#!/usr/bin/env python3 import base64, concurrent.futures, json, mimetypes, os, re, subprocess, time, uuid, urllib.request from pathlib import Path PROJECT=Path('/data/video-pipeline/Channel-Loi-Co-Nhan/project/13062026-024-Vo-Chong-Hon-Nhau-O-Chu-Nhuong') IMG_URL='http://192.168.40.11:20128/v1/images/generations' TTS_URL='http://192.168.1.201:3900/generate' REF_AUDIO=Path('/data/video-pipeline/Youtube-Video-Maker/data/voices/template1/refaudio.mp3') REF_TEXT=Path('/data/video-pipeline/Youtube-Video-Maker/data/voices/template1/reftext.txt') BGM=Path('/data/video-pipeline/Channel-Loi-Co-Nhan/data/bgm/the_mountain-history-archive-145197/the_mountain-history-archive-145197.mp3') LOGO=Path('/data/video-pipeline/Channel-Loi-Co-Nhan/data/logo/logoloiconhan.png') WORKERS=5 TTS_WORKERS=1 TTS_SPEED='0.86' SUB_FONT_SIZE=11 SCENE_PAUSE_SEC=0.5 INTRO_PAUSE_SEC=4.0 LOG=PROJECT/'logs/complete_short_project.log' LOG.parent.mkdir(parents=True, exist_ok=True) def log(s): line=time.strftime('%F %T ')+s print(line, flush=True) with LOG.open('a',encoding='utf-8') as f: f.write(line+'\n') def load_env(): p=Path('/home/hermes/.hermes/.env') if p.exists(): for line in p.read_text().splitlines(): if line.startswith('GEN_IMG_API_KEY=') and not os.environ.get('GEN_IMG_API_KEY'): os.environ['GEN_IMG_API_KEY']=line.split('=',1)[1].strip().strip('"').strip("'") def run(cmd): log('RUN '+' '.join(map(str,cmd))) subprocess.check_call(list(map(str,cmd))) def dur(p): out=subprocess.check_output(['ffprobe','-v','error','-show_entries','format=duration','-of','default=nw=1:nk=1',str(p)],text=True).strip() return float(out) def gen_image(prompt,out): load_env(); key=os.environ['GEN_IMG_API_KEY']; out.parent.mkdir(parents=True,exist_ok=True) while not out.exists() or out.stat().st_size==0: try: body={'model':'cx/gpt-5.5-image','prompt':prompt,'size':'1024x1792','n':1} req=urllib.request.Request(IMG_URL,data=json.dumps(body).encode(),headers={'Authorization':'Bearer '+key,'Content-Type':'application/json'},method='POST') with urllib.request.urlopen(req,timeout=600) as r: payload=json.loads(r.read().decode('utf-8','ignore')) b=payload['data'][0].get('b64_json') if b.startswith('data:'): b=b.split(',',1)[1] tmp=out.with_suffix('.tmp'); tmp.write_bytes(base64.b64decode(b)); tmp.replace(out) log(f'DONE image {out}') except Exception as e: log(f'RETRY image {out.name}: {type(e).__name__} {e}'); time.sleep(30) def multipart(fields, files): boundary='----OmniVoiceBoundary'+uuid.uuid4().hex; chunks=[] for k,v in fields.items(): chunks.append(f'--{boundary}\r\nContent-Disposition: form-data; name="{k}"\r\n\r\n{v}\r\n'.encode()) for k,path in files.items(): p=Path(path); mime=mimetypes.guess_type(str(p))[0] or 'application/octet-stream' chunks.append(f'--{boundary}\r\nContent-Disposition: form-data; name="{k}"; filename="{p.name}"\r\nContent-Type: {mime}\r\n\r\n'.encode()+p.read_bytes()+b'\r\n') chunks.append(f'--{boundary}--\r\n'.encode()); return boundary,b''.join(chunks) def tts_one(item): out=Path(item['audio_path']); out.parent.mkdir(parents=True,exist_ok=True) if out.exists() and out.stat().st_size>1000: log(f'SKIP existing tts {out}') return ref_text=REF_TEXT.read_text(encoding='utf-8') if REF_TEXT.exists() else '' text=re.sub(r'\s*[:;\-–—]\s*', ', ', item['text']).strip() attempt=0 while not out.exists() or out.stat().st_size<1000: attempt+=1 try: fields={'text':text,'language':'vi','ref_text':ref_text,'speed':TTS_SPEED,'num_step':'24','guidance_scale':'2.0','effect_preset':'broadcast'} boundary,body=multipart(fields, {'ref_audio':REF_AUDIO}) req=urllib.request.Request(TTS_URL,data=body,headers={'Content-Type':'multipart/form-data; boundary='+boundary},method='POST') with urllib.request.urlopen(req,timeout=180) as r: blob=r.read() tmp=out.with_suffix('.tmp'); tmp.write_bytes(blob) if not (blob[:4]==b'RIFF' and b'WAVE' in blob[:12]): raise RuntimeError(blob[:80]) tmp.replace(out) log(f'DONE tts {out}') except Exception as e: log(f'RETRY tts {item["id"]} attempt {attempt}: {type(e).__name__} {e}'); time.sleep(min(30*attempt,180)) def srt_time(sec): ms=int(round(sec*1000)); h=ms//3600000; ms%=3600000; m=ms//60000; ms%=60000; s=ms//1000; ms%=1000 return f'{h:02}:{m:02}:{s:02},{ms:03}' def main(): LOG.write_text('',encoding='utf-8') data=json.loads((PROJECT/'script/scenes.json').read_text(encoding='utf-8')) # Hard rule: generate scene images with WORKERS=5; never run this step sequentially. image_jobs=[] for sc in data['scenes']: out=PROJECT/'images/scenes'/(sc['scene_id']+'.png') image_jobs.append((sc.get('visual_prompt') or sc.get('visual_beat') or sc['narration'], out)) with concurrent.futures.ThreadPoolExecutor(max_workers=WORKERS) as ex: list(ex.map(lambda job: gen_image(*job), image_jobs)) thumb=PROJECT/'output/thumbnails/thumb_9x16_final.png' title=data.get('title') or data.get('metadata',{}).get('title','Cổ nhân dạy: Người khôn nói ít, kẻ dại nói nhiều') thumb_prompt=f"Vertical 9:16 thumbnail, cổ phong giấy vàng, aged yellow parchment, ink-wash cinematic Vietnamese moral story poster. Large readable Vietnamese brush title: {title}. Symbolic overflowing tea cup and calm ancient scholar, sepia bronze gold accents, mobile-safe composition, dramatic but dignified, no modern neon, no horizontal layout." gen_image(thumb_prompt, thumb) # TTS items: keep existing good segments on rerun; tts_one skips files >1000 bytes. items=[] for sc in data['scenes']: sid=sc['scene_id']; text=sc['narration'] items.append({'id':sid,'scene_id':sid,'text':text,'audio_path':str(PROJECT/'audio/voice_segments'/sid/'seg_001.wav')}) (PROJECT/'audio/voice_segments').mkdir(parents=True,exist_ok=True) (PROJECT/'audio/voice_segments/voice_segments_manifest.json').write_text(json.dumps({'items':items},ensure_ascii=False,indent=2),encoding='utf-8') with concurrent.futures.ThreadPoolExecutor(max_workers=TTS_WORKERS) as ex: list(ex.map(tts_one, items)) # concatenate audio and make timeline/srt with short subtitle cues only. concat=PROJECT/'audio/concat.txt'; lines=[]; timeline=[]; t=0.0; srt=[]; cue_debug=[]; cue_idx=1 silence=PROJECT/'audio/silence_scene_pause.wav' run(['ffmpeg','-y','-v','error','-f','lavfi','-i',f'anullsrc=r=44100:cl=mono','-t',SCENE_PAUSE_SEC,'-c:a','pcm_s16le',silence]) def word_chunks(text, n=5): words=text.replace('\n',' ').split(); out=[]; i=0 while i < len(words): remain=len(words)-i; take=n if remain>n else remain if remain==n+1: take=3 out.append(' '.join(words[i:i+take])); i+=take return out for idx,it in enumerate(items,1): ap=Path(it['audio_path']) d=dur(ap); visual_d=d+SCENE_PAUSE_SEC lines.append(f"file '{ap}'\n") lines.append(f"file '{silence}'\n") timeline.append({'scene_id':it['scene_id'],'start':t,'end':t+visual_d,'speech_end':t+d,'duration':visual_d,'audio_duration':d,'pause_sec':SCENE_PAUSE_SEC,'audio_path':str(ap),'image_path':str(PROJECT/'images/scenes'/f"{it['scene_id']}.png")}) chunks=word_chunks(it['text'],5) if chunks: step=d/len(chunks) for ci,txt in enumerate(chunks): cs=t+ci*step; ce=t+(ci+1)*step srt.append(f"{cue_idx}\n{srt_time(cs)} --> {srt_time(ce)}\n{txt}\n") cue_debug.append({'idx':cue_idx,'scene_id':it['scene_id'],'start':cs,'end':ce,'text':txt}) cue_idx+=1 t+=visual_d concat.write_text(''.join(lines),encoding='utf-8') master=PROJECT/'audio/master_narration.wav'; master.parent.mkdir(exist_ok=True) run(['ffmpeg','-y','-v','error','-f','concat','-safe','0','-i',concat,'-ar','44100','-ac','1','-c:a','pcm_s16le',master]) (PROJECT/'subtitles').mkdir(parents=True, exist_ok=True) (PROJECT/'audio/master_narration_timeline.json').write_text(json.dumps({'duration_sec': dur(master), 'items': timeline}, ensure_ascii=False, indent=2), encoding='utf-8') (PROJECT/'subtitles/subtitle_cue_debug_short.json').write_text(json.dumps(cue_debug,ensure_ascii=False,indent=2),encoding='utf-8') (PROJECT/'subtitles/master_subtitles_clamped.srt').write_text('\n'.join(srt), encoding='utf-8') intro_text = (data.get('metadata', {}).get('topic_sentence') or title).strip() intro_audio = PROJECT / 'output/intro/intro_audio.wav' intro_audio.parent.mkdir(parents=True, exist_ok=True) intro_audio.unlink(missing_ok=True) tts_one({'id': 'intro', 'text': intro_text, 'audio_path': str(intro_audio)}) log('STOP after audio/subtitle rebuild; production render/package must use worker-only script.') return # render main one-pass seglist=PROJECT/'render_onepass/segments.txt'; seglist.parent.mkdir(exist_ok=True) segtxt=[] for it in timeline: segtxt.append(f"file '{PROJECT/'images/scenes'/(it['scene_id']+'.png')}'\n") segtxt.append(f"duration {it['duration']:.3f}\n") segtxt.append(f"file '{PROJECT/'images/scenes'/(timeline[-1]['scene_id']+'.png')}'\n") seglist.write_text(''.join(segtxt),encoding='utf-8') mainv=PROJECT/'output/main_story.mp4'; mainv.parent.mkdir(parents=True,exist_ok=True) # Use filter_complex for the watermark; `-vf` cannot safely mix movie labels with the implicit input label. fc=( f"[0:v]scale=1080:1920:force_original_aspect_ratio=increase," f"crop=1080:1920,subtitles='{sub}':force_style='FontSize={SUB_FONT_SIZE},Outline=1,Alignment=2,MarginV=80'[base];" f"movie='{LOGO}',scale=160:-1,format=rgba,colorchannelmixer=aa=0.55[wm];" f"[base][wm]overlay=W-w-35:35[v]" ) run(['ffmpeg','-y','-v','error','-f','concat','-safe','0','-i',seglist,'-i',master,'-filter_complex',fc,'-map','[v]','-map','1:a','-r','30','-c:v','libx264','-pix_fmt','yuv420p','-c:a','aac','-shortest',mainv]) # intro one thumbnail frame + topic sentence audio intro_text=(data.get('metadata',{}).get('topic_sentence') or 'Cổ nhân dạy: Người khôn nói ít, kẻ dại nói nhiều.'); intro_audio=PROJECT/'output/intro/intro_audio.wav'; intro_audio.parent.mkdir(parents=True,exist_ok=True) tts_one({'id':'intro','text':intro_text,'audio_path':str(intro_audio)}) intro=PROJECT/'output/intro/intro.mp4' run(['ffmpeg','-y','-v','error','-loop','1','-i',thumb,'-i',intro_audio,'-vf','scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920','-r','30','-c:v','libx264','-pix_fmt','yuv420p','-c:a','aac','-shortest',intro]) pause=PROJECT/'output/intro/intro_pause_4s.mp4' run(['ffmpeg','-y','-v','error','-loop','1','-i',thumb,'-f','lavfi','-i','anullsrc=r=44100:cl=stereo','-t',INTRO_PAUSE_SEC,'-vf','scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920','-r','30','-c:v','libx264','-pix_fmt','yuv420p','-c:a','aac','-shortest',pause]) # final concat intro + 4s pause + main and bgm listp=PROJECT/'output/final/final_concat.txt'; listp.parent.mkdir(parents=True,exist_ok=True) listp.write_text(f"file '{intro}'\nfile '{pause}'\nfile '{mainv}'\n",encoding='utf-8') tmp=PROJECT/'output/final/final_no_bgm.mp4'; final=PROJECT/'output/final/final_video.mp4' concat_fc='[0:v]scale=1080:1920,setsar=1[v0];[0:a]aformat=sample_rates=44100:channel_layouts=stereo[a0];[1:v]scale=1080:1920,setsar=1[v1];[1:a]aformat=sample_rates=44100:channel_layouts=stereo[a1];[2:v]scale=1080:1920,setsar=1[v2];[2:a]aformat=sample_rates=44100:channel_layouts=stereo[a2];[v0][a0][v1][a1][v2][a2]concat=n=3:v=1:a=1[v][voice]' run(['ffmpeg','-y','-v','error','-i',intro,'-i',pause,'-i',mainv,'-filter_complex',concat_fc,'-map','[v]','-map','[voice]','-r','30','-c:v','libx264','-pix_fmt','yuv420p','-c:a','aac',tmp]) mix_fc='[0:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo[voice];[1:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,volume=0.18[bgm];[voice][bgm]amix=inputs=2:duration=first:dropout_transition=2[a]' run(['ffmpeg','-y','-v','error','-i',tmp,'-stream_loop','-1','-i',BGM,'-filter_complex',mix_fc,'-map','0:v','-map','[a]','-c:v','copy','-c:a','aac','-shortest',final]) info={'project_id':PROJECT.name,'status':'completed','thumbnail':str(thumb),'main_story':str(mainv),'final_video':str(final),'duration_sec':dur(final),'scenes':len(data['scenes']),'format':'9:16','tts_speed':TTS_SPEED,'subtitle_font_size':SUB_FONT_SIZE,'scene_pause_sec':SCENE_PAUSE_SEC,'intro_pause_sec':INTRO_PAUSE_SEC} (PROJECT/'output/info.json').write_text(json.dumps(info,ensure_ascii=False,indent=2),encoding='utf-8') log('COMPLETE '+str(final)) if __name__=='__main__': main()