClipProps(clip.props)
|
||
|
||
// 构建资产信息
|
||
const charactersLibName = novelPromotionData.characters.map((c) => c.name).join(', ') || '无'
|
||
const locationsLibName = novelPromotionData.locations.map((l) => l.name).join(', ') || '无'
|
||
const filteredAppearanceList = getFilteredAppearanceList(novelPromotionData.characters, clipCharacters)
|
||
const filteredFullDescription = getFilteredFullDescription(novelPromotionData.characters, clipCharacters)
|
||
const filteredPropsDescription = compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters: [],
|
||
locations: [],
|
||
props: novelPromotionData.props || [],
|
||
clipCharacters: [],
|
||
clipLocation: null,
|
||
clipProps,
|
||
})).propsDescriptionText
|
||
const charactersIntroduction = buildCharactersIntroduction(novelPromotionData.characters)
|
||
|
||
// 构建clip JSON
|
||
const clipJson = JSON.stringify({
|
||
id: clip.id,
|
||
content: clip.content,
|
||
characters: clipCharacters,
|
||
location: clipLocation,
|
||
props: clipProps,
|
||
}, null, 2)
|
||
|
||
// 读取剧本
|
||
const screenplay = parseScreenplay(clip.screenplay)
|
||
if (clip.screenplay && !screenplay) {
|
||
_ulogWarn(`[Phase 1] Clip ${clipId}: 剧本JSON解析失败`)
|
||
}
|
||
|
||
// 构建提示词
|
||
let planPrompt = planPromptTemplate
|
||
.replace('{characters_lib_name}', charactersLibName)
|
||
.replace('{locations_lib_name}', locationsLibName)
|
||
.replace('{characters_introduction}', charactersIntroduction)
|
||
.replace('{characters_appearance_list}', filteredAppearanceList)
|
||
.replace('{characters_full_description}', filteredFullDescription)
|
||
.replace('{props_description}', filteredPropsDescription)
|
||
.replace('{clip_json}', clipJson)
|
||
|
||
if (screenplay) {
|
||
planPrompt = planPrompt.replace('{clip_content}', `【剧本格式】\n${JSON.stringify(screenplay, null, 2)}`)
|
||
} else {
|
||
planPrompt = planPrompt.replace('{clip_content}', clip.content || '')
|
||
}
|
||
|
||
// 记录发送给 AI 的完整 prompt
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'STORYBOARD_PHASE1_PROMPT',
|
||
input: { 片段标识: clipId, 完整提示词: planPrompt },
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
// 调用AI（失败后重试一次）
|
||
let planPanels: StoryboardPanel[] = []
|
||
|
||
for (let attempt = 1; attempt <= 2; attempt++) {
|
||
try {
|
||
const planResult = await executeAiTextStep({
|
||
userId: session.user.id,
|
||
model: novelPromotionData.analysisModel,
|
||
messages: [{ role: 'user', content: planPrompt }],
|
||
reasoning: true,
|
||
projectId,
|
||
action: 'storyboard_phase1_plan',
|
||
meta: {
|
||
stepId: 'storyboard_phase1_plan',
|
||
stepTitle: '分镜规划',
|
||
stepIndex: 1,
|
||
stepTotal: 1,
|
||
},
|
||
})
|
||
|
||
const planResponseText = planResult.text
|
||
if (!planResponseText) {
|
||
throw new Error(`Phase 1: 无响应 clip ${clipId}`)
|
||
}
|
||
|
||
planPanels = parseJsonResponse<StoryboardPanel>(planResponseText, clipId, 1)
|
||
|
||
// 统计有效分镜数量
|
||
const validPanelCount = planPanels.filter(panel =>
|
||
panel.description && panel.description !== '无' && panel.location !== '无'
|
||
).length
|
||
|
||
_ulogInfo(`[Phase 1] Clip ${clipId}: 共 ${planPanels.length} 个分镜，其中 ${validPanelCount} 个有效分镜`)
|
||
|
||
if (validPanelCount === 0) {
|
||
throw new Error(`Phase 1: 返回全部为空分镜 clip ${clipId}`)
|
||
}
|
||
|
||
// ========== 检测 source_text 字段，缺失则重试 ==========
|
||
const missingSourceText = planPanels.some(panel => !panel.source_text)
|
||
if (missingSourceText && attempt === 1) {
|
||
_ulogWarn(`[Phase 1] Clip ${clipId}: 有分镜缺少source_text，尝试重试...`)
|
||
continue
|
||
}
|
||
|
||
// 成功，跳出循环
|
||
break
|
||
} catch (error: unknown) {
|
||
const message = error instanceof Error ? error.message : String(error)
|
||
_ulogError(`[Phase 1] Clip ${clipId}: 第${attempt}次尝试失败: ${message}`)
|
||
if (attempt === 2) throw error
|
||
}
|
||
}
|
||
|
||
// 记录第一阶段完整输出
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'STORYBOARD_PHASE1_OUTPUT',
|
||
output: {
|
||
片段标识: clipId,
|
||
总分镜数: planPanels.length,
|
||
第一阶段完整结果: planPanels
|
||
},
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
_ulogInfo(`[Phase 1] Clip ${clipId}: 生成 ${planPanels.length} 个基础分镜`)
|
||
|
||
return { clipId, planPanels }
|
||
}
|
||
|
||
// ========== Phase 2: 摄影规则生成 ==========
|
||
export async function executePhase2(
|
||
clip: ClipAsset,
|
||
planPanels: StoryboardPanel[],
|
||
novelPromotionData: NovelPromotionAssetData,
|
||
session: SessionAsset,
|
||
projectId: string,
|
||
projectName: string,
|
||
locale: Locale,
|
||
taskId?: string
|
||
): Promise<PhaseResult> {
|
||
const clipId = formatClipId(clip)
|
||
void taskId
|
||
_ulogInfo(`[Phase 2] Clip ${clipId}: 开始生成摄影规则...`)
|
||
|
||
// 读取提示词
|
||
const cinematographerPromptTemplate = getPromptTemplate(PROMPT_IDS.NP_AGENT_CINEMATOGRAPHER, locale)
|
||
|
||
// 解析clip数据
|
||
const clipCharacters = parseClipCharacters(clip.characters)
|
||
const clipLocation = clip.location || null
|
||
const clipProps = parseClipProps(clip.props)
|
||
|
||
const filteredFullDescription = getFilteredFullDescription(novelPromotionData.characters, clipCharacters)
|
||
const filteredLocationsDescription = getFilteredLocationsDescription(
|
||
novelPromotionData.locations,
|
||
clipLocation,
|
||
locale,
|
||
)
|
||
const filteredPropsDescription = compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters: [],
|
||
locations: [],
|
||
props: novelPromotionData.props || [],
|
||
clipCharacters: [],
|
||
clipLocation: null,
|
||
clipProps,
|
||
})).propsDescriptionText
|
||
|
||
// 构建提示词
|
||
const cinematographerPrompt = cinematographerPromptTemplate
|
||
.replace('{panels_json}', JSON.stringify(planPanels, null, 2))
|
||
.replace('{panel_count}', planPanels.length.toString())
|
||
.replace(/\{panel_count\}/g, planPanels.length.toString())
|
||
.replace('{locations_description}', filteredLocationsDescription)
|
||
.replace('{characters_info}', filteredFullDescription)
|
||
.replace('{props_description}', filteredPropsDescription)
|
||
|
||
let photographyRules: PhotographyRule[] = []
|
||
|
||
// 失败后重试一次
|
||
for (let attempt = 1; attempt <= 2; attempt++) {
|
||
try {
|
||
const cinematographerResult = await executeAiTextStep({
|
||
userId: session.user.id,
|
||
model: novelPromotionData.analysisModel,
|
||
messages: [{ role: 'user', content: cinematographerPrompt }],
|
||
reasoning: true,
|
||
projectId,
|
||
action: 'storyboard_phase2_cinematography',
|
||
meta: {
|
||
stepId: 'storyboard_phase2_cinematography',
|
||
stepTitle: '摄影规则',
|
||
stepIndex: 1,
|
||
stepTotal: 1,
|
||
},
|
||
})
|
||
|
||
const responseText = cinematographerResult.text
|
||
if (!responseText) {
|
||
throw new Error(`Phase 2: 无响应 clip ${clipId}`)
|
||
}
|
||
|
||
photographyRules = parseJsonResponse<PhotographyRule>(responseText, clipId, 2)
|
||
|
||
_ulogInfo(`[Phase 2] Clip ${clipId}: 成功生成 ${photographyRules.length} 个镜头的摄影规则`)
|
||
|
||
// 记录摄影方案生成结果
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'CINEMATOGRAPHER_PLAN',
|
||
output: {
|
||
片段标识: clipId,
|
||
镜头数量: planPanels.length,
|
||
摄影规则数量: photographyRules.length,
|
||
摄影规则: photographyRules
|
||
},
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
// 成功，跳出循环
|
||
break
|
||
} catch (e: unknown) {
|
||
const message = e instanceof Error ? e.message : String(e)
|
||
_ulogError(`[Phase 2] Clip ${clipId}: 第${attempt}次尝试失败: ${message}`)
|
||
if (attempt === 2) throw e
|
||
}
|
||
}
|
||
|
||
return { clipId, planPanels, photographyRules }
|
||
}
|
||
|
||
// ========== Phase 2-Acting: 演技指导生成 ==========
|
||
export async function executePhase2Acting(
|
||
clip: ClipAsset,
|
||
planPanels: StoryboardPanel[],
|
||
novelPromotionData: NovelPromotionAssetData,
|
||
session: SessionAsset,
|
||
projectId: string,
|
||
projectName: string,
|
||
locale: Locale,
|
||
taskId?: string
|
||
): Promise<PhaseResult> {
|
||
const clipId = formatClipId(clip)
|
||
void taskId
|
||
_ulogInfo(`[Phase 2-Acting] ==========================================`)
|
||
_ulogInfo(`[Phase 2-Acting] Clip ${clipId}: 开始生成演技指导...`)
|
||
_ulogInfo(`[Phase 2-Acting] planPanels 数量: ${planPanels.length}`)
|
||
_ulogInfo(`[Phase 2-Acting] projectId: ${projectId}, projectName: ${projectName}`)
|
||
|
||
// 读取提示词
|
||
const actingPromptTemplate = getPromptTemplate(PROMPT_IDS.NP_AGENT_ACTING_DIRECTION, locale)
|
||
|
||
// 解析clip数据
|
||
const clipCharacters = parseClipCharacters(clip.characters)
|
||
|
||
const filteredFullDescription = getFilteredFullDescription(novelPromotionData.characters, clipCharacters)
|
||
|
||
// 构建提示词
|
||
const actingPrompt = actingPromptTemplate
|
||
.replace('{panels_json}', JSON.stringify(planPanels, null, 2))
|
||
.replace('{panel_count}', planPanels.length.toString())
|
||
.replace(/\{panel_count\}/g, planPanels.length.toString())
|
||
.replace('{characters_info}', filteredFullDescription)
|
||
|
||
let actingDirections: ActingDirection[] = []
|
||
|
||
// 失败后重试一次
|
||
for (let attempt = 1; attempt <= 2; attempt++) {
|
||
try {
|
||
const actingResult = await executeAiTextStep({
|
||
userId: session.user.id,
|
||
model: novelPromotionData.analysisModel,
|
||
messages: [{ role: 'user', content: actingPrompt }],
|
||
reasoning: true,
|
||
projectId,
|
||
action: 'storyboard_phase2_acting',
|
||
meta: {
|
||
stepId: 'storyboard_phase2_acting',
|
||
stepTitle: '演技指导',
|
||
stepIndex: 1,
|
||
stepTotal: 1,
|
||
},
|
||
})
|
||
|
||
const responseText = actingResult.text
|
||
if (!responseText) {
|
||
throw new Error(`Phase 2-Acting: 无响应 clip ${clipId}`)
|
||
}
|
||
|
||
actingDirections = parseJsonResponse<ActingDirection>(responseText, clipId, 2)
|
||
|
||
_ulogInfo(`[Phase 2-Acting] Clip ${clipId}: 成功生成 ${actingDirections.length} 个镜头的演技指导`)
|
||
|
||
// 记录演技指导生成结果
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'ACTING_DIRECTION_PLAN',
|
||
output: {
|
||
片段标识: clipId,
|
||
镜头数量: planPanels.length,
|
||
演技指导数量: actingDirections.length,
|
||
演技指导: actingDirections
|
||
},
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
// 成功，跳出循环
|
||
break
|
||
} catch (e: unknown) {
|
||
const message = e instanceof Error ? e.message : String(e)
|
||
_ulogError(`[Phase 2-Acting] Clip ${clipId}: 第${attempt}次尝试失败: ${message}`)
|
||
if (attempt === 2) throw e
|
||
}
|
||
}
|
||
|
||
return { clipId, planPanels, actingDirections }
|
||
}
|
||
|
||
// ========== Phase 3: 补充细节和video_prompt ==========
|
||
export async function executePhase3(
|
||
clip: ClipAsset,
|
||
planPanels: StoryboardPanel[],
|
||
photographyRules: PhotographyRule[],
|
||
novelPromotionData: NovelPromotionAssetData,
|
||
session: SessionAsset,
|
||
projectId: string,
|
||
projectName: string,
|
||
locale: Locale,
|
||
taskId?: string
|
||
): Promise<PhaseResult> {
|
||
const clipId = formatClipId(clip)
|
||
void taskId
|
||
_ulogInfo(`[Phase 3] Clip ${clipId}: 开始补充镜头细节...`)
|
||
|
||
// 读取提示词
|
||
const detailPromptTemplate = getPromptTemplate(PROMPT_IDS.NP_AGENT_STORYBOARD_DETAIL, locale)
|
||
|
||
// 解析clip数据
|
||
const clipCharacters = parseClipCharacters(clip.characters)
|
||
const clipLocation = clip.location || null
|
||
const clipProps = parseClipProps(clip.props)
|
||
|
||
const filteredFullDescription = getFilteredFullDescription(novelPromotionData.characters, clipCharacters)
|
||
const filteredLocationsDescription = getFilteredLocationsDescription(
|
||
novelPromotionData.locations,
|
||
clipLocation,
|
||
locale,
|
||
)
|
||
const filteredPropsDescription = compileAssetPromptFragments(buildPromptAssetContext({
|
||
characters: [],
|
||
locations: [],
|
||
props: novelPromotionData.props || [],
|
||
clipCharacters: [],
|
||
clipLocation: null,
|
||
clipProps,
|
||
})).propsDescriptionText
|
||
|
||
// 构建提示词
|
||
const detailPrompt = detailPromptTemplate
|
||
.replace('{panels_json}', JSON.stringify(planPanels, null, 2))
|
||
.replace('{characters_age_gender}', filteredFullDescription) // 改用完整描述
|
||
.replace('{locations_description}', filteredLocationsDescription)
|
||
.replace('{props_description}', filteredPropsDescription)
|
||
|
||
// 记录发送给 AI 的完整 prompt
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'STORYBOARD_PHASE3_PROMPT',
|
||
input: { 片段标识: clipId, 完整提示词: detailPrompt },
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
void photographyRules
|
||
let finalPanels: StoryboardPanel[] = []
|
||
|
||
// 失败后重试一次
|
||
for (let attempt = 1; attempt <= 2; attempt++) {
|
||
try {
|
||
const detailResult = await executeAiTextStep({
|
||
userId: session.user.id,
|
||
model: novelPromotionData.analysisModel,
|
||
messages: [{ role: 'user', content: detailPrompt }],
|
||
reasoning: true,
|
||
projectId,
|
||
action: 'storyboard_phase3_detail',
|
||
meta: {
|
||
stepId: 'storyboard_phase3_detail',
|
||
stepTitle: '镜头细化',
|
||
stepIndex: 1,
|
||
stepTotal: 1,
|
||
},
|
||
})
|
||
|
||
const detailResponseText = detailResult.text
|
||
if (!detailResponseText) {
|
||
throw new Error(`Phase 3: 无响应 clip ${clipId}`)
|
||
}
|
||
|
||
finalPanels = parseJsonResponse<StoryboardPanel>(detailResponseText, clipId, 3)
|
||
|
||
// 记录第三阶段完整输出（过滤前）
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'STORYBOARD_PHASE3_OUTPUT',
|
||
output: {
|
||
片段标识: clipId,
|
||
总分镜数: finalPanels.length,
|
||
第三阶段完整结果_过滤前: finalPanels
|
||
},
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
// 过滤掉"无"的空分镜
|
||
const beforeFilterCount = finalPanels.length
|
||
finalPanels = finalPanels.filter((panel) =>
|
||
panel.description && panel.description !== '无' && panel.location !== '无'
|
||
)
|
||
_ulogInfo(`[Phase 3] Clip ${clipId}: 过滤空分镜 ${beforeFilterCount} -> ${finalPanels.length} 个有效分镜`)
|
||
|
||
if (finalPanels.length === 0) {
|
||
throw new Error(`Phase 3: 过滤后无有效分镜 clip ${clipId}`)
|
||
}
|
||
|
||
// 注意：photographyRules的合并已移至route.ts中，与并行执行的Phase 2结果合并
|
||
|
||
// 记录最终输出
|
||
logAIAnalysis(session.user.id, session.user.name, projectId, projectName, {
|
||
action: 'STORYBOARD_FINAL_OUTPUT',
|
||
output: {
|
||
片段标识: clipId,
|
||
过滤前总数: beforeFilterCount,
|
||
过滤后有效数: finalPanels.length,
|
||
最终有效分镜: finalPanels
|
||
},
|
||
model: novelPromotionData.analysisModel
|
||
})
|
||
|
||
// 成功，跳出循环
|
||
break
|
||
} catch (e: unknown) {
|
||
const message = e instanceof Error ? e.message : String(e)
|
||
_ulogError(`[Phase 3] Clip ${clipId}: 第${attempt}次尝试失败: ${message}`)
|
||
if (attempt === 2) throw e
|
||
}
|
||
}
|
||
|
||
_ulogInfo(`[Phase 3] Clip ${clipId}: 完成 ${finalPanels.length} 个镜头细节`)
|
||
|
||
return { clipId, finalPanels }
|
||
}
|