import os
import tempfile
import gradio as gr
import requests
import json
from loguru import logger
from typing import Optional, Tuple
import base64
import time
def call_gradio_client_api(video_file, text_prompt, guidance_scale, inference_steps, sample_nums):
"""调用官方Hugging Face Space的API"""
try:
from gradio_client import Client
logger.info("连接到官方 HunyuanVideo-Foley Space...")
# 连接到官方Space
client = Client("tencent/HunyuanVideo-Foley")
logger.info("发送推理请求...")
# 调用推理函数
result = client.predict(
video_file, # 视频文件
text_prompt, # 文本提示
guidance_scale, # CFG scale
inference_steps, # 推理步数
sample_nums, # 样本数量
api_name="/infer_single_video" # API端点名称
)
return result, "✅ 成功通过官方API生成音频!"
except Exception as e:
error_msg = str(e)
logger.error(f"Gradio Client API 调用失败: {error_msg}")
if "not found" in error_msg.lower():
return None, "❌ 官方Space的API端点未找到，可能接口已更改"
elif "connection" in error_msg.lower():
return None, "❌ 无法连接到官方Space，请检查网络"
elif "queue" in error_msg.lower():
return None, "⏳ 官方Space繁忙，请稍后重试"
else:
return None, f"❌ API调用错误: {error_msg}"
def call_huggingface_inference_api(video_file, text_prompt):
"""调用Hugging Face Inference API"""
try:
logger.info("尝试Hugging Face Inference API...")
API_URL = "https://api-inference.huggingface.co/models/tencent/HunyuanVideo-Foley"
# 读取视频文件
with open(video_file, "rb") as f:
video_data = f.read()
# 准备请求数据
headers = {
"Authorization": f"Bearer {os.environ.get('HF_TOKEN', '')}",
}
# 发送请求
response = requests.post(
API_URL,
headers=headers,
json={"inputs": {"video": base64.b64encode(video_data).decode(), "text": text_prompt}},
timeout=300
)
if response.status_code == 200:
# 保存结果
temp_dir = tempfile.mkdtemp()
audio_path = os.path.join(temp_dir, "generated_audio.wav")
with open(audio_path, 'wb') as f:
f.write(response.content)
return [audio_path], "✅ 通过Hugging Face API生成成功!"
else:
logger.error(f"HF API错误: {response.status_code}")
return None, f"❌ Hugging Face API返回错误: {response.status_code}"
except Exception as e:
logger.error(f"HF API调用失败: {str(e)}")
return None, f"❌ Hugging Face API调用失败: {str(e)}"
def try_alternative_apis(video_file, text_prompt):
"""尝试其他可能的API服务"""
# 1. 尝试通过公开的demo接口
try:
logger.info("尝试demo接口...")
# 这里可以尝试其他公开的API服务
# 比如Replicate、RunPod等
return None, "❌ 暂无可用的替代API服务"
except Exception as e:
return None, f"❌ 替代API调用失败: {str(e)}"
def smart_api_inference(video_file, text_prompt, guidance_scale=4.5, inference_steps=50, sample_nums=1):
"""智能API推理 - 尝试多种API调用方式"""
if video_file is None:
return [], "❌ 请上传视频文件!"
if not text_prompt:
text_prompt = "audio for this video"
logger.info(f"开始API推理: {video_file}")
logger.info(f"文本提示: {text_prompt}")
status_updates = []
# 方法1: 尝试Gradio Client (最可能成功)
status_updates.append("🔄 尝试连接官方Space API...")
try:
result, status = call_gradio_client_api(
video_file, text_prompt, guidance_scale, inference_steps, sample_nums
)
if result:
return result, "\n".join(status_updates + [status])
status_updates.append(status)
except ImportError:
status_updates.append("⚠️ gradio_client未安装，跳过官方API调用")
# 方法2: 尝试Hugging Face Inference API
status_updates.append("🔄 尝试Hugging Face Inference API...")
result, status = call_huggingface_inference_api(video_file, text_prompt)
if result:
return result, "\n".join(status_updates + [status])
status_updates.append(status)
# 方法3: 尝试其他API
status_updates.append("🔄 尝试替代API服务...")
result, status = try_alternative_apis(video_file, text_prompt)
status_updates.append(status)
# 所有方法都失败了
final_message = "\n".join(status_updates + [
"",
"💡 **解决方案建议:**",
"• 安装 gradio_client: pip install gradio_client",
"• 配置 HF_TOKEN 环境变量",
"• 等待官方Space负载降低",
"• 本地运行完整模型(需24GB+ RAM)",
"",
"🔗 **官方Space**: https://huggingface.co/spaces/tencent/HunyuanVideo-Foley"
])
return [], final_message
def create_real_api_interface():
"""创建真实API调用界面"""
css = """
.api-status {
background: #f0f8ff;
border: 2px solid #4169e1;
border-radius: 10px;
padding: 1rem;
margin: 1rem 0;
color: #191970;
}
"""
with gr.Blocks(css=css, title="HunyuanVideo-Foley API Client") as app:
# Header
gr.HTML("""
🎵 HunyuanVideo-Foley
API客户端 - 调用真实模型推理
""")
# API Status Notice
gr.HTML("""
🌐 真实API调用模式: 这个版本会通过API调用真实的HunyuanVideo-Foley模型进行推理。
优点: 真实AI音频生成，无需本地大内存
缺点: 依赖外部服务可用性，可能需要等待队列
""")
with gr.Row():
# 输入区域
with gr.Column(scale=1):
gr.Markdown("### 📹 视频输入")
video_input = gr.Video(
label="上传视频",
info="支持MP4、AVI、MOV等格式"
)
text_input = gr.Textbox(
label="🎯 音频描述",
placeholder="描述你想要的音频效果，例如：脚步声、雨声、车辆行驶等",
lines=3,
value="audio sound effects for this video"
)
with gr.Row():
guidance_scale = gr.Slider(
minimum=1.0,
maximum=10.0,
value=4.5,
step=0.1,
label="🎚️ CFG Scale"
)
inference_steps = gr.Slider(
minimum=10,
maximum=100,
value=50,
step=5,
label="⚡ 推理步数"
)
sample_nums = gr.Slider(
minimum=1,
maximum=6,
value=1,
step=1,
label="🎲 样本数量"
)
generate_btn = gr.Button(
"🎵 调用API生成音频",
variant="primary",
size="lg"
)
# 输出区域
with gr.Column(scale=1):
gr.Markdown("### 🎵 生成结果")
audio_outputs = []
for i in range(6):
audio_output = gr.Audio(
label=f"样本 {i+1}",
visible=(i == 0) # 只显示第一个
)
audio_outputs.append(audio_output)
status_output = gr.Textbox(
label="API状态",
interactive=False,
lines=10,
placeholder="等待API调用..."
)
# 事件处理
def process_with_api(video_file, text_prompt, guidance_scale, inference_steps, sample_nums):
# 调用API推理
results, status_msg = smart_api_inference(
video_file, text_prompt, guidance_scale, inference_steps, int(sample_nums)
)
# 准备输出
outputs = [None] * 6
visibilities = [False] * 6
if results and isinstance(results, list):
for i, result in enumerate(results[:6]):
outputs[i] = result
visibilities[i] = True
return outputs + visibilities + [status_msg]
# 动态显示样本数量
def update_visibility(sample_nums):
sample_nums = int(sample_nums)
return [gr.update(visible=(i < sample_nums)) for i in range(6)]
# 连接事件
sample_nums.change(
fn=update_visibility,
inputs=[sample_nums],
outputs=audio_outputs
)
generate_btn.click(
fn=process_with_api,
inputs=[video_input, text_input, guidance_scale, inference_steps, sample_nums],
outputs=audio_outputs + [gr.update(visible=(i < 6)) for i in range(6)] + [status_output]
)
# Footer
gr.HTML("""
""")
return app
if __name__ == "__main__":
# 设置日志
logger.remove()
logger.add(lambda msg: print(msg, end=''), level="INFO")
logger.info("启动 HunyuanVideo-Foley API 客户端...")
# 检查依赖
try:
import gradio_client
logger.info("✅ gradio_client 已安装")
except ImportError:
logger.warning("⚠️ gradio_client 未安装，API调用功能可能受限")
# 创建并启动应用
app = create_real_api_interface()
logger.info("API客户端就绪，准备调用真实模型...")
app.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
debug=False,
show_error=True
)