Update app.py

app.py

@@ -18,6 +18,8 @@ install_package("whisper", "openai-whisper")
 install_package("PIL", "Pillow") 
 install_package("numpy")          
 install_package("opencc", "opencc-python-reimplemented") 
+install_package("gradio")
+install_package("requests")
 
 # ==========================================
 # 1. 正常导入库
@@ -27,9 +29,7 @@ import re
 import time
 import json
 import requests
-import zipfile
 import queue
-import math
 import numpy as np
 from datetime import datetime
 from concurrent.futures import ThreadPoolExecutor, as_completed
@@ -59,9 +59,9 @@ VIDEOS_PER_BATCH = 10
 
 # 4. 系统配置
 BASE_OUTPUT_DIR = "project_thangka_final_pipeline" 
-VIDEO_WORKERS = 150      
-LLM_WORKERS = 5           
-VIDEO_TIMEOUT_SECONDS = 360 # 🔥 修改点：单个视频超时设定为 6分钟
+VIDEO_WORKERS = 150       
+LLM_WORKERS = 5            
+VIDEO_TIMEOUT_SECONDS = 360 # 🔥 单个视频超时设定为 6分钟
 
 # 5. 字幕配置
 FONT_URL = "https://github.com/googlefonts/noto-cjk/raw/main/Sans/OTF/TraditionalChinese/NotoSansCJKtc-Bold.otf"
@@ -327,10 +327,24 @@ class AudioAgent:
                 time.sleep(5)
         return None
 
-# 🔥 VideoAgent (优化超时逻辑)
+# 🔥 VideoAgent (API格式修正版：Multipart/form-data)
 class VideoAgent:
     def __init__(self):
-        self.headers = {"Authorization": f"Bearer {YUNWU_API_KEY}"}
+        # 1. 修正：只保留鉴权，requests会自动处理 multipart boundary
+        self.headers = {
+            "Authorization": f"Bearer {YUNWU_API_KEY}"
+        }
+
+    def ensure_dummy_image(self):
+        # 创建一个黑色占位图，用于满足 API 的 input_reference 必填项
+        dummy_path = "_temp_blank_ref.png"
+        if not os.path.exists(dummy_path):
+            try:
+                img = Image.new('RGB', (720, 1280), (0, 0, 0))
+                img.save(dummy_path)
+            except Exception:
+                pass
+        return dummy_path
 
     def worker_task(self, prompt, global_video_idx, log_queue, save_dir):
         save_path = os.path.join(save_dir, f"clip_{global_video_idx+1:04d}.mp4")
@@ -340,21 +354,40 @@ class VideoAgent:
         final_prompt = (f"{FORCE_STYLE_PREFIX} {prompt} {FORCE_MOTION_BOOSTER} {FORCE_STYLE_SUFFIX} {FORCE_NEGATIVE_PROMPT}")
         final_prompt = re.sub(r'\s+', ' ', final_prompt).replace(", ,", ",").replace("..", ".")
 
-        # 🔥 重试逻辑：失败直接重试
+        # 准备占位图
+        dummy_ref_path = self.ensure_dummy_image()
+
+        # 🔥 重试逻辑
         for attempt in range(5):
             try:
-                # 提交任务
                 create_url = f"{MERCHANT_BASE_URL}/v1/videos"
+                
+                # 2. 修正：所有字段必须是字符串 (data 部分)
                 form_data = {
                     "model": VEO_MODEL,             
                     "prompt": final_prompt,
-                    "seconds": VIDEO_DURATION_STR,  
+                    "seconds": str(VIDEO_DURATION_STR), 
                     "size": VIDEO_SIZE,             
-                    "watermark": "false"
+                    "watermark": "false" # API要求 string 'false'
                 }
                 
-                req = requests.post(create_url, headers=self.headers, data=form_data, timeout=60)
+                # 3. 修正：文件部分 (input_reference 必填)
+                # 使用 files 参数会自动触发 multipart/form-data
+                files_payload = {}
+                if os.path.exists(dummy_ref_path):
+                    files_payload["input_reference"] = (
+                        os.path.basename(dummy_ref_path), 
+                        open(dummy_ref_path, "rb"), 
+                        "image/png"
+                    )
+                
+                # 发送请求
+                req = requests.post(create_url, headers=self.headers, data=form_data, files=files_payload, timeout=60)
                 
+                # 手动关闭文件句柄（如果打开了的话，requests通常会自动处理，但为安全起见）
+                if "input_reference" in files_payload:
+                    files_payload["input_reference"][1].close()
+
                 if req.status_code != 200:
                     try: err_msg = req.json()
                     except: err_msg = req.text[:100]
@@ -374,10 +407,9 @@ class VideoAgent:
                 # 轮询查询状态
                 start_time = time.time()
                 while True:
-                    # 🔥 核心修改1：严格的 6分钟超时
                     if time.time() - start_time > VIDEO_TIMEOUT_SECONDS:
                          log_queue.put(f"⚠️ [超时] 视频{global_video_idx+1} 耗时超过6分钟，视为失败，正在重试...")
-                         break # 跳出 while，进入下一次 for attempt 循环（即重新提交）
+                         break 
 
                     query_url = f"{MERCHANT_BASE_URL}/v1/videos/{task_id}"
                     status_resp = requests.get(query_url, headers=self.headers, timeout=30)
@@ -396,7 +428,6 @@ class VideoAgent:
                             else:
                                 break
                         elif status == "failed":
-                            # 捕获敏感词错误
                             log_queue.put(f"❌ [生成失败] 视频{global_video_idx+1}: {status_data}")
                             break 
                         time.sleep(2) 
@@ -549,7 +580,7 @@ def process_pipeline(full_text, audio_file, gen_system_prompt):
         if new_msg: logs.append(f"[{datetime.now().strftime('%H:%M:%S')}] {new_msg}")
         return "\n".join(logs[-100:])
 
-    yield update_ui("🚀 系统启动 (Async Veo API 版)..."), None
+    yield update_ui("🚀 系统启动 (Async Veo API V3.2 Fix版)..."), None
     download_font_if_missing() 
     
     raw_segments = strict_text_splitter(full_text)
@@ -676,8 +707,8 @@ def process_pipeline(full_text, audio_file, gen_system_prompt):
 # ================= UI 界面 =================
 
 with gr.Blocks(title="Veo Pipeline Pro") as demo:
-    gr.Markdown("## 🎬 Veo 全自动音画工厂 (Async V3.2版)")
-    gr.Markdown("🌟 **特性**：实时渲染进度 | 6分钟超时重试 | 极速合成")
+    gr.Markdown("## 🎬 Veo 全自动音画工厂 (Multipart 修正版)")
+    gr.Markdown("🌟 **特性**：自动生成垫图 | 修复API格式 | 实时渲染")
     with gr.Row():
         with gr.Column(scale=3):
             in_text = gr.Textbox(label="1. 输入长文案", lines=8)