""" AVE 视频工厂 — 统一生产入口 按策略路由到不同管线, 提供统一的输入接口和进度报告。 策略: 口播 文案 → TTS → Pexels素材 → BGM → 字幕 → 数字人片头片尾 卡点 BGM → 节拍检测 → 素材 → 拍点切换 → xfade 数字人 照片+音频 → OmniHuman 或 照片+视频 → DreamActor 用法: python main.py video-factory --strategy 口播 --script script.yaml python main.py video-factory --strategy 卡点 --bgm bgm.wav --search 关键词 python main.py video-factory --strategy 数字人 --image photo.jpg --text 文案 """ import sys, os; sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) import json import os import time from pathlib import Path from lib.config import load_config from lib.cost_tracker import get_tracker, print_history from lib.logger import get_logger logger = get_logger("factory") # ── 策略配置 ───────────────────────────────────────────── STRATEGIES = { "口播": { "desc": "文案→TTS→素材→BGM→字幕→数字人片头片尾", "default_output": "/tmp/ave_oral.mp4", }, "卡点": { "desc": "BGM→节拍检测→素材→拍点切换→xfade过渡", "default_output": "/tmp/ave_beat.mp4", }, "数字人": { "desc": "OmniHuman(对口型) / DreamActor(动作模仿)", "default_output": "/tmp/ave_digital_human.mp4", }, "口播+卡点": { "desc": "人声锚点+BGM能量变速→帧锁定拼接", "default_output": "/tmp/ave_hybrid.mp4", }, "故事": { "desc": "剧本→场景分解→Kling批量生成→角色一致性拼接", "default_output": "/tmp/ave_story.mp4", }, } def run_oral(script: str, output: str, clips_per_segment: int, bgm: str | None, duck: bool, anchor: bool, subtitles: bool, mix_engine: str = "ffmpeg") -> str: """口播策略: 调用 main.py generate""" from main import main as cli_main # 直接调用 generate 的逻辑 cfg = load_config() import yaml print(f"[视频工厂] 口播策略: {script}") # Dashboard 埋点: 创建 production 记录 try: from lib.dashboard import init_db, log_production, log_step, complete_production init_db() dash_pid = log_production( strategy="口播", script_path=script, script_name=os.path.basename(script) if script else "", ) if dash_pid > 0: from lib.cost_tracker import set_current_production_id set_current_production_id(dash_pid) except Exception: dash_pid = -1 # 1. 读取脚本 if script.endswith((".yaml", ".yml")): with open(script, encoding="utf-8") as f: script_data = yaml.safe_load(f) else: from director_parser.parser import parse_script script = parse_script(script) with open(script, encoding="utf-8") as f: script_data = yaml.safe_load(f) segments = script_data.get("segments", []) total_duration = sum(s.get("duration_sec", 10) for s in segments) print(f" 脚本: {len(segments)} 段, 预计 {total_duration}s") # 2. TTS print(" [1/6] 合成人声...") from voice_synthesizer.aliyun import synthesize_with_timestamps ak = cfg.get("aliyun", {}).get("api_key", "") vid = cfg.get("aliyun", {}).get("voice_id", "") text_all = "\n".join(s["text"] for s in segments) voice_path, word_ts = synthesize_with_timestamps( text_all, "/tmp/ave_factory_voice.wav", api_key=ak, voice_id=vid, ) if dash_pid > 0: log_step(dash_pid, "tts", "completed", detail=f"{len(text_all)}字") tracker = get_tracker() tracker.log("CosyVoice", chars=len(text_all), note=f"口播TTS {len(segments)}段") # 3. 素材搜索 print(" [2/6] 搜索素材...") from material_producer.pexels.search import search_videos mat_cfg = cfg.get("pexels", {}) all_clips = [] for seg in segments: keyword = seg.get("material", {}).get("search", "") if not keyword: continue clips = search_videos(keyword, count=clips_per_segment, api_key=mat_cfg.get("api_key", ""), orientation="portrait") all_clips.extend((c["path"], c["duration"]) for c in clips) tracker.log("Pexels", note=f"搜索: {keyword}") clip_paths = [c[0] for c in all_clips] print(f" 素材: {len(clip_paths)} 个") if dash_pid > 0: log_step(dash_pid, "search_material", "completed", detail=f"搜索到 {len(clip_paths)} 个素材") # 3.5 自动 BGM 选择 (基于情绪分析) if not bgm: print(" [▸] 自动选择 BGM (情绪分析)...") try: from audio_line.layer2_analysis.emotion_analyzer import EmotionAnalyzer from music_selector import BGMSelector emotions_analyzer = EmotionAnalyzer() emotion_result = emotions_analyzer.analyze(voice_path) # 情绪 → BGM 风格映射 EMOTION_STYLE_MAP = { "energetic": "燃系", "happy": "轻快", "calm": "舒缓", "sad": "治愈", "aggressive": "燃系", "neutral": "治愈", } target_style = EMOTION_STYLE_MAP.get(emotion_result.mood, "治愈") print(f" 检测到情绪: {emotion_result.mood} ({emotion_result.description})") print(f" 匹配 BGM 风格: {target_style}") selector = BGMSelector() candidates = selector.get_by_style(target_style) + selector.get_trending(limit=3) # 去重 seen = set() unique_candidates = [] for c in candidates: if c["id"] not in seen: seen.add(c["id"]) unique_candidates.append(c) if unique_candidates: chosen = unique_candidates[0] bgm_path_auto = chosen.get("file", "") if os.path.exists(bgm_path_auto): bgm = bgm_path_auto print(f" ✅ 自动选中 BGM: {chosen['name']} - {chosen['artist']} ({chosen['style']})") else: print(f" ⚠ BGM 文件不存在，跳过: {chosen['name']} ({bgm_path_auto})") else: print(" ⚠ 未找到匹配的 BGM") except ImportError as e: print(f" ⚠ 情绪分析模块不可用: {e} (跳过自动选BGM)") except Exception as e: print(f" ⚠ 自动选 BGM 失败: {e}") # 4. BGM + 避让 + 混音 print(" [3/6] 混音...") from composer.ffmpeg import mix_audio, duck_bgm final_audio = voice_path if bgm and os.path.exists(bgm): if duck: bgm_path = duck_bgm(voice_path, bgm, "/tmp/ave_factory_ducked.wav", word_timestamps=word_ts) bgm_for_mix = bgm_path print(f" BGM避让: ON") else: bgm_for_mix = bgm print(f" BGM: ON (固定音量)") # 混音引擎选择: ffmpeg (默认) | audio_line (MixModule) if mix_engine == "audio_line": try: from audio_line.layer3_creation.module_d_mix import MixModule mix_mod = MixModule() mix_result = mix_mod.mix( stems={"voice": voice_path, "bgm": bgm_for_mix}, output_path="/tmp/ave_factory_master.wav", stem_levels={"voice": 0.0, "bgm": -6.0}, ) final_audio = mix_result.master_path print(f" AudioLine 混音: {mix_result.loudness} LUFS, " f"峰值={mix_result.peak_dB} dBFS") except Exception as e: print(f" ⚠ AudioLine 混音失败: {e} (回退 ffmpeg)") final_audio = mix_audio(voice_path, bgm_for_mix, bgm_volume=1.0) else: final_audio = mix_audio(voice_path, bgm_for_mix, bgm_volume=1.0) # 5. 字幕 print(" [4/6] 生成字幕...") from composer.ffmpeg import create_subtitles subtitles_path = None if subtitles and segments and word_ts: full_text_from_ts = "".join(w["text"] for w in word_ts) char_pos = [] acc = 0 for w in word_ts: char_pos.append(acc) acc += len(w["text"]) cursor = 0 for seg in segments: seg_text = seg["text"] try: pos = full_text_from_ts.index(seg_text, cursor) char_end = pos + len(seg_text) seg_words = [word_ts[i] for i in range(len(word_ts)) if char_pos[i] >= pos and char_pos[i] + len(word_ts[i]["text"]) <= char_end] except ValueError: seg_words = [] if seg_words: seg["start_sec"] = seg_words[0]["begin_time"] / 1000.0 seg["end_sec"] = seg_words[-1]["end_time"] / 1000.0 cursor = pos + len(seg_text) subtitles_path = "/tmp/ave_factory_subtitles.ass" create_subtitles(segments, subtitles_path) print(f" 字幕: {len(segments)} 段") # 6. 合成视频 print(" [5/6] 合成视频...") from composer.ffmpeg import compose_video, _get_media_duration from composer.ffmpeg import segment_render, concat_segments audio_duration = _get_media_duration(final_audio) if audio_duration > 180: import tempfile, shutil seg_dir = tempfile.mkdtemp(prefix="ave_fac_") seg_files = segment_render(clip_paths, final_audio, output_dir=seg_dir, subtitles_path=subtitles_path) if len(seg_files) > 1: concat_segments(seg_files, output) shutil.rmtree(seg_dir, ignore_errors=True) else: shutil.move(seg_files[0], output) shutil.rmtree(seg_dir, ignore_errors=True) else: if anchor: print(" 锚点画面切换...") from anchor_extractor.extractor import get_silence_periods from composer.ffmpeg import compose_with_anchors silences = get_silence_periods(final_audio, min_silence_sec=0.1) compose_with_anchors(clip_paths, final_audio, output, silence_periods=silences, total_duration=audio_duration, subtitles_path=subtitles_path) else: compose_video(clip_paths, final_audio, output, resolution="1080x1920", subtitles_path=subtitles_path) if dash_pid > 0: log_step(dash_pid, "compose", "completed", cost=0, detail=f"{audio_duration:.0f}s 视频") # 7. LipSync 后处理 (如果脚本指定了角色且有定妆照) print(" [6/7] 检查 LipSync 后处理...") character_refs = script_data.get("meta", {}).get("character_refs", []) lipsync_needed = False for seg in segments: if seg.get("character_ref") and any( cr.get("lip_sync") for cr in character_refs if cr.get("name") == seg["character_ref"] ): lipsync_needed = True break if lipsync_needed: print(" 角色 LipSync 已启用, 尝试后处理...") try: cfg = load_config() fal_key = cfg.get("fal", {}).get("api_key", "") if fal_key: from composer.lipsync import lipsync_audio_to_video lipsync_result = lipsync_audio_to_video( video_path=output, audio_path=voice_path, output_path=output.replace(".mp4", "_lipsync.mp4"), ) # 替换原视频 import shutil shutil.move(lipsync_result, output) print(f" ✅ LipSync 完成") if dash_pid > 0: log_step(dash_pid, "lipsync", "completed", detail="口型对齐") else: print(" ⏭️ 未配置 fal.ai API Key, 跳过 LipSync") except Exception as e: print(f" ⚠️ LipSync 失败: {e} (不影响主线流程)") # 8. 检查数字人片头片尾 (可选) print(" [7/7] 检查数字人片头片尾...") avatar_cfg = script_data.get("meta", {}).get("avatar", {}) opening_text = avatar_cfg.get("opening_text", "") closing_text = avatar_cfg.get("closing_text", "") avatar_image = avatar_cfg.get("image", "") if avatar_image and opening_text: print(" 需要生成片头 + 片尾数字人 (需手动调用 digital-human)") print(f" 片头: {opening_text}") print(f" 片尾: {closing_text}") sz_mb = os.path.getsize(output) / 1024 / 1024 print(f"\n✅ 口播完成: {output} ({sz_mb:.0f}MB, {audio_duration:.0f}s)") # Dashboard: 完成 production if dash_pid > 0: try: complete_production( dash_pid, status="completed", output_path=output, duration_sec=audio_duration, total_cost=get_tracker()._total_cost(), ) from lib.cost_tracker import set_current_production_id set_current_production_id(0) except Exception: pass return output def run_beat(bgm: str, output: str, search: str, group_size: int, clips: list[str], texts: list[str], target_bpm: float = 0.0) -> str: """卡点策略""" print(f"[视频工厂] 卡点策略: {bgm}") from composer.beat_sync import compose_beat_sync cfg = load_config() pexels_key = cfg.get("pexels", {}).get("api_key", "") result = compose_beat_sync( bgm_path=bgm, output_path=output, material_clips=clips or None, group_size=group_size, texts=texts or None, pexels_api_key=pexels_key, pexels_search=search, target_bpm=target_bpm, ) return result def run_digital_human(image: str, text: str, output: str, resolution: str, mode: str, video: str | None, character: str = "") -> str: """数字人策略""" from voice_synthesizer.aliyun import synthesize cfg = load_config() ak = cfg.get("aliyun", {}).get("api_key", "") vid = cfg.get("aliyun", {}).get("voice_id", "") # 角色注册 → 音色选择 char_voice_style = "" if character: try: from character_registry import CharacterRegistry registry = CharacterRegistry() try: char_obj = registry.get_character(character) char_voice_style = char_obj.voice_style print(f" 角色 '{character}' 音色: {char_voice_style}") except KeyError: print(f" ⚠ 未找到角色 '{character}', 使用默认音色") except Exception as e: print(f" ⚠ 角色注册读取失败: {e}") print(f"[视频工厂] 数字人策略: mode={mode}") # 角色 voice_style → voice_id 映射 (占位, 可根据实际音色库扩展) VOICE_STYLE_MAP = { "甜美": "wan2.2-s2v", # 示例映射 "成熟": "cosyvoice-v3.5-plus", "知性": "cosyvoice-v3.5-plus", "默认": vid, } if char_voice_style and char_voice_style in VOICE_STYLE_MAP: vid = VOICE_STYLE_MAP[char_voice_style] print(f" → 映射音色: {char_voice_style} → {vid}") if mode == "对口型": from material_producer.wan2_2.wan2_2 import generate_digital_human audio_path = "/tmp/ave_factory_dh_voice.wav" print(" [1/2] 合成人声...") synthesize(text, audio_path, api_key=ak, voice_id=vid) print(" [2/2] 生成数字人 (OmniHuman)...") result = generate_digital_human( image, audio_path, ak, output_path=output, resolution=resolution, text=text, ) elif mode == "动作模仿": if not video: raise ValueError("动作模仿模式需要 --video 参数") # DreamActor M1 调用 from lib.ghvideo_upload import upload_and_get_url import httpx, json, hmac, hashlib, datetime from datetime import timezone ak_v = cfg.get("volcano", {}).get("access_key_id", ak) sk_v = cfg.get("volcano", {}).get("secret_access_key", "") host='visual.volcengineapi.com';region='cn-north-1';service='cv' def sr(p,b): t=datetime.datetime.now(timezone.utc);cd=t.strftime('%Y%m%dT%H%M%SZ');ds=t.strftime('%Y%m%d') q='&'.join(f'{k}={p[k]}'for k in sorted(p));ph=hashlib.sha256(b.encode()).hexdigest() kd=hmac.new(sk_v.encode(),ds.encode(),hashlib.sha256).digest();kr=hmac.new(kd,region.encode(),hashlib.sha256).digest() ks=hmac.new(kr,service.encode(),hashlib.sha256).digest();skey=hmac.new(ks,b'request',hashlib.sha256).digest() sig=hmac.new(skey,f'HMAC-SHA256\n{cd}\n{ds}/{region}/{service}/request\n{hashlib.sha256(f"POST\n/\n{q}\ncontent-type:application/json\nhost:{host}\nx-content-sha256:{ph}\nx-date:{cd}\n\ncontent-type;host;x-content-sha256;x-date\n{ph}".encode()).hexdigest()}'.encode(),hashlib.sha256).hexdigest() return {'X-Date':cd,'Authorization':f'HMAC-SHA256 Credential={ak_v}/{ds}/{region}/{service}/request, SignedHeaders=content-type;host;x-content-sha256;x-date, Signature={sig}','X-Content-Sha256':ph,'Content-Type':'application/json'} # 获取公网URL img_url = upload_and_get_url(image) if not image.startswith("http") else image vid_url = upload_and_get_url(video) if not video.startswith("http") else video print(f" 头像URL: {img_url[:60]}...") print(f" 参考视频URL: {vid_url[:60]}...") p={'Action':'CVSync2AsyncSubmitTask','Version':'2022-08-31'} b=json.dumps({'req_key':'jimeng_dream_actor_m1_gen_video_cv','image_url':img_url,'video_url':vid_url}) r=httpx.post(f'https://{host}/',params=p,headers=sr(p,b),data=b,timeout=30).json() tid=r.get('data',{}).get('task_id','') print(f" 任务提交: {tid}") for i in range(60): time.sleep(10) p2={'Action':'CVSync2AsyncGetResult','Version':'2022-08-31'} b2=json.dumps({'req_key':'jimeng_dream_actor_m1_gen_video_cv','task_id':tid}) try: r2=httpx.post(f'https://{host}/',params=p2,headers=sr(p2,b2),data=b2,timeout=30).json() s=r2.get('data',{}).get('status','?');u=r2.get('data',{}).get('video_url','') if u: r3=httpx.get(u,timeout=120) with open(output,'wb') as f:f.write(r3.content) dur=r2.get('usage',{}).get('duration',0) get_tracker().log("DreamActor",duration=dur,resolution=resolution.replace("P","")+"P",note="动作模仿") break print(f' [{i+1}/60] {s}') if s in ('FAILED','failed'): print(f' 失败: {json.dumps(r2,ensure_ascii=False)[:300]}') break except Exception as e: print(f' [{i+1}/60] {e}') time.sleep(5) else: raise ValueError(f"未知数字人模式: {mode}") sz_mb = os.path.getsize(result) / 1024 / 1024 print(f"\n✅ 数字人完成: {result} ({sz_mb:.1f}MB)") return result def run_hybrid( voice: str, bgm: str, output: str, search: str = "", group_size: int = 4, clips: list[str] | None = None, texts: list[str] | None = None, enable_speed_ramp: bool = True, base_speed: float = 1.0, high_speed: float = 1.5, low_speed: float = 0.7, min_silence: float = 0.15, ) -> str: """口播+卡点融合策略""" from composer.hybrid import compose_hybrid cfg = load_config() pexels_key = cfg.get("pexels", {}).get("api_key", "") print(f"[视频工厂] 口播+卡点策略") print(f" 人声: {voice}") print(f" BGM: {bgm}") print(f" 变速: {'关闭' if not enable_speed_ramp else f'{low_speed:.1f}~{high_speed:.1f}x'}") # Dashboard 埋点 try: from lib.dashboard import init_db, log_production, log_step, complete_production init_db() prod_id = log_production("hybrid", f"混合_{Path(voice).stem}") log_step(prod_id, "compose", "开始") except Exception: pass result = compose_hybrid( voice_path=voice, bgm_path=bgm, output_path=output, material_clips=clips or None, texts=texts or None, pexels_api_key=pexels_key, pexels_search=search, enable_speed_ramp=enable_speed_ramp, base_speed=base_speed, high_speed=high_speed, low_speed=low_speed, group_size=group_size, min_silence_sec=min_silence, ) try: complete_production(prod_id) except Exception: pass return result def run_story( script: str, output: str, character: str = "", block: str = "", story_model: str = "turbo", story_duration: int = 5, story_lang: str = "en", seed: int = 42, dry_run: bool = False, force: bool = False, ) -> str: """故事策略: 剧本→场景分解→Kling批量生成→拼接""" from story_director.scene_planner import plan_scenes, export_scenes from story_director.temporal_bridge import enrich_scenes_with_bridges from story_director.batch_generator import run_story_pipeline cfg = load_config() print(f"[视频工厂] 故事策略: {script}") print(f" 角色: {character or '(无)'} Seed: {seed}") # Dashboard 埋点 try: from lib.dashboard import init_db, log_production, log_step, complete_production init_db() prod_id = log_production("故事", os.path.basename(script)) if prod_id > 0: from lib.cost_tracker import set_current_production_id set_current_production_id(prod_id) log_step(prod_id, "plan_scenes", "开始") except Exception: prod_id = -1 # 加载角色描述块 character_block = block if not character_block and character: from character_sheet import load_character char = load_character(character) if char: character_block = char.get("description", "") print(f" 角色库: '{character}' → {character_block[:40]}...") # Step 1: 场景分解 print(" [1/3] 场景分解...") scenes = plan_scenes( script_path=script, character_name=character or None, character_block=character_block, lang=story_lang, ) print(f" → {len(scenes)} 个场景") if prod_id > 0: log_step(prod_id, "plan_scenes", "completed", detail=f"{len(scenes)} 场景") # Step 2: 桥接 print(" [2/3] 过渡桥接...") enriched = enrich_scenes_with_bridges(scenes, character_block=character_block) import time tmp_scenes_path = f"/tmp/ave_story_scenes_factory_{int(time.time())}.json" export_scenes(scenes, tmp_scenes_path, character_block=character_block, lang=story_lang, seed=seed) if prod_id > 0: log_step(prod_id, "build_bridges", "completed") if dry_run: print("\n⏸️ Dry-run — 场景分解:") for s in enriched: bridge = s.get("bridge_to_next", {}) print(f" Scene {s['scene_id']}: {s.get('duration_sec', '?')}s " f"char={s.get('character_ref', 'none')} " f"trans={bridge.get('type', '-') if bridge else '-'}") print(f" Prompt: {s.get('prompt', '')[:80]}...") return "" # Step 3: 批量生成 print(" [3/3] 批量 Kling 生成...") pipeline_result = run_story_pipeline( scenes_path=tmp_scenes_path, output_dir="", model=story_model, duration=story_duration, force=force, ) # 尝试拼接最终视频 import subprocess concat_script = pipeline_result.get("concat_script", "") if concat_script and os.path.exists(concat_script): print(f" [后处理] 执行拼接脚本...") try: subprocess.run(["bash", concat_script, output], check=True, timeout=600) except subprocess.TimeoutExpired: print(" ⚠️ 拼接超时") except subprocess.CalledProcessError as e: print(f" ⚠️ 拼接失败: {e}") if prod_id > 0: try: complete_production(prod_id, output_path=output, total_cost=get_tracker()._total_cost()) set_current_production_id(0) except Exception: pass # 汇总 for r in pipeline_result.get("results", []): icon = "✅" if r["status"] == "success" else "💾" if r["status"] == "cached" else "❌" print(f" {icon} Scene {r['scene_id']:2d}: {r.get('path', 'N/A')}") return output def show_status(): """显示工厂状态""" print("=" * 40) print("🎬 AVE 视频工厂 · 状态") print("=" * 40) print() for name, info in STRATEGIES.items(): print(f" {name:5s}: {info['desc']}") print() tracker = get_tracker() tracker.summary()