Download app.py from Komyo1122/Recap: direct link, hf CLI and curl.
- Browser
- Download file 97.5 kB
-
https://huggingface.co/spaces/Komyo1122/Recap/resolve/main/app.py
- Command line
-
hf download hf://spaces/Komyo1122/Recap/app.py
-
curl -L -o app.py https://huggingface.co/spaces/Komyo1122/Recap/resolve/main/app.py
97.5 kB
import os
import uuid
import time
import asyncio
import gc
import re
import requests
import base64
import wave
import edge_tts
import shutil
import textwrap
import math
import traceback
import subprocess
from contextlib import asynccontextmanager
from fastapi import FastAPI, UploadFile, File, Form, BackgroundTasks, Header
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import FileResponse, JSONResponse
import google.generativeai as genai
from pydantic import BaseModel
from typing import List
import rabbit
import logging
import hashlib
# =========================================================================
# 💡 [TECH MONSTAR FIX] Bot များ၏ 404 Error များကို Log တွင် ဖျောက်ထားရန်
# =========================================================================
class HideBotLogsFilter(logging.Filter):
def filter(self, record):
msg = record.getMessage()
if " 404 " in msg:
return False
return True
logging.getLogger("uvicorn.access").addFilter(HideBotLogsFilter())
TMP_DIR = "/tmp/app_io"
os.makedirs(TMP_DIR, exist_ok=True)
USER_FONT_DIR = os.path.expanduser("~/.fonts")
os.makedirs(USER_FONT_DIR, exist_ok=True)
LOCAL_FONTS = [
"Zawgyi-One.ttf",
"Kanit-Regular.ttf",
"NotoSansSC-Regular.ttf",
"NotoSansMyanmar-Regular.ttf"
]
for f_name in LOCAL_FONTS:
local_path = os.path.join(os.getcwd(), f_name)
dest_path = os.path.join(USER_FONT_DIR, f_name)
if os.path.exists(local_path):
if not os.path.exists(dest_path):
try: shutil.copy(local_path, dest_path)
except: pass
os.system("fc-cache -f -v")
jobs = {}
upload_tokens = {}
running_tasks = {}
active_user_jobs = {}
MAX_CONCURRENT_JOBS = 3
queue_semaphore = asyncio.Semaphore(MAX_CONCURRENT_JOBS)
tts_semaphore = asyncio.Semaphore(5)
api_key_counter = 0
BACKEND_SECRET = os.getenv("BACKEND_SECRET", "MY_SUPER_SECRET_TOKEN_123").strip()
VOICE_MAP = {
"thiha": "my-MM-ThihaNeural", "nilar": "my-MM-NilarNeural",
"thai_male": "th-TH-NiwatNeural", "thai_female": "th-TH-PremwadeeNeural",
"english_male": "en-US-BrianNeural", "english_female": "en-US-EmmaNeural",
"chinese_male": "zh-CN-YunxiNeural", "chinese_female": "zh-CN-XiaoxiaoNeural",
"gemini_aoede": "Aoede", "gemini_charon": "Charon", "gemini_kore": "Kore",
"gemini_fenrir": "Fenrir", "gemini_puck": "Puck", "gemini_zephyr": "Zephyr",
"gemini_leda": "Leda", "gemini_orus": "Orus", "gemini_autonoe": "Autonoe",
"gemini_callirrhoe": "Callirrhoe", "gemini_enceladus": "Enceladus", "gemini_iapetus": "Iapetus",
"gemini_umbriel": "Umbriel", "gemini_algieba": "Algieba", "gemini_despina": "Despina",
"gemini_erinome": "Erinome", "gemini_algenib": "Algenib", "gemini_rasalgethi": "Rasalgethi",
"gemini_laomedeia": "Laomedeia", "gemini_achernar": "Achernar", "gemini_alnilam": "Alnilam",
"gemini_schedar": "Schedar", "gemini_gacrux": "Gacrux", "gemini_pulcherrima": "Pulcherrima",
"gemini_achird": "Achird", "gemini_zubenelgenubi": "Zubenelgenubi", "gemini_vindemiatrix": "Vindemiatrix",
"gemini_sadachbia": "Sadachbia", "gemini_sadaltager": "Sadaltager", "gemini_sulafat": "Sulafat"
}
DUAL_VOICE_MAP = {
"thiha": {"edge": "my-MM-ThihaNeural", "gemini": "Charon"},
"nilar": {"edge": "my-MM-NilarNeural", "gemini": "Aoede"},
"thai_male": {"edge": "th-TH-NiwatNeural", "gemini": "Charon"},
"thai_female": {"edge": "th-TH-PremwadeeNeural", "gemini": "Aoede"},
"english_male": {"edge": "en-US-BrianNeural", "gemini": "Charon"},
"english_female": {"edge": "en-US-EmmaNeural", "gemini": "Aoede"},
"chinese_male": {"edge": "zh-CN-YunxiNeural", "gemini": "Charon"},
"chinese_female": {"edge": "zh-CN-XiaoxiaoNeural", "gemini": "Aoede"}
}
def emergency_disk_cleanup():
try:
total_size = sum(os.path.getsize(os.path.join(TMP_DIR, f)) for f in os.listdir(TMP_DIR) if os.path.isfile(os.path.join(TMP_DIR, f)))
if total_size < 50000 * 1024 * 1024: return
current_time = time.time()
for f in os.listdir(TMP_DIR):
f_path = os.path.join(TMP_DIR, f)
if os.path.isfile(f_path) and (current_time - os.path.getmtime(f_path) > 10800):
os.remove(f_path)
gc.collect()
except: pass
# 💡 [THE FIX] Command များ ပိတ်မိမနေစေရန် Timeout ထည့်သွင်းထားပါသည်
async def run_cmd_async(cmd_list, retries=3, timeout=300):
for attempt in range(retries):
try:
process = await asyncio.create_subprocess_exec(*cmd_list, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
stdout, stderr = await asyncio.wait_for(process.communicate(), timeout=timeout)
if process.returncode == 0: return stdout.decode()
await asyncio.sleep(2)
except asyncio.TimeoutError:
try: process.kill()
except: pass
except asyncio.CancelledError:
try: process.kill()
except: pass
raise
except Exception:
try: process.kill()
except: pass
raise Exception(f"FFmpeg Task Failed/Timeout after {retries} attempts.")
async def get_media_duration(file_path):
cmd = ["ffprobe", "-v", "error", "-show_entries", "format=duration", "-of", "default=noprint_wrappers=1:nokey=1", file_path]
try: out = await run_cmd_async(cmd); return float(out.strip())
except: return 0.0
def get_api_keys_pool():
keys = []
main_key = os.getenv("GEMINI_API_KEY", "").strip()
if main_key and main_key.startswith("AIza"): keys.append(main_key.replace('"', '').replace("'", "").strip())
for i in range(2, 101):
backup_key = os.getenv(f"GEMINI_API_KEY_{i}", "").strip()
if backup_key and backup_key.startswith("AIza"): keys.append(backup_key.replace('"', '').replace("'", "").strip())
return keys
def get_working_api_key(user_key: str):
global api_key_counter
keys = []
if user_key and user_key.strip(): keys.extend([k.strip() for k in user_key.split(",") if k.strip().startswith("AIza")])
keys.extend(get_api_keys_pool())
if not keys: return None
selected_key = keys[api_key_counter % len(keys)]
api_key_counter += 1
return selected_key
@asynccontextmanager
async def lifespan(app: FastAPI):
async def periodic_cleanup():
while True:
await asyncio.sleep(1800)
current_time = time.time()
for j_id in list(jobs.keys()):
if current_time - jobs[j_id].get("created_at", 0) > 10800:
try:
for prefix in ["temp_", "out_", "ext_", "tts_", "wm_", "bgm_", "list_", "trim_", "sub_", "sil_", "concat_", "ov_", "bp_", "ai_vid_"]:
for f in os.listdir(TMP_DIR):
if f.startswith(prefix) and j_id in f: os.remove(os.path.join(TMP_DIR, f))
except: pass
del jobs[j_id]
for cache_f in os.listdir(TMP_DIR):
c_p = os.path.join(TMP_DIR, cache_f)
if cache_f.startswith("cache_") and current_time - os.path.getmtime(c_p) > 10800:
try: os.remove(c_p)
except: pass
for t in list(upload_tokens.keys()):
if current_time > upload_tokens[t]: del upload_tokens[t]
gc.collect()
asyncio.create_task(periodic_cleanup())
yield
app = FastAPI(lifespan=lifespan)
app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"])
class TokenRequest(BaseModel): secret: str
@app.post("/generate-upload-token")
async def generate_upload_token(req: TokenRequest):
if req.secret != BACKEND_SECRET: return JSONResponse(status_code=403, content={"error": "Unauthorized Server"})
new_otp_token = str(uuid.uuid4())
upload_tokens[new_otp_token] = time.time() + 3600
return JSONResponse(content={"token": new_otp_token})
async def generate_tts_audio(text, voice, rate_str, pitch_str, output_path):
communicate = edge_tts.Communicate(text.strip(), voice, rate=rate_str, pitch=pitch_str)
await communicate.save(output_path)
def generate_gemini_tts_audio_single(text, voice_name, user_api_key, output_path):
for attempt in range(3):
time.sleep(1)
api_key = get_working_api_key(user_api_key)
if not api_key: raise Exception("API Key Missing!")
url = f"https://generativelanguage.googleapis.com/v1alpha/models/gemini-3.1-flash-tts-preview:generateContent?key={api_key}"
payload = {"contents": [{"parts": [{"text": text}]}], "generationConfig": {"responseModalities": ["AUDIO"], "speechConfig": {"voiceConfig": {"prebuiltVoiceConfig": {"voiceName": voice_name}}}}}
response = requests.post(url, json=payload, timeout=60)
if response.status_code == 200:
try:
audio_base64 = response.json()['candidates'][0]['content']['parts'][0]['inlineData']['data']
audio_base64 += "=" * ((4 - len(audio_base64) % 4) % 4)
with wave.open(output_path, "wb") as wf:
wf.setnchannels(1); wf.setsampwidth(2); wf.setframerate(24000)
wf.writeframes(base64.b64decode(audio_base64))
return
except: raise Exception("Gemini Data Error")
elif attempt == 2: raise Exception(f"Gemini API Error: {response.status_code}")
async def process_single_line_tts(idx, text, time_ms, orig_end_ms, unique_job_id, v_name_check, edge_voice, gemini_voice, user_api_key, pitch, lang_key, tts_speed, cache_job_id=None):
clean_text = re.sub(r'\[\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?\s*-\s*\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?\]', '', text).strip()
clean_text = re.sub(r'\[.*?\]|\(.*?\)|<.*?>', '', clean_text).strip()
clean_text = re.sub(r'["\'“”‘’`\-—_]', ' ', clean_text).strip()
clean_text = re.sub(r'\.+', ' ', clean_text).strip()
clean_text = re.sub(r'[?!,;]', ' ', clean_text).strip()
if not clean_text: return None
sub_wav = os.path.join(TMP_DIR, f"tts_{unique_job_id}_{idx}.wav")
try: f_speed = float(tts_speed); f_speed = f_speed if f_speed > 0 else 1.15
except: f_speed = 1.15
ffmpeg_speed = str(f_speed)
tts_text = clean_text
if not v_name_check.startswith("gemini_"): tts_text = re.sub(r'\s+', ' ', tts_text).strip()
is_success = False
async with tts_semaphore:
for attempt in range(20):
try:
if v_name_check.startswith("gemini_"):
raw_gemini = os.path.join(TMP_DIR, f"tts_{unique_job_id}_{idx}_raw_gemini.wav")
if cache_job_id:
c_raw = os.path.join(TMP_DIR, f"tts_{cache_job_id}_{idx}_raw_gemini.wav")
if os.path.exists(c_raw): raw_gemini = c_raw
if not os.path.exists(raw_gemini):
await asyncio.to_thread(generate_gemini_tts_audio_single, clean_text, gemini_voice, user_api_key, raw_gemini)
if os.path.exists(raw_gemini):
silence_filter = f"silenceremove=start_periods=1:start_threshold=-60dB,areverse,silenceremove=start_periods=1:start_threshold=-60dB,areverse,atempo={ffmpeg_speed}"
await run_cmd_async(["ffmpeg", "-y", "-i", raw_gemini, "-filter:a", silence_filter, sub_wav])
else:
raw_edge = os.path.join(TMP_DIR, f"tts_{unique_job_id}_{idx}_raw_edge.wav")
if cache_job_id:
c_raw = os.path.join(TMP_DIR, f"tts_{cache_job_id}_{idx}_raw_edge.wav")
if os.path.exists(c_raw): raw_edge = c_raw
if not os.path.exists(raw_edge):
await generate_tts_audio(tts_text, edge_voice, "+0%", f"{int(pitch):+d}Hz", raw_edge)
if os.path.exists(raw_edge):
silence_filter = f"silenceremove=start_periods=1:start_threshold=-55dB,areverse,silenceremove=start_periods=1:start_threshold=-55dB,areverse,atempo={ffmpeg_speed}"
await run_cmd_async(["ffmpeg", "-y", "-i", raw_edge, "-filter:a", silence_filter, sub_wav])
if os.path.exists(sub_wav) and os.path.getsize(sub_wav) > 0:
is_success = True; break
else: await asyncio.sleep(2)
except Exception: await asyncio.sleep(2)
if not is_success: raise Exception(f"❌ အသံဖန်တီးမှု မအောင်မြင်ပါ။ စာသား: '{clean_text[:20]}...'")
dur = int((await get_media_duration(sub_wav)) * 1000)
return {"idx": idx, "path": sub_wav, "actual_dur": dur, "time_ms": time_ms, "orig_end_ms": orig_end_ms, "clean_text": clean_text}
async def wrapper_process_task(*args, **kwargs):
async with queue_semaphore: await process_video_task(*args, **kwargs)
async def process_video_task(job_id: str, user_id: str, file_path: str, wm_path: str, bgm_path: str, ov_path: str, voice_name: str, user_api_key: str, ratio: str, resolution: str, voice_toggle: str, hide_subtitle: str, volume: str, pitch: str, bypass_mode: str, logo_pos: str, logo_size: str, logo_opacity: str, custom_prompt: str, genre: str, bgm_volume: str, target_chunk: str, chunk_minutes: str, mask_toggle: str, mask_y: str, mask_h: str, mask_w: str, sub_toggle: str, sub_y: str, sub_size: str, sub_color: str, tts_speed: str, bgm_toggle: str, logo_toggle: str, logo_shape: str, logo_custom_x: str, logo_custom_y: str, special_effect: str, marquee_toggle: str, marquee_text: str, marquee_y: str, pip_shape: str, pip_pos: str, overlay_size: str, pip_custom_x: str, pip_custom_y: str, bouncing_pip_toggle: str, bp_path: str, bouncing_pip_shape: str, bouncing_pip_size: str, bouncing_pip_opacity: str, jumpcut_mode: str = "off", is_remerge: bool = False, original_job_id: str = None, force_recreate: str = "false", tab_mode: str = "translate"):
output_video_path = os.path.join(TMP_DIR, f"out_{job_id}.mp4")
working_video_path = file_path
v_check = voice_name.lower()
is_thai_mode = "thai" in v_check or "th-" in v_check or "ထိုင်း" in v_check
is_chinese_mode = "chinese" in v_check or "zh-" in v_check or "တရုတ်" in v_check
is_eng_mode = "english" in v_check or "en-" in v_check or "အင်္ဂလိပ်" in v_check
lang_key = "thai" if is_thai_mode else "chinese" if is_chinese_mode else "english" if is_eng_mode else "myanmar"
try:
chunks_info = []
is_full_auto = False
if is_remerge and original_job_id in jobs:
print(f"🚀 [{job_id}] ပြန်လည်ပေါင်းစပ်ခြင်း စတင်ပါပြီ...", flush=True)
orig_job = jobs[original_job_id]
working_video_path = orig_job.get("working_video_path", file_path)
chunks_info = orig_job.get("chunks_info", [])
is_full_auto = len(chunks_info) > 1
v_dur_full = orig_job.get("v_dur_full", 1.0)
else:
if jobs.get(job_id, {}).get("status") == "cancelled": raise Exception("လုပ်ငန်းစဉ်ကို ရပ်ဆိုင်းလိုက်ပါသည်။")
v_dur_full = await get_media_duration(working_video_path)
if v_dur_full <= 0: v_dur_full = 1.0
try:
c_mins = int(chunk_minutes)
except:
c_mins = 2
if target_chunk.startswith("auto_"):
try:
parts = target_chunk.split("_")
if len(parts) > 1:
c_mins = int(parts[1])
except:
pass
chunk_duration_sec = c_mins * 60
if target_chunk.startswith("auto"):
is_full_auto = True
total_chunks = math.ceil(v_dur_full / chunk_duration_sec)
for i in range(total_chunks):
chunks_info.append({"start_time": i * chunk_duration_sec, "duration": chunk_duration_sec, "real_idx": i})
elif target_chunk.isdigit():
is_full_auto = False
chunk_num = int(target_chunk)
start_time = (chunk_num - 1) * chunk_duration_sec
chunks_info.append({"start_time": start_time, "duration": chunk_duration_sec, "real_idx": chunk_num - 1})
else:
is_full_auto = True
total_chunks = math.ceil(v_dur_full / chunk_duration_sec)
for i in range(total_chunks):
chunks_info.append({"start_time": i * chunk_duration_sec, "duration": chunk_duration_sec, "real_idx": i})
jobs[job_id]["chunks_info"] = chunks_info
jobs[job_id]["v_dur_full"] = v_dur_full
completed_chunk_videos = []
full_transcript_list = []
total_cost_accum = 0
total_chunks_count = len(chunks_info)
video_file_ai = None
jobs[job_id]["total_chunks"] = total_chunks_count
previous_chunk_memory = ""
character_sheet_memory = ""
base_res = 720 if resolution == "720p" else 1080
if ratio == "9:16":
target_w = base_res; target_h = int(target_w * 16 / 9)
elif ratio == "1:1":
target_w = base_res; target_h = base_res
elif ratio == "4:5":
target_w = base_res; target_h = int(target_w * 5 / 4)
elif ratio == "4:3":
target_h = base_res; target_w = int(target_h * 4 / 3)
else: # 16:9
target_h = base_res; target_w = int(target_h * 16 / 9)
target_w = (target_w // 2) * 2
target_h = (target_h // 2) * 2
for c_idx, chunk_data in enumerate(chunks_info):
jobs[job_id]["current_chunk"] = c_idx + 1
# 💡 [THE FIX] UI တန်းတက်သွားစေရန် status ကို အရင်ဆုံး သတ်မှတ်မည်
jobs[job_id]["status"] = "generating_audio" if is_remerge else "transcribing"
jobs[job_id]["progress_msg"] = f" [အပိုင်း {c_idx+1}/{total_chunks_count}] ဗီဒီယို ဖြတ်တောက်နေပါသည်..."
if jobs.get(job_id, {}).get("status") == "cancelled": raise Exception("လုပ်ငန်းစဉ်ကို ရပ်ဆိုင်းလိုက်ပါသည်။")
start_time = chunk_data["start_time"]
chunk_duration_sec = chunk_data["duration"]
if start_time >= v_dur_full: break
actual_chunk_dur = min(chunk_duration_sec, v_dur_full - start_time)
msg_prefix = f" [အပိုင်း {c_idx+1}/{total_chunks_count}] " if is_full_auto else " "
real_idx = chunk_data.get("real_idx", c_idx)
cache_str = f"{working_video_path}_{wm_path}_{ov_path}_{bp_path}_{real_idx}_{actual_chunk_dur}_{voice_name}_{tts_speed}_{genre}_{custom_prompt}_{sub_toggle}_{sub_y}_{sub_size}_{sub_color}_{bypass_mode}_{voice_toggle}_{bgm_toggle}_{bgm_volume}_{logo_toggle}_{logo_pos}_{logo_size}_{logo_opacity}_{logo_shape}_{logo_custom_x}_{logo_custom_y}_{mask_toggle}_{mask_y}_{mask_h}_{mask_w}_{special_effect}_{marquee_toggle}_{marquee_text}_{marquee_y}_{bouncing_pip_toggle}_{bouncing_pip_shape}_{bouncing_pip_size}_{bouncing_pip_opacity}_{pip_shape}_{pip_pos}_{overlay_size}_{pip_custom_x}_{pip_custom_y}_{jumpcut_mode}_{ratio}_{resolution}_{volume}_{pitch}_{tab_mode}"
chunk_cache_id = hashlib.md5(cache_str.encode()).hexdigest()
trim_vid = os.path.join(TMP_DIR, f"trim_{job_id}_{c_idx}.mp4")
tts_audio = os.path.join(TMP_DIR, f"tts_{job_id}_{c_idx}_final.wav")
sub_file = os.path.join(TMP_DIR, f"sub_{job_id}_{c_idx}.ass")
chunk_out_vid = os.path.join(TMP_DIR, f"chunk_out_{chunk_cache_id}.mp4")
chunk_txt_file = os.path.join(TMP_DIR, f"chunk_txt_{chunk_cache_id}.txt")
if not is_remerge and force_recreate == "true":
try:
if os.path.exists(chunk_out_vid): os.remove(chunk_out_vid)
if os.path.exists(chunk_txt_file): os.remove(chunk_txt_file)
except: pass
if not is_remerge and os.path.exists(chunk_out_vid) and os.path.getsize(chunk_out_vid) > 0 and os.path.exists(chunk_txt_file):
print(f"⏩ [RESUME] Skipping Chunk {c_idx+1}, loading from memory.")
completed_chunk_videos.append(chunk_out_vid)
with open(chunk_txt_file, "r", encoding="utf-8") as f:
memory_text = f.read()
full_transcript_list.append(memory_text)
matches = list(re.finditer(r'\[(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\s*-\s*(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\]\s*([^\[]+)', memory_text))
if matches:
last_few_lines = [re.sub(r'\[.*?\]', '', m.group(3)).strip() for m in matches[-4:]]
previous_chunk_memory = " ".join(last_few_lines).strip()
continue
if is_remerge:
old_trim = os.path.join(TMP_DIR, f"trim_{original_job_id}_{c_idx}.mp4")
if os.path.exists(old_trim):
trim_vid = old_trim
else:
await run_cmd_async(['ffmpeg', '-y', '-ss', str(start_time), '-i', working_video_path, '-t', str(actual_chunk_dur), '-c', 'copy', trim_vid])
else:
await run_cmd_async(['ffmpeg', '-y', '-ss', str(start_time), '-i', working_video_path, '-t', str(actual_chunk_dur), '-c', 'copy', trim_vid])
c_v_dur = await get_media_duration(trim_vid)
if c_v_dur <= 0: c_v_dur = 1.0
parsed_lines = []
if is_remerge:
parsed_lines = chunk_data.get("parsed_lines", [])
else:
# 💡 [THE FIX] 10% တွင် ရပ်မနေစေရန် AI သို့ပို့မည့် ဗီဒီယိုအား (480p, 10fps) သို့ အမြန်ဆုံး ချုံ့ပစ်ပါမည်
ai_vid = os.path.join(TMP_DIR, f"ai_vid_{job_id}_{c_idx}.mp4")
jobs[job_id]["progress_msg"] = f"{msg_prefix}AI အတွက် ဗီဒီယိုအကြမ်း ပြင်ဆင်နေပါသည်..."
await run_cmd_async(['ffmpeg', '-y', '-i', trim_vid, '-vf', 'scale=-2:480', '-r', '10', '-c:v', 'libx264', '-preset', 'ultrafast', '-crf', '35', '-c:a', 'libmp3lame', '-b:a', '64k', ai_vid])
genre_instructions = {
"original": "မူရင်းအတိုင်း သဘာဝကျသော အသုံးအနှုန်းများဖြင့် ပြောပါ။",
"comedy": "⚠️ ဤကားသည် 'ချိုင့်ကေကို (ဟာသ)' စတိုင်ဖြစ်သည်။ အလွန်ရယ်စရာကောင်းသော၊ ဟာသမြောက်သော ဗန်းစကားများ နှင့် ရယ်သံများကို သုံးပါ။",
"action": "⚠️ 'အက်ရှင်' စတိုင်ဖြစ်သည်။ မြန်ဆန်သွက်လက်ပြီး စိတ်လှုပ်ရှားဖွယ်ကောင်းအောင် အသက်ဝင်ဝင် ပြောပါ။",
"horror": "⚠️ 'မဖဲဝါ (သင်္ချိုင်းစောင့်သူရဲမ)' စတိုင်ဖြစ်သည်။ ခြောက်ခြားဖွယ်ကောင်းသော၊ ထစ်ငေါ့နေသော လေသံဖြင့် ကျောချမ်းအောင် ရေးပါ။",
"motivational": "⚠️ 'တက်ကြွတဲ့ကြက်ဖ' စတိုင်ဖြစ်သည်။ မလျှော့သော ဇွဲလုံ့လနှင့် ရင်ထဲကို ထိမှန်စေမည့် အားပေးစကားများ ထည့်သုံးပါ။",
"happy": "⚠️ 'ပျော်ရွှင်ဖွယ်လူငယ်' စတိုင်ဖြစ်သည်။ တက်ကြွရွှင်လန်းပြီး ပျော်စရာကောင်းသော လေသံဖြင့် ပြောပါ။",
"sad": "⚠️ 'ဝမ်းနည်းဖွယ်လူငယ်' စတိုင်ဖြစ်သည်။ လွမ်းဆွေးဖွယ်ကောင်းသော၊ သနားစရာကောင်းသော လေသံဖြင့် ပြောပါ။",
"story": "⚠️ 'ပုံပြင်' စတိုင်ဖြစ်သည်။ အဖိုးအဖွားတစ်ဦးက မြေးလေးများကို ပုံပြင်ပြောပြနေသကဲ့သို့ ချိုသာစွာ ဇာတ်ကြောင်းပြန်ပါ။"
}
genre_prompt = genre_instructions.get(genre, genre_instructions["original"])
custom_prompt_text = ""
if custom_prompt and custom_prompt.strip():
custom_prompt_text = f"⚠️ [USER'S SPECIAL INSTRUCTION - OVERRIDE ALL]: {custom_prompt.strip()}"
memory_injection = ""
if character_sheet_memory:
memory_injection += f"⚠️ [CRITICAL MEMORY]: Always remember these key details: '{character_sheet_memory}'. Keep all names consistent.\n"
if previous_chunk_memory:
memory_injection += f"⚠️ [PREVIOUS STORY CONTEXT]: The last part ended with: '{previous_chunk_memory}'. Continue naturally from here."
current_is_first = (chunk_data.get("real_idx", c_idx) == 0)
if tab_mode == "recap":
my_translation_rule = "⚠️ [RECAP MODE - CRITICAL]: ဇာတ်ကောင်များပြောသော စကားများကို တစ်ကြောင်းချင်း တိုက်ရိုက် ဘာသာမပြန်ပါနှင့်။ ဗီဒီယို၏ ဇာတ်လမ်းသွားကို နားလည်အောင်ကြည့်ပြီး၊ မြင်ကွင်းများကို 'ရုပ်ရှင်အကျဉ်းချုပ် (Movie Recap)' စတိုင်ဖြင့် ကိုယ်ပိုင်အသုံးအနှုန်းများ သုံးကာ ဆွဲဆောင်မှုရှိအောင် ဇာတ်ကြောင်းပြန်ပါ။"
en_translation_rule = "⚠️ [RECAP MODE - CRITICAL]: Do NOT translate dialogues line-by-line. Write a highly engaging 'Movie Recap' script describing the scenes and plot in your own words."
th_translation_rule = "⚠️ [RECAP MODE - CRITICAL]: ห้ามแปลบทสนทนาแบบตรงตัวทุกบรรทัด ให้สรุปและเขียนบท 'สปอยหนัง (Movie Recap)' ที่น่าติดตามโดยใช้คำพูดของคุณเอง"
zh_translation_rule = "⚠️ [RECAP MODE - CRITICAL]: 严禁逐句翻译对话。请用你自己的语言撰写引人入胜的“电影解说”脚本。"
else:
my_translation_rule = "မူရင်းက စာရှည်လျှင် ရှည်သလို၊ တိုလျှင် တိုသလို (Line-by-line) ဘာသာပြန်ပါ။ အလွန်အမင်း အကျဉ်းချုံ့ပစ်ခြင်း လုံးဝ မလုပ်ပါနှင့်။ အစအဆုံး အပြည့်အစုံ ဘာသာပြန်ပါ။"
en_translation_rule = "Translate the video audio line-by-line completely. Do not over-summarize."
th_translation_rule = "แปลเสียงวิดีโอแบบบรรทัดต่อบรรทัดอย่างสมบูรณ์ ห้ามสรุปมากเกินไป"
zh_translation_rule = "完整地逐句翻译视频音频。不要过度总结。"
my_human_narrator_rule = "⚠️ [HUMAN NARRATOR - CRITICAL]: သင်သည် ပုံပြောကောင်းသော၊ လူသားစစ်စစ် ဇာတ်ကြောင်းပြန်သူ (Human Storyteller) ဖြစ်သည်။ စက်ရုပ်ဆန်သော စကားလုံးခြောက်များ၊ တိုပြတ်သော ဝါကျများကို လုံးဝမသုံးရ။ စိတ်ခံစားမှုပါသော စကားလုံးများ၊ ဆွဲဆောင်မှုရှိသော ဝါကျအရှည်အတိုများဖြင့် ပရိသတ်ကို ဇာတ်လမ်းထဲ စီးမျောသွားအောင် ပြောပါ။"
my_consistency_rule = "⚠️ [MEMORY & CONSISTENCY - CRITICAL]: သင်သည် စုံထောက်တစ်ဦးကဲ့သို့ အသေးစိတ်ကို အာရုံစိုက်ရမည်။ ဇာတ်ကောင်အမည်များ၊ နေရာဒေသများ၊ ခုနှစ်သက္ကရာဇ်များကို ဗီဒီယိုတစ်လျှောက်လုံး လုံးဝမမှားစေရ။ အစမှာ 'ဂျွန်' ဟု ပြောပြီး နောက်ပိုင်းတွင် သူ့နာမည်ကို မေ့ကျန်ခဲ့ခြင်းမျိုး လုံးဝမဖြစ်စေရ။"
en_human_narrator_rule = "⚠️ [HUMAN NARRATOR - CRITICAL]: You are a charismatic, human storyteller. Avoid robotic, dry, or simple sentences. Use vivid verbs, emotional adjectives, and varied sentence structures to immerse the audience in the story."
en_consistency_rule = "⚠️ [MEMORY & CONSISTENCY - CRITICAL]: You must act like a detective, paying close attention to every detail. Character names, locations, and dates must remain 100% consistent throughout the video. Do not mention a name like 'John' at the start and then forget it later."
th_human_narrator_rule = "⚠️ [HUMAN NARRATOR - CRITICAL]: คุณคือผู้บรรยายที่เป็นมนุษย์และมีเสน่ห์ หลีกเลี่ยงประโยคที่เหมือนหุ่นยนต์ แห้งแล้ง หรือเรียบง่าย ใช้คำกริยาที่ชัดเจน คำคุณศัพท์ที่แสดงอารมณ์ และโครงสร้างประโยคที่หลากหลายเพื่อให้ผู้ชมอินไปกับเรื่องราว"
th_consistency_rule = "⚠️ [MEMORY & CONSISTENCY - CRITICAL]: คุณต้องทำตัวเหมือนนักสืบ ใส่ใจทุกรายละเอียด ชื่อตัวละคร สถานที่ และวันที่ต้องสอดคล้องกัน 100% ตลอดทั้งวิดีโอ อย่าเอ่ยชื่อ 'จอห์น' ในตอนแรกแล้วลืมในภายหลัง"
zh_human_narrator_rule = "⚠️ [HUMAN NARRATOR - CRITICAL]: 你是一位富有魅力的人类故事讲述者。避免使用机器人般、枯燥或简单的句子。使用生动的动词、富有情感的形容词和多样的句式结构,让观众沉浸在故事中。"
zh_consistency_rule = "⚠️ [MEMORY & CONSISTENCY - CRITICAL]: 你必须像侦探一样,密切关注每一个细节。角色名称、地点和日期在整个视频中必须保持100%一致。不要在开头提到像“约翰”这样的名字,然后又忘记了。"
my_pacing = "⚠️ [PACING & CONCISENESS]: ဇာတ်လမ်းပြောရာတွင် ကျစ်လျစ်ပါစေ။ တိုက်ခိုက်ခန်းများ သို့မဟုတ် လှုပ်ရှားမှုများကို တစ်ခုချင်းစီ အသေးစိတ် လိုက်မပြောပါနှင့်။ အဖြစ်အပျက်ကို အကျဉ်းချုပ်ပြီးသာ ပြောပါ။"
en_pacing = "⚠️ [PACING & CONCISENESS]: Keep narration CONCISE. Do NOT over-describe every single action. Summarize briefly."
th_pacing = "⚠️ [PACING & CONCISENESS]: เล่าเรื่องให้กระชับ ห้ามอธิบายทุกการกระทำโดยละเอียด สรุปสั้นๆ"
zh_pacing = "⚠️ [PACING & CONCISENESS]: 保持叙述简明扼要。不要过度描述每一个动作。简要总结事件。"
if is_eng_mode:
intro_rule = "⚠️ [INTRO RULE]: The very first line MUST be exactly [00:00 - 00:02] Hello." if current_is_first else "⚠️ [NO INTRO RULE]: Continuation chunk. NO greetings."
prompt = (f"You are a professional movie recap narrator and storyteller. Watch the video carefully and narrate the story in English.\n"
f"⚠️ CRITICAL INSTRUCTIONS:\n"
f"1. {en_translation_rule}\n"
f"2. {en_human_narrator_rule}\n"
f"3. {en_consistency_rule}\n"
f"4. {en_pacing}\n"
f"5. ⚠️ [VIDEO ANALYSIS]: Watch the characters' actions, emotions, and key events. Describe the story seamlessly, even if there is no dialogue.\n"
f"6. {intro_rule}\n"
f"{memory_injection}{custom_prompt_text} MOVIE GENRE: {genre_prompt}\n"
f"FINAL INSTRUCTION: Use only [MM:SS - MM:SS] format for EVERY line. ACT AS A CHARISMATIC HUMAN NARRATOR.")
elif is_thai_mode:
intro_rule = "⚠️ [INTRO RULE]: บรรทัดแรกสุดต้องเป็น [00:00 - 00:02] สวัสดี อย่างเคร่งครัด" if current_is_first else "⚠️ [NO INTRO RULE]: ส่วนต่อขยาย ห้ามทักทาย"
prompt = (f"You are a professional movie recap narrator. Watch the video carefully and narrate the story in Thai Language (ภาษาไทย).\n"
f"⚠️ CRITICAL INSTRUCTIONS:\n"
f"1. {th_translation_rule}\n"
f"2. {th_human_narrator_rule}\n"
f"3. {th_consistency_rule}\n"
f"4. {th_pacing}\n"
f"5. ⚠️ [VIDEO ANALYSIS]: ดูการกระทำ อารมณ์ และเหตุการณ์สำคัญในวิดีโออย่างละเอียด เล่าเรื่องอย่างต่อเนื่องแม้จะไม่มีคนพูดก็ตาม\n"
f"6. {intro_rule}\n"
f"{memory_injection}{custom_prompt_text} MOVIE GENRE: {genre_prompt}\n"
f"FINAL INSTRUCTION: ใช้รูปแบบ [MM:SS - MM:SS] สำหรับทุกบรรทัดเท่านั้น ทำตัวเป็นผู้บรรยายที่มีเสน่ห์")
elif is_chinese_mode:
intro_rule = "⚠️ [INTRO RULE]: 第一行必须严格是 [00:00 - 00:02] 你好。" if current_is_first else "⚠️ [NO INTRO RULE]: 后续片段。严禁问候。"
prompt = (f"You are a professional movie recap narrator. Watch the video carefully and narrate the story in Simplified Chinese (中文).\n"
f"⚠️ CRITICAL INSTRUCTIONS:\n"
f"1. {zh_translation_rule}\n"
f"2. {zh_human_narrator_rule}\n"
f"3. {zh_consistency_rule}\n"
f"4. {zh_pacing}\n"
f"5. ⚠️ [VIDEO ANALYSIS]: 仔细观察视频中的人物动作、情感和关键事件。即使没有对话,也要连贯地讲述故事。\n"
f"6. {intro_rule}\n"
f"{memory_injection}{custom_prompt_text} MOVIE GENRE: {genre_prompt}\n"
f"FINAL INSTRUCTION: 每一行仅使用 [MM:SS - MM:SS] 格式。扮演一个有魅力的人类叙述者。")
else:
intro_rule = "⚠️ [INTRO RULE]: ပထမဆုံး စာကြောင်းကို [00:00 - 00:02] မင်္ဂလာပါ ဟု အချိန်အတိအကျဖြင့် နှုတ်ဆက်ပါ။" if current_is_first else "⚠️ [NO INTRO RULE]: နောက်ဆက်တွဲအပိုင်းဖြစ်၍ နှုတ်ဆက်စကားများ လုံးဝ မထည့်ပါနှင့်။"
prompt = (f"You are a professional movie recap narrator and storyteller. Watch the video carefully and narrate the story in Burmese Language (မြန်မာဘာသာစကား).\n"
f"⚠️ အရေးကြီးဆုံး ညွှန်ကြားချက်များ:\n"
f"၁။ {my_translation_rule}\n"
f"၂။ {my_human_narrator_rule}\n"
f"၃။ {my_consistency_rule}\n"
f"၄။ {my_pacing}\n"
f"၅။ ⚠️ [VIDEO ANALYSIS - CRITICAL]: ဗီဒီယိုထဲရှိ လူများ၏ လှုပ်ရှားမှုများ၊ ခံစားချက်များ၊ အရေးကြီးသော အဖြစ်အပျက်များကို သေချာကြည့်ရှုပြီး ပရိသတ်ကို ဇာတ်လမ်းပြန်ပြောပြသည့် (Movie Recap) ပုံစံဖြင့် ရေးပါ။ စကားပြောသံမရှိလျှင်လည်း ဇာတ်လမ်း အဆက်မပြတ်သွားစေရန် မြင်ကွင်းကို ဇာတ်ကြောင်းပြန်ပြောပြပါ။\n"
f"၆။ {intro_rule}\n"
f"{memory_injection}{custom_prompt_text} MOVIE GENRE: {genre_prompt}\n"
f"FINAL INSTRUCTION: စာကြောင်းတိုင်းအတွက် [MM:SS - MM:SS] ပုံစံကိုသာ မဖြစ်မနေ သုံးပါ။ လူသားစစ်စစ်ကဲ့သို့ ဆွဲဆောင်မှုရှိစွာ ဇာတ်လမ်းပြောပါ။")
keys_pool = get_api_keys_pool()
max_retries = max(5, len(keys_pool) + 2)
transcript = ""
jobs[job_id]["progress_msg"] = f"{msg_prefix}AI မှ ဗီဒီယိုကို ကြည့်ရှု၍ ဇာတ်လမ်းစဉ်းစားနေပါသည်..."
for attempt in range(max_retries):
try:
await asyncio.sleep(1)
active_key = get_working_api_key(user_api_key)
if not active_key: raise Exception("API Key Missing!")
genai.configure(api_key=active_key)
if video_file_ai:
try: await asyncio.to_thread(genai.delete_file, video_file_ai.name)
except: pass
# 💡 [THE FIX] ai_vid (ချုံ့ထားသော ဗီဒီယို) ကိုသာ AI သို့ ပေးပို့ပါမည်
video_file_ai = await asyncio.wait_for(asyncio.to_thread(genai.upload_file, path=ai_vid), timeout=120)
await asyncio.sleep(3)
for _ in range(6):
video_file_ai = await asyncio.wait_for(asyncio.to_thread(genai.get_file, video_file_ai.name), timeout=20)
if video_file_ai.state.name != "PROCESSING": break
await asyncio.sleep(3)
try:
model = genai.GenerativeModel("models/gemini-3.1-flash")
response = await asyncio.wait_for(asyncio.to_thread(model.generate_content, [prompt, video_file_ai]), timeout=120)
transcript = response.text.strip()
except Exception as e1:
print(f"⚠️ Model 3.1 failed: {e1}. Trying Model 2.5...", flush=True)
model = genai.GenerativeModel("models/gemini-2.5-flash")
response = await asyncio.wait_for(asyncio.to_thread(model.generate_content, [prompt, video_file_ai]), timeout=120)
transcript = response.text.strip()
if c_idx == 0 and transcript and not is_remerge:
try:
summary_prompt = f"From the following recap, list main characters, key locations, and primary plot points in a single paragraph. This is for memory.\n\nTranscript:\n{transcript}"
summary_response = await asyncio.to_thread(model.generate_content, summary_prompt)
character_sheet_memory = summary_response.text.strip()
except: pass
matches = list(re.finditer(r'\[(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\s*-\s*(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\]\s*([^\[]+)', transcript))
if not matches and transcript.strip(): raise Exception("Timestamp Format Error")
break
except Exception as e:
if attempt == max_retries - 1: raise e
await asyncio.sleep(2)
matches = list(re.finditer(r'\[(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\s*-\s*(\d{1,2}:\d{2}(?::\d{2})?(?:\.\d+)?)\]\s*([^\[]+)', transcript))
def time_to_ms(time_str):
parts = time_str.split(':')
if len(parts) == 2: return int((int(parts[0])*60) * 1000 + float(parts[1])*1000)
elif len(parts) == 3: return int((int(parts[0])*3600 + int(parts[1])*60) * 1000 + float(parts[2])*1000)
return 0
for m in matches:
start_str = m.group(1); end_str = m.group(2); raw_txt = m.group(3).strip()
clean_txt = re.sub(r'MOVIE\s*GENRE\s*:?', '', raw_txt, flags=re.IGNORECASE)
if is_eng_mode or is_thai_mode: clean_txt = re.sub(r'[\t\n\r]+', ' ', clean_txt)
elif is_chinese_mode: clean_txt = re.sub(r'\s+', '', clean_txt)
else:
clean_txt = re.sub(r'[\u0E00-\u0E7F]+', '', clean_txt)
clean_txt = re.sub(r'[\u4E00-\u9FFF]+', '', clean_txt)
clean_txt = re.sub(r'[\t\n\r]+', ' ', clean_txt)
if clean_txt.strip():
ms_start = time_to_ms(start_str); ms_end = time_to_ms(end_str)
if ms_end <= ms_start: ms_end = ms_start + 1500
parsed_lines.append({"time_ms": int(ms_start), "orig_end_ms": int(ms_end), "text": f"[{start_str} - {end_str}] {clean_txt.strip()}"})
if not parsed_lines:
clean_txt = re.sub(r'MOVIE\s*GENRE\s*:?', '', transcript, flags=re.IGNORECASE)
clean_txt = re.sub(r'\[.*?\]|\d{1,2}:\d{2}', '', clean_txt)
parsed_lines.append({"time_ms": 0, "orig_end_ms": 2000, "text": clean_txt.strip()})
chunk_data["parsed_lines"] = parsed_lines
chunk_data["transcript"] = " ".join([l["text"] for l in parsed_lines])
chunk_text = " ".join([l["text"] for l in parsed_lines])
full_transcript_list.append(chunk_text)
if parsed_lines:
last_few_lines = [re.sub(r'\[.*?\]', '', l["text"]).strip() for l in parsed_lines[-4:]]
previous_chunk_memory = " ".join(last_few_lines).strip()
if not is_remerge:
with open(chunk_txt_file, "w", encoding="utf-8") as f:
f.write(chunk_text)
# --- 4. Generate TTS & Build Video ---
if not is_remerge:
jobs[job_id]["status"] = "generating_audio"
jobs[job_id]["progress_msg"] = f"{msg_prefix}အသံနှင့် Timestamps ပေါင်းစပ်နေပါသည်..."
dynamic_filter_str = ""
tts_idx = -1
if voice_toggle == "on":
if v_check.startswith("gemini_"):
g_voice = VOICE_MAP.get(v_check, "Aoede"); edge_voice = g_voice; gemini_voice = g_voice
else:
cfg = DUAL_VOICE_MAP.get(v_check, DUAL_VOICE_MAP["nilar"])
edge_voice = cfg["edge"]; gemini_voice = cfg["gemini"]
tasks = []
for idx, line in enumerate(parsed_lines):
c_id = f"{original_job_id}_{c_idx}" if is_remerge else None
tasks.append(process_single_line_tts(idx, line["text"], line["time_ms"], line.get("orig_end_ms", line["time_ms"]+2000), f"{job_id}_{c_idx}", v_check, edge_voice, gemini_voice, user_api_key, pitch, lang_key, tts_speed, c_id))
chunk_results = await asyncio.gather(*tasks)
chunk_results = sorted([r for r in chunk_results if r is not None], key=lambda x: x["time_ms"])
if chunk_results:
dyn_filters = []
concat_labels = ""
seg_idx = 0
current_orig_time = 0.0
current_out_time = 0.0
safe_v_dur = float(c_v_dur)
try:
speed_math_map = {"1.0": 1.20, "1.15": 1.25, "1.25": 1.30, "1.30": 1.35}
tts_speed_float = speed_math_map.get(tts_speed, 1.20)
except:
tts_speed_float = 1.20
for i, res in enumerate(chunk_results):
orig_start = min(res['time_ms'] / 1000.0, safe_v_dur)
if orig_start < current_orig_time:
orig_start = current_orig_time
orig_end = min(res.get('orig_end_ms', res['time_ms'] + 1500) / 1000.0, safe_v_dur)
if orig_end <= orig_start: orig_end = min(orig_start + 1.0, safe_v_dur)
tts_dur = res['actual_dur'] / 1000.0
orig_dur = orig_end - orig_start
if orig_dur < 0.1: orig_dur = 0.1
# Gap Processing
if orig_start > current_orig_time:
gap_dur = orig_start - current_orig_time
if gap_dur > 0.3:
if jumpcut_mode == "on" and gap_dur > 1.5:
keep_gap = 0.5
gap_start = orig_start - keep_gap
if gap_start < current_orig_time: gap_start = current_orig_time
lbl = f"[vdyn_{seg_idx}]"
dyn_filters.append(f"[0:v]trim=start={gap_start}:end={orig_start},setpts=PTS-STARTPTS,fps=24{lbl}")
current_out_time += keep_gap
concat_labels += lbl
seg_idx += 1
else:
lbl = f"[vdyn_{seg_idx}]"
dyn_filters.append(f"[0:v]trim=start={current_orig_time}:end={orig_start},setpts=PTS-STARTPTS,fps=24{lbl}")
current_out_time += gap_dur
concat_labels += lbl
seg_idx += 1
# Speech Stretching
stretch_factor = tts_dur / orig_dur
if stretch_factor < 0.95:
stretch_factor = 1.0
elif stretch_factor > 2.2:
stretch_factor = 2.2
actual_vid_dur = orig_dur * stretch_factor
lbl_sp = f"[vdyn_{seg_idx}]"
if stretch_factor > 1.05:
dyn_filters.append(f"[0:v]trim=start={orig_start}:end={orig_end},setpts=({stretch_factor})*(PTS-STARTPTS),minterpolate=fps=24:mi_mode=blend{lbl_sp}")
else:
dyn_filters.append(f"[0:v]trim=start={orig_start}:end={orig_end},setpts=({stretch_factor})*(PTS-STARTPTS),fps=24{lbl_sp}")
res['final_start_ms'] = int(current_out_time * 1000)
current_out_time += actual_vid_dur
concat_labels += lbl_sp
seg_idx += 1
current_orig_time = orig_end
# Tail gap compression
if current_orig_time < safe_v_dur:
gap_dur = safe_v_dur - current_orig_time
if gap_dur > 0.05:
if jumpcut_mode == "on" and gap_dur > 1.5:
keep_gap = 0.5
gap_end = current_orig_time + keep_gap
lbl = f"[vdyn_{seg_idx}]"
dyn_filters.append(f"[0:v]trim=start={current_orig_time}:end={gap_end},setpts=PTS-STARTPTS,fps=24{lbl}")
current_out_time += keep_gap
else:
lbl = f"[vdyn_{seg_idx}]"
dyn_filters.append(f"[0:v]trim=start={current_orig_time}:end={safe_v_dur},setpts=PTS-STARTPTS,fps=24{lbl}")
current_out_time += gap_dur
concat_labels += lbl
seg_idx += 1
c_v_dur = current_out_time
if seg_idx > 0:
dyn_filters.append(f"{concat_labels}concat=n={seg_idx}:v=1:a=0[v_concat]")
else:
dyn_filters.append(f"[0:v]copy[v_concat]")
if resolution == "720p":
scale_filter = "scale=-2:720,setsar=1"
dyn_filters.append(f"[v_concat]{scale_filter}[v_dyn_stretched]")
else:
dyn_filters.append(f"[v_concat]copy[v_dyn_stretched]")
dynamic_filter_str = ";".join(dyn_filters) + ";"
mix_cmd = ["ffmpeg", "-y"]; filter_str = ""; mix_labels = ""; valid_chunks = 0
for i, res in enumerate(chunk_results):
start_ms = res.get("final_start_ms", 0)
mix_cmd.extend(["-i", res['path']])
filter_str += f"[{valid_chunks}:a]aformat=sample_rates=24000:channel_layouts=mono,adelay={int(start_ms)}|{int(start_ms)}[d{valid_chunks}];"
mix_labels += f"[d{valid_chunks}]"; valid_chunks += 1
if valid_chunks > 0:
if valid_chunks == 1: filter_str += f"{mix_labels}anull[a_main]"
else: filter_str += f"{mix_labels}amix=inputs={valid_chunks}:normalize=0[a_main]"
mix_cmd.extend(["-filter_complex", filter_str, "-map", "[a_main]", "-c:a", "pcm_s16le", tts_audio])
await run_cmd_async(mix_cmd)
tts_idx = 1
else:
dynamic_filter_str = "[0:v]copy[v_dyn_stretched];"
jobs[job_id]["status"] = "rendering"
jobs[job_id]["progress_msg"] = f"{msg_prefix}ဗီဒီယို အချောသတ်နေပါသည်..."
s_y_pos = float(sub_y) / 100.0 if sub_y else 0.85
s_size = int(sub_size) * 3 if sub_size else 48
s_color = str(sub_color).strip() if sub_color else ""
if s_color.startswith("&H"): ass_color = s_color
else:
s_color = s_color.lstrip('#')
if len(s_color) == 6: ass_color = f"&H00{s_color[4:6]}{s_color[2:4]}{s_color[0:2]}&"
else: ass_color = "&H0000FFFF&"
base_res = 720 if resolution == "720p" else 1080
if ratio == "9:16":
target_w = base_res
target_h = int(target_w * 16 / 9)
elif ratio == "1:1":
target_w = base_res
target_h = base_res
elif ratio == "4:5":
target_w = base_res
target_h = int(target_w * 5 / 4)
elif ratio == "4:3":
target_h = base_res
target_w = int(target_h * 4 / 3)
else: # 16:9
target_h = base_res
target_w = int(target_h * 16 / 9)
target_w = (target_w // 2) * 2
target_h = (target_h // 2) * 2
s_size = int(s_size * (target_h / 720.0))
m_y_val = float(mask_y) / 100.0 if mask_y else 0.80
m_h_val = float(mask_h) / 100.0 if mask_h else 0.15
m_w_val = float(mask_w) / 100.0 if mask_w else 0.50
crop_y = int(target_h * m_y_val); crop_h = int(target_h * m_h_val); crop_w = int(target_w * m_w_val)
crop_x = int(target_w * (1.0 - m_w_val) / 2)
crop_w = max(2, (crop_w // 2) * 2); crop_h = max(2, (crop_h // 2) * 2)
crop_x = (crop_x // 2) * 2; crop_y = (crop_y // 2) * 2
if crop_y + crop_h > target_h: crop_h = ((target_h - crop_y) // 2) * 2
if crop_x + crop_w > target_w: crop_w = ((target_w - crop_x) // 2) * 2
font_name = "Zawgyi-One"
if is_thai_mode: font_name = "Kanit"
elif is_chinese_mode: font_name = "Noto Sans SC"
elif is_eng_mode: font_name = "Kanit"
with open(sub_file, "w", encoding="utf-8") as f:
f.write(f"[Script Info]\nScriptType: v4.00+\nPlayResX: {target_w}\nPlayResY: {target_h}\nWrapStyle: 2\nScaledBorderAndShadow: yes\n\n[V4+ Styles]\n")
f.write("Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding\n")
f.write(f"Style: Default,{font_name},{s_size},{ass_color},&H000000FF,&H00000000,&H99000000,-1,0,0,0,100,100,0,0,3,4,0,5,10,10,0,1\n\n[Events]\n")
f.write("Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\n")
pos_x = target_w // 2
pos_y = int(target_h * s_y_pos) + (int(target_h * m_h_val) // 2)
def fmt_a(ms):
h, r = divmod(ms, 3600000); m, r = divmod(r, 60000); s, ml = divmod(r, 1000)
return f"{int(h)}:{int(m):02d}:{int(s):02d}.{int(ml/10):02d}"
if voice_toggle == "on" and 'chunk_results' in locals() and chunk_results:
for res in chunk_results:
sync_start_ms = res.get("final_start_ms", 0); sync_dur_ms = res.get("actual_dur", 0); clean_text = res.get("clean_text", "")
if not clean_text: continue
if not is_thai_mode and not is_chinese_mode and not is_eng_mode:
try: clean_text = rabbit.Rabbit.uni2zg(clean_text)
except: pass
if is_chinese_mode:
zh_tokens = re.findall(r"[a-zA-Z0-9]+|[^\sa-zA-Z0-9]", clean_text); short_chunks = []; temp_str = ""; count = 0
for tk in zh_tokens:
temp_str += tk; count += 1
if count >= 15: short_chunks.append(temp_str.strip()); temp_str = ""; count = 0
if temp_str: short_chunks.append(temp_str.strip())
elif is_thai_mode:
thai_chars = re.findall(r"[\u0E00-\u0E7F](?:[\u0E31-\u0E3A\u0E47-\u0E4E])*|[^\u0E00-\u0E7F\s]+", clean_text); short_chunks = []; temp_str = ""; count = 0
for char in thai_chars:
temp_str += char; count += 1
if count >= 20: short_chunks.append(temp_str.strip()); temp_str = ""; count = 0
if temp_str: short_chunks.append(temp_str.strip())
elif is_eng_mode: short_chunks = textwrap.wrap(clean_text, width=35)
else:
max_syllables = 12 if ratio in ["9:16", "4:5"] else 22
base_chars = r'\u1000-\u102A\u103F\u1040-\u1049\u104E-\u1055\u105A-\u105D\u1061\u1065\u1066\u106E-\u1070\u1075-\u1081\u108E\u1090-\u1099\u109E\u109F'
pattern = f"(?:[{base_chars}][^{base_chars}\s]*)|[^\u1000-\u109F\s]+"
words = clean_text.split()
short_chunks = []
temp_str = ""
for w in words:
if not w: continue
syllables = re.findall(pattern, w)
if not syllables: syllables = [w]
current_syl_count = len(re.findall(pattern, temp_str))
if current_syl_count + len(syllables) > max_syllables:
if temp_str: short_chunks.append(temp_str.strip())
if len(syllables) > max_syllables:
for i in range(0, len(syllables), max_syllables):
short_chunks.append("".join(syllables[i:i+max_syllables]))
temp_str = ""
else:
temp_str = w + " "
else:
temp_str += w + " "
if temp_str:
short_chunks.append(temp_str.strip())
total_chars = sum(len(c) for c in short_chunks)
chunk_start_ms = sync_start_ms
for s_chunk in short_chunks:
c_dur = (len(s_chunk) / max(1, total_chars)) * sync_dur_ms
chunk_end_ms = chunk_start_ms + c_dur
f.write(f"Dialogue: 0,{fmt_a(chunk_start_ms)},{fmt_a(chunk_end_ms)},Default,,0,0,0,,{{\\pos({pos_x},{pos_y})}}{s_chunk}\n")
chunk_start_ms += c_dur
else:
for line_data in parsed_lines:
current_start = line_data.get("time_ms", 0); current_end = line_data.get("orig_end_ms", current_start + 2000)
clean_text = re.sub(r'\[.*?\]', '', line_data.get("text", "")).strip()
if not clean_text: continue
if not is_thai_mode and not is_chinese_mode and not is_eng_mode:
try: clean_text = rabbit.Rabbit.uni2zg(clean_text)
except: pass
if is_chinese_mode:
zh_tokens = re.findall(r"[a-zA-Z0-9]+|[^\sa-zA-Z0-9]", clean_text); short_chunks = []; temp_str = ""; count = 0
for tk in zh_tokens:
temp_str += tk; count += 1
if count >= 10: short_chunks.append(temp_str.strip()); temp_str = ""; count = 0
if temp_str: short_chunks.append(temp_str.strip())
elif is_thai_mode:
thai_chars = re.findall(r"[\u0E00-\u0E7F](?:[\u0E31-\u0E3A\u0E47-\u0E4E])*|[^\u0E00-\u0E7F\s]+", clean_text); short_chunks = []; temp_str = ""; count = 0
for char in thai_chars:
temp_str += char; count += 1
if count >= 15: short_chunks.append(temp_str.strip()); temp_str = ""; count = 0
if temp_str: short_chunks.append(temp_str.strip())
elif is_eng_mode: short_chunks = textwrap.wrap(clean_text, width=25)
else:
max_syllables = 12 if ratio in ["9:16", "4:5"] else 22
base_chars = r'\u1000-\u102A\u103F\u1040-\u1049\u104E-\u1055\u105A-\u105D\u1061\u1065\u1066\u106E-\u1070\u1075-\u1081\u108E\u1090-\u1099\u109E\u109F'
pattern = f"(?:[{base_chars}][^{base_chars}\s]*)|[^\u1000-\u109F\s]+"
words = clean_text.split()
short_chunks = []
temp_str = ""
for w in words:
if not w: continue
syllables = re.findall(pattern, w)
if not syllables: syllables = [w]
current_syl_count = len(re.findall(pattern, temp_str))
if current_syl_count + len(syllables) > max_syllables:
if temp_str: short_chunks.append(temp_str.strip())
if len(syllables) > max_syllables:
for i in range(0, len(syllables), max_syllables):
short_chunks.append("".join(syllables[i:i+max_syllables]))
temp_str = ""
else:
temp_str = w + " "
else:
temp_str += w + " "
if temp_str:
short_chunks.append(temp_str.strip())
total_chars = sum(len(c) for c in short_chunks)
chunk_start_ms = current_start
for s_chunk in short_chunks:
c_dur = (len(s_chunk) / max(1, total_chars)) * (current_end - current_start)
chunk_end_ms = chunk_start_ms + c_dur - 30
f.write(f"Dialogue: 0,{fmt_a(chunk_start_ms)},{fmt_a(chunk_end_ms)},Default,,0,0,0,,{{\\pos({pos_x},{pos_y})}}{s_chunk}\n")
chunk_start_ms += c_dur
cmd = ["ffmpeg", "-y", "-i", trim_vid]
input_idx = 1; wm_idx = bgm_idx = ov_idx = bp_idx = -1
if voice_toggle == "on" and os.path.exists(tts_audio):
cmd.extend(["-itsoffset", "0", "-i", tts_audio]); tts_idx = input_idx; input_idx += 1
if logo_toggle == "on" and wm_path and os.path.exists(wm_path):
cmd.extend(["-i", wm_path]); wm_idx = input_idx; input_idx += 1
if bgm_toggle == "on" and bgm_path and os.path.exists(bgm_path):
cmd.extend(["-stream_loop", "-1", "-i", bgm_path]); bgm_idx = input_idx; input_idx += 1
if ov_path and os.path.exists(ov_path):
cmd.extend(["-stream_loop", "-1", "-i", ov_path]); ov_idx = input_idx; input_idx += 1
if bouncing_pip_toggle == "on" and bp_path and os.path.exists(bp_path):
cmd.extend(["-stream_loop", "-1", "-i", bp_path]); bp_idx = input_idx; input_idx += 1
pan_period = min(60.0, c_v_dur); pan_period = max(10.0, pan_period)
pre_filters = []
if bypass_mode in ["flip", "ultra"]: pre_filters.append("hflip")
if bypass_mode in ["color", "ultra"]: pre_filters.append("eq=brightness=0.03:saturation=1.1")
# 💡 [NEW ZOOM FIX] ရီဝေဝေဖြစ်သော Circular Pan အစား သဘာဝကျသော Cinematic Zoom-In ကို ပြောင်းလဲအသုံးပြုထားသည်
if bypass_mode in ["zoom", "ultra"]:
zoom_rate = 0.15 / max(1.0, c_v_dur)
pre_filters.append(f"zoompan=z='min(1.0+({zoom_rate:.5f}*time),1.15)':d=1:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':fps=24:s={target_w}x{target_h}")
pre_filter_str = ",".join(pre_filters) if pre_filters else "copy"
if special_effect == "split_screen" and ov_idx != -1:
v_complex = dynamic_filter_str + f"[v_dyn_stretched]{pre_filter_str},fps=24[fg_in];"
sz = float(overlay_size) / 100.0
split_h = int(target_h * sz); split_h = (split_h // 2) * 2
top_h = target_h - split_h; top_h = (top_h // 2) * 2
v_complex += f"[fg_in]scale='max({target_w},iw*({top_h}/ih))':'max({top_h},ih*({target_w}/iw))',crop={target_w}:{top_h}:(in_w-out_w)/2:(in_h-out_h)/2,setsar=1,format=yuv420p[main_scaled];"
v_complex += f"[{ov_idx}:v]fps=24,setpts=PTS-STARTPTS,scale='max({target_w},iw*({split_h}/ih))':'max({split_h},ih*({target_w}/iw))',crop={target_w}:{split_h}:(in_w-out_w)/2:(in_h-out_h)/2,setsar=1,format=yuv420p[bot_scaled];"
v_complex += f"[main_scaled][bot_scaled]vstack=shortest=1[v_base];"
current_v_node = "[v_base]"
else:
v_complex = dynamic_filter_str + f"[v_dyn_stretched]{pre_filter_str},fps=24,split[bg_in][fg_in];"
v_complex += f"[bg_in]scale=160:-1,boxblur=5:1,scale={target_w}:{target_h},setsar=1[bg];"
v_complex += f"[fg_in]scale='min({target_w},iw*({target_h}/ih))':'min({target_h},ih*({target_w}/iw))',setsar=1[fg_base];"
v_complex += f"[bg][fg_base]overlay=(W-w)/2:(H-h)/2[v_base];"
current_v_node = "[v_base]"
if mask_toggle == "on":
v_complex += (
f"{current_v_node}split=2[main_vid][glass_src];"
f"[glass_src]crop={crop_w}:{crop_h}:{crop_x}:{crop_y},boxblur=12:2,drawbox=x=0:y=0:w=iw:h=ih:color=white@0.2:t=fill[glass_eff];"
f"[main_vid][glass_eff]overlay=x={crop_x}:y={crop_y}[v_glass];"
)
current_v_node = "[v_glass]"
if marquee_toggle == "on" and marquee_text:
font_path = os.path.join(USER_FONT_DIR, "Zawgyi-One.ttf").replace('\\', '/')
safe_text = marquee_text.replace("'", "").replace(":", "\\:")
if not is_thai_mode and not is_chinese_mode and not is_eng_mode:
try: safe_text = rabbit.Rabbit.uni2zg(safe_text)
except: pass
mq_y = float(marquee_y) / 100.0 * target_h
v_complex += f"{current_v_node}drawtext=fontfile='{font_path}':text='{safe_text}':fontcolor=gold:fontsize=40:box=1:boxcolor=black@0.5:boxborderw=10:y={mq_y}:x='w-mod(t*150,w+tw)'[v_mq];"
current_v_node = "[v_mq]"
if wm_idx != -1:
opac = float(logo_opacity) / 100.0; sz = float(logo_size) / 100.0
pos_map = {"top-left": "10:10", "top-right": "W-w-10:10", "bottom-left": "10:H-h-10", "bottom-right": "W-w-10:H-h-10"}
if special_effect == "bouncing_logo": overlay_pos = "x='abs(sin(t*0.5))*(W-w)':y='abs(cos(t*0.3))*(H-h)'"
elif logo_pos == "custom":
lx = float(logo_custom_x) / 100.0 * target_w; ly = float(logo_custom_y) / 100.0 * target_h
overlay_pos = f"x={int(lx)}:y={int(ly)}"
else: overlay_pos = pos_map.get(logo_pos, "10:10")
w_logo = int(target_w * sz) // 2 * 2
if logo_shape == "circle": v_complex += f"[{wm_idx}:v]crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_logo}:{w_logo},setsar=1,format=rgba,geq=r='r(X,Y)':g='g(X,Y)':b='b(X,Y)':a='if(lt(pow(X-(W/2),2)+pow(Y-(H/2),2),pow(W/2,2)),255,0)',colorchannelmixer=aa={opac}[wm];{current_v_node}[wm]overlay={overlay_pos}[v_wm_out];"
else: v_complex += f"[{wm_idx}:v]crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_logo}:{w_logo},setsar=1,format=rgba,colorchannelmixer=aa={opac}[wm];{current_v_node}[wm]overlay={overlay_pos}[v_wm_out];"
current_v_node = "[v_wm_out]"
if sub_toggle == "on":
safe_sub_file = sub_file.replace('\\', '/').replace(':', '\\:')
v_complex += f"{current_v_node}subtitles='{safe_sub_file}'[v_sub];"
current_v_node = "[v_sub]"
if bouncing_pip_toggle == "on" and bp_idx != -1:
bp_sz = float(bouncing_pip_size) / 100.0; bp_opac = float(bouncing_pip_opacity) / 100.0 if bouncing_pip_opacity else 1.0
w_bp = int(target_w * bp_sz) // 2 * 2
if bouncing_pip_shape == "circle": v_complex += f"[{bp_idx}:v]fps=24,setpts=PTS-STARTPTS,crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_bp}:{w_bp},setsar=1,format=rgba,geq=r='r(X,Y)':g='g(X,Y)':b='b(X,Y)':a='if(lt(pow(X-(W/2),2)+pow(Y-(H/2),2),pow(W/2,2)),255,0)',colorchannelmixer=aa={bp_opac}[bp];{current_v_node}[bp]overlay=x='abs(sin(t*0.5))*(W-w)':y='abs(cos(t*0.3))*(H-h)':shortest=1[v_bp_out];"
else: v_complex += f"[{bp_idx}:v]fps=24,setpts=PTS-STARTPTS,crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_bp}:{w_bp},setsar=1,format=rgba,colorchannelmixer=aa={bp_opac}[bp];{current_v_node}[bp]overlay=x='abs(sin(t*0.5))*(W-w)':y='abs(cos(t*0.3))*(H-h)':shortest=1[v_bp_out];"
current_v_node = "[v_bp_out]"
if special_effect == "pip" and ov_idx != -1:
sz = float(overlay_size) / 100.0; w_pip = int(target_w * sz) // 2 * 2
pos_map = {"top-left": "10:10", "top-right": "W-w-10:10", "bottom-left": "10:H-h-10", "bottom-right": "W-w-10:H-h-10"}
if pip_pos == "custom":
px = float(pip_custom_x) / 100.0 * target_w; py = float(pip_custom_y) / 100.0 * target_h
overlay_pos = f"x={int(px)}:y={int(py)}"
else: overlay_pos = pos_map.get(pip_pos, "W-w-10:H-h-10")
if pip_shape == "circle": v_complex += f"[{ov_idx}:v]fps=24,setpts=PTS-STARTPTS,crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_pip}:{w_pip},setsar=1,format=rgba,geq=r='r(X,Y)':g='g(X,Y)':b='b(X,Y)':a='if(lt(pow(X-(W/2),2)+pow(Y-(H/2),2),pow(W/2,2)),255,0)'[pip];{current_v_node}[pip]overlay={overlay_pos}:shortest=1[v_pip_out];"
else: v_complex += f"[{ov_idx}:v]fps=24,setpts=PTS-STARTPTS,crop='min(in_w,in_h)':'min(in_w,in_h)',scale={w_pip}:{w_pip},setsar=1,format=yuva420p[pip];{current_v_node}[pip]overlay={overlay_pos}:shortest=1[v_pip_out];"
current_v_node = "[v_pip_out]"
v_complex += f"{current_v_node}copy[v_out];"
a_complex = ""
if tts_idx != -1:
vol_float = float(volume) / 100.0
if bgm_idx != -1:
bgm_vol_float = float(bgm_volume) / 100.0
a_complex = (f"[{tts_idx}:a]volume={vol_float}[a_tts];"
f"[{bgm_idx}:a]aresample=44100,sidechaincompress=threshold=0.02:ratio=30:attack=10:release=200,volume={bgm_vol_float}*5[a_bgm];"
f"[a_tts][a_bgm]amix=inputs=2:duration=first:dropout_transition=2:normalize=0[a_out]")
else: a_complex = f"[{tts_idx}:a]volume={vol_float}[a_out]"
audio_map = "[a_out]"
else:
if bgm_idx != -1:
bgm_vol_float = float(bgm_volume) / 100.0
a_complex = f"[0:a]volume=1.0[a_main];[{bgm_idx}:a]volume={bgm_vol_float}[a_bgm];[a_main][a_bgm]amix=inputs=2:duration=first:dropout_transition=2:normalize=0[a_out]"
audio_map = "[a_out]"
else: audio_map = "0:a"
filter_complex = v_complex + a_complex
cmd.extend(["-filter_complex", filter_complex, "-map", "[v_out]"])
cmd.extend([
"-map", f"{audio_map}", "-c:v", "libx264", "-loglevel", "error", "-r", "24", "-crf", "24",
"-preset", "ultrafast", "-profile:v", "baseline", "-level", "3.0", "-pix_fmt", "yuv420p",
"-movflags", "+faststart", "-c:a", "aac", "-b:a", "128k", "-ar", "44100",
"-t", str(c_v_dur), chunk_out_vid
])
await run_cmd_async(cmd)
completed_chunk_videos.append(chunk_out_vid)
if jobs.get(job_id, {}).get("status") == "cancelled": raise Exception("လုပ်ငန်းစဉ်ကို ရပ်ဆိုင်းလိုက်ပါသည်။")
jobs[job_id]["status"] = "merging"
jobs[job_id]["progress_msg"] = "အပိုင်းများ အားလုံးကို ပြန်လည်ပေါင်းစပ်နေပါသည်..."
temp_output_path = os.path.join(TMP_DIR, f"temp_final_{job_id}.mp4")
if len(completed_chunk_videos) > 1:
list_file = os.path.join(TMP_DIR, f"list_{job_id}.txt")
with open(list_file, "w", encoding="utf-8") as f:
for cv in completed_chunk_videos: f.write(f"file '{os.path.abspath(cv)}'\n")
await run_cmd_async(["ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", list_file, "-c", "copy", temp_output_path])
elif len(completed_chunk_videos) == 1:
shutil.copy(completed_chunk_videos[0], temp_output_path)
else:
raise Exception("ဗီဒီယိုဖန်တီးမှု မအောင်မြင်ပါ။")
if os.path.exists(temp_output_path):
shutil.move(temp_output_path, output_video_path)
total_processed_seconds = sum([min(c["duration"], v_dur_full - c["start_time"]) for c in chunks_info])
actual_mins = int(math.ceil(total_processed_seconds / 60.0))
if actual_mins < 1: actual_mins = 1
rate = 10 if "gemini" in voice_name.lower() else 5
if is_remerge:
final_cost = 0
else:
final_cost = actual_mins * rate
jobs[job_id]["status"] = "completed"
jobs[job_id]["video_url"] = f"/download/{job_id}"
jobs[job_id]["transcript"] = "\n\n".join(full_transcript_list)
jobs[job_id]["final_cost"] = final_cost
jobs[job_id]["v_dur_mins"] = actual_mins
jobs[job_id]["tts_type"] = "Gemini TTS" if "gemini" in voice_name.lower() else "Microsoft Edge TTS"
except Exception as e:
jobs[job_id]["status"] = "failed"
if is_remerge:
jobs[job_id]["error"] = f"အမြန်နှုန်းပြောင်းခြင်း မအောင်မြင်ပါ။ ({str(e)})"
else:
jobs[job_id]["error"] = str(e)
finally:
if video_file_ai:
try: await asyncio.to_thread(genai.delete_file, video_file_ai.name)
except: pass
gc.collect()
@app.post("/adjust-speed")
async def adjust_speed(
background_tasks: BackgroundTasks,
job_id: str = Form(...),
tts_speed: str = Form(...),
x_secure_token: str = Header(None)
):
try:
is_valid_token = False
if x_secure_token == BACKEND_SECRET: is_valid_token = True
elif x_secure_token in upload_tokens:
if time.time() <= upload_tokens[x_secure_token]:
is_valid_token = True
del upload_tokens[x_secure_token]
if not is_valid_token: return JSONResponse(status_code=403, content={"error": "Token မမှန်ကန်ပါ။"})
if job_id not in jobs: return JSONResponse(status_code=404, content={"error": "မူရင်းဗီဒီယို မှတ်ဉာဏ် မရှိတော့ပါ။"})
old_job = jobs[job_id]
real_original_id = old_job.get("original_job_id") or job_id
source_video = os.path.join(TMP_DIR, f"out_{real_original_id}.mp4")
if not os.path.exists(source_video):
return JSONResponse(status_code=404, content={"error": "ဆာဗာပေါ်တွင် ဗီဒီယိုအဟောင်း မရှိတော့ပါ။ (အချိန်ကျော်လွန်သွားပါသည်)"})
new_job_id = str(uuid.uuid4())
jobs[new_job_id] = {
"status": "pending",
"created_at": time.time(),
"user_id": old_job.get("user_id", "default_user"),
"transcript": old_job.get("transcript", ""),
"original_job_id": real_original_id,
"final_cost": 0,
"tts_type": "Fast Re-edit Mode",
"v_dur_mins": 0
}
async def fast_ffmpeg_speed_task(n_id, src_vid, speed_str):
jobs[n_id]["status"] = "rendering"
jobs[n_id]["progress_msg"] = "ဗီဒီယို တစ်ခုလုံး၏ အမြန်နှုန်းကို ပြောင်းလဲနေပါသည်... (စက္ကန့်ပိုင်းသာ ကြာပါမည်)"
out_vid = os.path.join(TMP_DIR, f"out_{n_id}.mp4")
try:
s_val = float(speed_str)
if s_val <= 0: s_val = 1.15
v_pts = 1.0 / s_val
cmd = [
"ffmpeg", "-y", "-i", src_vid,
"-filter_complex", f"[0:v]setpts={v_pts}*PTS[v];[0:a]atempo={s_val}[a]",
"-map", "[v]", "-map", "[a]",
"-c:v", "libx264", "-preset", "ultrafast", "-crf", "24",
out_vid
]
await run_cmd_async(cmd)
jobs[n_id]["status"] = "completed"
jobs[n_id]["video_url"] = f"/download/{n_id}"
except Exception as e:
jobs[n_id]["status"] = "failed"
jobs[n_id]["error"] = f"အမြန်နှုန်းပြောင်းလဲခြင်း မအောင်မြင်ပါ။ ({str(e)})"
task = asyncio.create_task(fast_ffmpeg_speed_task(new_job_id, source_video, tts_speed))
running_tasks[new_job_id] = task
active_user_jobs[jobs[new_job_id]["user_id"]] = new_job_id
return JSONResponse(content={"success": True, "job_id": new_job_id})
except Exception as e:
return JSONResponse(status_code=400, content={"error": f"API Error: {str(e)}"})
class MergePayload(BaseModel): job_ids: List[str]
@app.post("/merge-videos")
async def merge_cart_videos(payload: MergePayload):
try:
if not payload.job_ids or len(payload.job_ids) < 2:
return JSONResponse(status_code=400, content={"error": "အနည်းဆုံး ဗီဒီယို (၂) ခု လိုအပ်ပါသည်။"})
valid_files = [os.path.join(TMP_DIR, f"out_{j_id}.mp4") for j_id in payload.job_ids if os.path.exists(os.path.join(TMP_DIR, f"out_{j_id}.mp4"))]
if len(valid_files) < 2:
return JSONResponse(status_code=400, content={"error": "ဗီဒီယိုများ ရှာမတွေ့ပါ။ (အချိန်ကျော်လွန်၍ ဖျက်သွားခြင်း ဖြစ်နိုင်ပါသည်)"})
m_id = str(uuid.uuid4())
o_path = os.path.join(TMP_DIR, f"out_{m_id}.mp4")
l_path = os.path.join(TMP_DIR, f"list_{m_id}.txt")
merged_transcripts = []
for j_id in payload.job_ids:
if j_id in jobs and jobs[j_id].get("transcript"):
merged_transcripts.append(jobs[j_id]["transcript"])
else:
for f in os.listdir(TMP_DIR):
if f.startswith("chunk_txt_") and f.endswith(".txt"):
try:
with open(os.path.join(TMP_DIR, f), "r", encoding="utf-8") as tf:
merged_transcripts.append(tf.read())
break
except: pass
with open(l_path, "w", encoding="utf-8") as f:
for vf in valid_files: f.write(f"file '{os.path.abspath(vf)}'\n")
await run_cmd_async(["ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", l_path, "-c", "copy", o_path])
final_merged_text = "\n\n=== ဇာတ်လမ်းအစအဆုံး စာသားများ ===\n\n" + "\n\n".join(merged_transcripts) if merged_transcripts else "✅ အပိုင်းများ အားလုံး ပေါင်းစပ်ပြီးပါပြီ။"
jobs[m_id] = {
"status": "completed",
"created_at": time.time(),
"transcript": final_merged_text
}
for vf in valid_files:
try: os.remove(vf)
except: pass
return JSONResponse(content={"job_id": m_id, "transcript": final_merged_text})
except Exception as e:
return JSONResponse(status_code=500, content={"error": f"Server Error: {str(e)}"})
def write_chunk_sync(chunk_file, path, mode):
with open(path, mode) as f: shutil.copyfileobj(chunk_file, f)
@app.post("/upload-chunk")
async def upload_chunk(chunk: UploadFile = File(...), file_hash: str = Form(...), chunk_index: int = Form(...)):
emergency_disk_cleanup()
chunk_path = os.path.join(TMP_DIR, f"cache_{file_hash}.mp4")
mode = "ab" if chunk_index > 0 else "wb"
await asyncio.to_thread(write_chunk_sync, chunk.file, chunk_path, mode)
return JSONResponse(content={"success": True})
@app.post("/delete-cache")
async def delete_cache(file_hash: str = Form(...)):
try:
p = os.path.join(TMP_DIR, f"cache_{file_hash}.mp4")
if os.path.exists(p): os.remove(p)
gc.collect(); return JSONResponse(content={"success": True})
except: return JSONResponse(content={"success": False})
def save_upload_sync(upload_file, path):
with open(path, "wb") as f: shutil.copyfileobj(upload_file.file, f)
@app.post("/process-video")
async def process_video(
background_tasks: BackgroundTasks,
video: UploadFile = File(None), watermark: UploadFile = File(None), bgm: UploadFile = File(None), overlay_video: UploadFile = File(None),
voice: str = Form("thiha"), api_key: str = Form(""), ratio: str = Form("16:9"), resolution: str = Form("original"),
voice_toggle: str = Form("on"), hide_subtitle: str = Form("off"), volume: str = Form("100"), pitch: str = Form("0"),
bypass_mode: str = Form("none"), logo_pos: str = Form("top-left"), logo_size: str = Form("5"), logo_opacity: str = Form("100"),
custom_prompt: str = Form(""), genre: str = Form("original"), bgm_volume: str = Form("20"), target_chunk: str = Form("auto"), chunk_minutes: str = Form("2"),
user_id: str = Form("default_user"), x_secure_token: str = Header(None), mask_toggle: str = Form("off"), mask_y: str = Form("80"),
mask_h: str = Form("15"), mask_w: str = Form("50"), sub_toggle: str = Form("on"), sub_y: str = Form("85"), sub_size: str = Form("16"),
sub_color: str = Form("D4AF37"), file_hash: str = Form(""), tts_speed: str = Form("1.15"),
bgm_toggle: str = Form("on"), logo_toggle: str = Form("on"), logo_shape: str = Form("square"), logo_custom_x: str = Form("10"), logo_custom_y: str = Form("10"),
special_effect: str = Form("none"), marquee_toggle: str = Form("off"), marquee_text: str = Form(""), marquee_y: str = Form("90"), pip_shape: str = Form("square"), pip_pos: str = Form("bottom-right"), overlay_size: str = Form("15"),
pip_custom_x: str = Form("10"), pip_custom_y: str = Form("10"),
bouncing_pip_toggle: str = Form("off"), bouncing_pip_video: UploadFile = File(None), bouncing_pip_shape: str = Form("circle"), bouncing_pip_size: str = Form("20"), bouncing_pip_opacity: str = Form("100"),
jumpcut_mode: str = Form("off"),
force_recreate: str = Form("false"),
tab_mode: str = Form("translate")
):
try:
is_valid_token = False
if x_secure_token == BACKEND_SECRET: is_valid_token = True
elif x_secure_token in upload_tokens:
if time.time() <= upload_tokens[x_secure_token]:
is_valid_token = True; del upload_tokens[x_secure_token]
if not is_valid_token: return JSONResponse(status_code=403, content={"error": "Token မမှန်ကန်ပါ။"})
emergency_disk_cleanup()
job_id = str(uuid.uuid4())
working_video_path = ""
if user_id in active_user_jobs:
old_job = active_user_jobs[user_id]
if old_job in running_tasks:
running_tasks[old_job].cancel()
del running_tasks[old_job]
if old_job in jobs:
jobs[old_job]["status"] = "cancelled"
if file_hash:
cache_path = os.path.join(TMP_DIR, f"cache_{file_hash}.mp4")
if video and video.filename:
await asyncio.to_thread(save_upload_sync, video, cache_path)
working_video_path = cache_path
elif os.path.exists(cache_path): working_video_path = cache_path
else: return JSONResponse(status_code=400, content={"error": "ဗီဒီယိုဖိုင် ရှာမတွေ့ပါ။"})
else:
if not video or not video.filename: return JSONResponse(status_code=400, content={"error": "ဗီဒီယိုဖိုင် လိုအပ်ပါသည်။"})
working_video_path = os.path.join(TMP_DIR, f"temp_{job_id}_{video.filename}")
await asyncio.to_thread(save_upload_sync, video, working_video_path)
wm_path = ""; bgm_path = ""; ov_path = ""; bp_path = ""
if watermark and watermark.filename:
wm_path = os.path.join(TMP_DIR, f"wm_{job_id}_{watermark.filename}"); await asyncio.to_thread(save_upload_sync, watermark, wm_path)
if bgm and bgm.filename:
bgm_path = os.path.join(TMP_DIR, f"bgm_{job_id}_{bgm.filename}"); await asyncio.to_thread(save_upload_sync, bgm, bgm_path)
if overlay_video and overlay_video.filename:
ov_path = os.path.join(TMP_DIR, f"ov_{job_id}_{overlay_video.filename}"); await asyncio.to_thread(save_upload_sync, overlay_video, ov_path)
if bouncing_pip_video and bouncing_pip_video.filename:
bp_path = os.path.join(TMP_DIR, f"bp_{job_id}_{bouncing_pip_video.filename}"); await asyncio.to_thread(save_upload_sync, bouncing_pip_video, bp_path)
settings = {
"voice": voice, "api_key": api_key, "ratio": ratio, "resolution": resolution,
"voice_toggle": voice_toggle, "hide_subtitle": hide_subtitle, "volume": volume, "pitch": pitch,
"bypass_mode": bypass_mode, "logo_pos": logo_pos, "logo_size": logo_size, "logo_opacity": logo_opacity,
"custom_prompt": custom_prompt, "genre": genre, "bgm_volume": bgm_volume, "target_chunk": target_chunk, "chunk_minutes": chunk_minutes,
"mask_toggle": mask_toggle, "mask_y": mask_y, "mask_h": mask_h, "mask_w": mask_w,
"sub_toggle": sub_toggle, "sub_y": sub_y, "sub_size": sub_size, "sub_color": sub_color,
"tts_speed": tts_speed, "bgm_toggle": bgm_toggle, "logo_toggle": logo_toggle, "logo_shape": logo_shape,
"logo_custom_x": logo_custom_x, "logo_custom_y": logo_custom_y, "special_effect": special_effect,
"marquee_toggle": marquee_toggle, "marquee_text": marquee_text, "marquee_y": marquee_y,
"pip_shape": pip_shape, "pip_pos": pip_pos, "overlay_size": overlay_size,
"pip_custom_x": pip_custom_x, "pip_custom_y": pip_custom_y,
"bouncing_pip_toggle": bouncing_pip_toggle, "bouncing_pip_shape": bouncing_pip_shape,
"bouncing_pip_size": bouncing_pip_size, "bouncing_pip_opacity": bouncing_pip_opacity,
"jumpcut_mode": jumpcut_mode, "force_recreate": force_recreate, "tab_mode": tab_mode
}
jobs[job_id] = {
"status": "pending", "created_at": time.time(), "video_path": working_video_path, "transcript": "",
"user_id": user_id, "wm_path": wm_path, "bgm_path": bgm_path, "ov_path": ov_path, "bp_path": bp_path, "settings": settings
}
task = asyncio.create_task(
wrapper_process_task(
job_id, user_id, working_video_path, wm_path, bgm_path, ov_path, voice, api_key, ratio, resolution, voice_toggle, hide_subtitle, volume, pitch, bypass_mode, logo_pos, logo_size, logo_opacity, custom_prompt, genre, bgm_volume, target_chunk, chunk_minutes, mask_toggle, mask_y, mask_h, mask_w, sub_toggle, sub_y, sub_size, sub_color, tts_speed, bgm_toggle, logo_toggle, logo_shape, logo_custom_x, logo_custom_y, special_effect, marquee_toggle, marquee_text, marquee_y, pip_shape, pip_pos, overlay_size, pip_custom_x, pip_custom_y, bouncing_pip_toggle, bp_path, bouncing_pip_shape, bouncing_pip_size, bouncing_pip_opacity, jumpcut_mode, False, None, force_recreate, tab_mode
)
)
running_tasks[job_id] = task
active_user_jobs[user_id] = job_id
return JSONResponse(content={"success": True, "job_id": job_id})
except Exception as e: return JSONResponse(status_code=400, content={"error": str(e)})
@app.get("/preview-voice")
async def preview_voice(voice: str = "thiha"):
v_name = voice.lower().strip()
preview_path = os.path.join(TMP_DIR, f"preview_{v_name}.wav")
if os.path.exists(preview_path): return FileResponse(preview_path, media_type="audio/wav")
female_voices = ["nilar", "thai_female", "english_female", "chinese_female", "gemini_aoede", "gemini_kore", "gemini_leda", "gemini_autonoe", "gemini_callirrhoe", "gemini_despina", "gemini_erinome", "gemini_laomedela", "gemini_pulcherrima", "gemini_vindemiatrix"]
name_map = {"thiha": "သီဟ", "nilar": "နီလာ"}
display_name = name_map.get(v_name, v_name.replace("gemini_", "").capitalize())
if "english" in v_name: text = "Hello, I am an English voice. Nice to meet you."
elif "thai" in v_name: text = "สวัสดีค่ะ ดิฉันคือเสียงภาษาไทย" if v_name in female_voices else "สวัสดีครับ ผมคือเสียงภาษาไทย"
elif "chinese" in v_name: text = "你好,我是中文语音。 很高兴认识你。"
else: text = f"{'ကျွန်မ' if v_name in female_voices else 'ကျွန်တော်'}က {display_name} ဖြစ်ပါတယ်။ တွေ့ရတာ ဝမ်းသာပါတယ်။"
try:
if v_name.startswith("gemini_"):
api_key = get_working_api_key("")
if not api_key: raise Exception("No API Key")
await asyncio.to_thread(generate_gemini_tts_audio_single, text, VOICE_MAP.get(v_name, "Aoede"), api_key, preview_path)
else: await generate_tts_audio(text, DUAL_VOICE_MAP.get(v_name, DUAL_VOICE_MAP["thiha"])["edge"], "+0%", "+0Hz", preview_path)
if os.path.exists(preview_path): return FileResponse(preview_path, media_type="audio/wav")
except Exception as e: return JSONResponse(status_code=500, content={"error": str(e)})
return JSONResponse(status_code=404, content={"error": "Not generated"})
@app.post("/cancel/{job_id}")
async def cancel_job_process(job_id: str):
if job_id in running_tasks:
running_tasks[job_id].cancel()
del running_tasks[job_id]
if job_id in jobs:
jobs[job_id]["status"] = "cancelled"
gc.collect()
return JSONResponse(content={"success": True})
return JSONResponse(status_code=404, content={"error": "Not found"})
@app.post("/delete-old-job")
async def delete_old_job(job_id: str = Form(...)):
try:
if job_id in running_tasks:
running_tasks[job_id].cancel()
del running_tasks[job_id]
for f in os.listdir(TMP_DIR):
if job_id in f:
try: os.remove(os.path.join(TMP_DIR, f))
except: pass
if job_id in jobs: del jobs[job_id]
gc.collect(); return JSONResponse(content={"success": True})
except Exception as e: return JSONResponse(status_code=500, content={"error": str(e)})
@app.get("/status/{job_id}")
async def get_status(job_id: str):
job_data = jobs.get(job_id, {"error": "Not found"})
return JSONResponse(content=job_data)
@app.get("/download/{job_id}")
async def download_video(job_id: str):
v_path = os.path.join(TMP_DIR, f"out_{job_id}.mp4")
if jobs.get(job_id, {}).get("status") == "completed" and os.path.exists(v_path):
headers = {"Content-Disposition": f'attachment; filename="Nabule_Recap_Video_{job_id[:6]}.mp4"'}
return FileResponse(v_path, media_type="video/mp4", headers=headers)
return JSONResponse(content={"error": "Not ready"})
@app.get("/download-audio/{job_id}")
async def download_audio(job_id: str):
v_path = os.path.join(TMP_DIR, f"out_{job_id}.mp4")
a_path = os.path.join(TMP_DIR, f"audio_{job_id}.mp3")
if jobs.get(job_id, {}).get("status") == "completed" and os.path.exists(v_path):
if not os.path.exists(a_path):
await run_cmd_async(["ffmpeg", "-y", "-i", v_path, "-q:a", "0", "-map", "a", a_path])
if os.path.exists(a_path):
headers = {"Content-Disposition": f'attachment; filename="Nabule_Recap_Audio_{job_id[:6]}.mp3"'}
return FileResponse(a_path, media_type="audio/mpeg", headers=headers)
return JSONResponse(content={"error": "Audio Not ready"})
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=7860)
import os
import uuid
import time
import asyncio
import gc
import re
import requests
import base64
import wave
import edge_tts
import shutil
import textwrap
import math
import traceback
import subprocess
from contextlib import asynccontextmanager
from fastapi import FastAPI, UploadFile, File, Form, BackgroundTasks, Header
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import FileResponse, JSONResponse
import google.generativeai as genai
from pydantic import BaseModel
from typing import List
import rabbit
import logging
import hashlib
# =========================================================================
# 💡 [TECH MONSTAR FIX] Bot များ၏ 404 Error များကို Log တွင် ဖျောက်ထားရန်
# =========================================================================
class HideBotLogsFilter(logging.Filter):
def filter(self, record):
msg = record.getMessage()
if " 404 " in msg:
return False
return True
logging.getLogger("uvicorn.access").addFilter(HideBotLogsFilter())
TMP_DIR = "/tmp/app_io"
os.makedirs(TMP_DIR, exist_ok=True)
USER_FONT_DIR = os.path.expanduser("~/.fonts")
os.makedirs(USER_FONT_DIR, exist_ok=True)
LOCAL_FONTS = [
"Zawgyi-One.ttf",
"Kanit-Regular.ttf",
"NotoSansSC-Regular.ttf",
"NotoSansMyanmar-Regular.ttf"
]
for f_name in LOCAL_FONTS:
local_path = os.path.join(os.getcwd(), f_name)
dest_path = os.path.join(USER_FONT_DIR, f_name)
if os.path.exists(local_path):
if not os.path.exists(dest_path):
try: shutil.copy(local_path, dest_path)
except: pass
os.system("fc-cache -f -v")
jobs = {}
upload_tokens = {}
running_tasks = {}
active_user_jobs = {}
MAX_CONCURRENT_JOBS = 3
queue_semaphore = asyncio.Semaphore(MAX_CONCURRENT_JOBS)
tts_semaphore = asyncio.Semaphore(5)
api_key_counter = 0
BACKEND_SECRET = os.getenv("BACKEND_SECRET", "MY_SUPER_SECRET_TOKEN_123").strip()
VOICE_MAP = {
"thiha": "my-MM-ThihaNeural", "nilar": "my-MM-NilarNeural",
"thai_male": "th-TH-NiwatNeural", "thai_female": "th-TH-PremwadeeNeural",
"english_male": "en-US-BrianNeural", "english_female": "en-US-EmmaNeural",
"chinese_male": "zh-CN-YunxiNeural", "chinese_female": "zh-CN-XiaoxiaoNeural",
"gemini_aoede": "Aoede", "gemini_charon": "Charon", "gemini_kore": "Kore",
"gemini_fenrir": "Fenrir", "gemini_puck": "Puck", "gemini_zephyr": "Zephyr",
"gemini_leda": "Leda", "gemini_orus": "Orus", "gemini_autonoe": "Autonoe",
"gemini_callirrhoe": "Callirrhoe", "gemini_enceladus": "Enceladus", "gemini_iapetus": "Iapetus",
"gemini_umbriel": "Umbriel", "gemini_algieba": "Algieba", "gemini_despina": "Despina",
"gemini_erinome": "Erinome", "gemini_algenib": "Algenib", "gemini_rasalgethi": "Rasalgethi",
"gemini_laomedeia": "Laomedeia", "gemini_achernar": "Achernar", "gemini_alnilam": "Alnilam",
"gemini_schedar": "Schedar", "gemini_gacrux": "Gacrux", "gemini_pulcherrima": "Pulcherrima",
"gemini_achird": "Achird", "gemini_zubenelgenubi": "Zubenelgenubi", "gemini_vindemiatrix": "Vindemiatrix",
"gemini_sadachbia": "Sadachbia", "gemini_sadaltager": "Sadaltager", "gemini_sulafat": "Sulafat"
}
DUAL_VOICE_MAP = {
"thiha": {"edge": "my-MM-ThihaNeural", "gemini": "Charon"},
"nilar": {"edge": "my-MM-NilarNeural", "gemini": "Aoede"},
"thai_male": {"edge": "th-TH-NiwatNeural", "gemini": "Charon"},
"thai_female": {"edge": "th-TH-PremwadeeNeural", "gemini": "Aoede"},
"english_male": {"edge": "en-US-BrianNeural", "gemini": "Charon"},
"english_female": {"edge": "en-US-EmmaNeural", "gemini": "Aoede"},
"chinese_male": {"edge": "zh-CN-YunxiNeural", "gemini": "Charon"},
"chinese_female": {"edge": "zh-CN-XiaoxiaoNeural", "gemini": "Aoede"}
}