图像转视频生成器（v3.0.0 – 正常运行的 Veo-3.0 REST API）

状态: ✅ 已测试并正常运行
最后更新时间: 2026-04-11
示例: “金色藏式供品” → 约 60 秒生成 2.4 MB 视频

功能说明

使用 Google 的 Veo-3.0，基于任意图像生成一段 5 秒钟的电影级视频：

1. Gemini Vision（gemini-2.5-flash）分析图像，提取运动特征与场景描述
2. Veo-3.0 predictLongRunning 通过 REST API 异步生成视频
3. 轮询循环 持续监测操作状态，直至视频就绪（约 60–90 秒）
4. 下载 将 MP4 文件保存至 ~/.openclaw/workspace/tibetanProc/ 目录，文件名前缀为 yymmddHHMM

关键修复（v3.0.0）: REST 请求载荷中必须使用 bytesBase64Encoded 字段（不可使用 data 或 inline_data）

输入参数

• 图像: 本地路径或 URL
• 时长: 5–10 秒（可选）
• 风格: 如 “cinematic”、“ethereal”、“smooth” 等提示词（可选）

快速开始

python3 << 'PYEOF'
import os
import sys
import json
import time
import base64
import requests
import google.generativeai as genai
from pathlib import Path
from datetime import datetime

GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")
WORKSPACE = Path.home() / ".openclaw" / "workspace" / "tibetanProc"
WORKSPACE.mkdir(parents=True, exist_ok=True)
TIMESTAMP = datetime.now().strftime("%y%m%d%H%M")

# 第一步：加载图像
IMAGE_PATH = WORKSPACE / f"{TIMESTAMP}_input_image.jpg"
if not IMAGE_PATH.exists():
    print("✗ 图像未找到")
    sys.exit(1)

print(f"✓ 图像: {IMAGE_PATH.name}")

# 第二步：使用 Gemini 分析
genai.configure(api_key=GOOGLE_API_KEY)
model = genai.GenerativeModel("gemini-2.5-flash")
image_file = genai.upload_file(str(IMAGE_PATH), mime_type="image/jpeg")

prompt = """Analyze for cinematic video: describe subject, setting, lighting, textures, 
suggested camera movements (dolly, pan, orbit, zoom, rack focus)."""

response = model.generate_content([prompt, image_file])
analysis = response.text

prompt_path = WORKSPACE / f"{TIMESTAMP}_prompt.md"
with open(prompt_path, "w") as f:
    f.write(analysis)
print(f"✓ 分析结果: {prompt_path.name}")

# 第三步：构建增强型提示词
enhanced = f"""VIDEO GENERATION PROMPT
Duration: 5 seconds
Quality: High Definition

SCENE ANALYSIS:
{analysis}

MOTION GUIDELINES:
- Smooth, deliberate camera movement
- Enhance visual depth with elegant transitions
- Maintain consistent lighting
- Cinematic color grading
"""

enhanced_path = WORKSPACE / f"{TIMESTAMP}_enhanced_prompt.txt"
with open(enhanced_path, "w") as f:
    f.write(enhanced)
print(f"✓ 增强提示词: {enhanced_path.name}")

# 第四步：调用 Veo API（注意字段命名规范）
with open(IMAGE_PATH, "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode("utf-8")

VEO_URL = f"https://generativelanguage.googleapis.com/v1beta/models/veo-3.0-generate-001:predictLongRunning?key={GOOGLE_API_KEY}"

# ✅ 正确的载荷结构（v3.0.0）
payload = {
    "instances": [{
        "prompt": enhanced,
        "image": {
            "bytesBase64Encoded": image_b64,  # ← 关键：不可用 "data" 或 "inline_data"
            "mimeType": "image/jpeg"
        }
    }]
}

print("n🎬 正在调用 Veo API...")
response = requests.post(VEO_URL, json=payload, timeout=60)

if response.status_code not in [200, 202]:
    print(f"✗ API 错误: {response.json()}")
    sys.exit(1)

result = response.json()
operation_name = result.get("name")
if not operation_name:
    print(f"✗ 未返回 operation name")
    sys.exit(1)

print(f"✓ 操作名称: {operation_name}")

# 第五步：轮询直至完成
POLL_URL = f"https://generativelanguage.googleapis.com/v1beta/{operation_name}?key={GOOGLE_API_KEY}"

for attempt in range(1, 121):
    time.sleep(5 if attempt > 1 else 2)
    
    poll_response = requests.get(POLL_URL, timeout=10)
    poll_result = poll_response.json()
    
    if poll_result.get("done"):
        print(f"✓ 在 {attempt * 5} 秒内完成")
        
        # 提取视频 URL
        try:
            video_uri = poll_result["response"]["generateVideoResponse"]["generatedSamples"][0]["video"]["uri"]
        except (KeyError, IndexError):
            print(f"✗ 响应中无视频数据")
            print(json.dumps(poll_result, indent=2)[:500])
            sys.exit(1)
        
        # 第六步：下载视频
        print(f"⬇️  正在下载...")
        video_response = requests.get(f"{video_uri}&key={GOOGLE_API_KEY}", timeout=120, stream=True)
        
        if video_response.status_code == 200:
            output_path = WORKSPACE / f"{TIMESTAMP}_video.mp4"
            with open(output_path, "wb") as f:
                for chunk in video_response.iter_content(chunk_size=8192):
                    if chunk:
                        f.write(chunk)
            
            size_mb = output_path.stat().st_size / (1024 * 1024)
            print(f"✓ 视频: {output_path.name} ({size_mb:.1f} MB)")
            sys.exit(0)
        else:
            print(f"✗ 下载失败: {video_response.status_code}")
            sys.exit(1)

print("✗ 超时")
sys.exit(1)

PYEOF

详细工作流程

第一步：准备图像

WORKSPACE="$HOME/.openclaw/workspace/tibetanProc"
mkdir -p "$WORKSPACE"
TIMESTAMP=$(date +%y%m%d%H%M)
cp ./my_image.jpg "$WORKSPACE/${TIMESTAMP}_input_image.jpg"

第二步：使用 Gemini Vision 分析图像

import google.generativeai as genai
from pathlib import Path

genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
model = genai.GenerativeModel("gemini-2.5-flash")

IMAGE_PATH = Path.home() / ".openclaw" / "workspace" / "tibetanProc" / "2604110411_input_image.jpg"
image_file = genai.upload_file(str(IMAGE_PATH), mime_type="image/jpeg")

analysis_prompt = """Analyze this image for cinematic video generation:
1. Main subject and focal point
2. Setting and environment
3. Lighting direction and mood
4. Materials and textures
5. Suggested camera movements (dolly, pan, orbit, zoom, rack focus)
6. Overall energy and pacing"""

response = model.generate_content([analysis_prompt, image_file])

# 保存分析结果
with open(WORKSPACE / "2604110411_prompt.md", "w") as f:
    f.write(response.text)

第三步：构建增强型提示词

# 读取分析结果
with open(WORKSPACE / "2604110411_prompt.md") as f:
    analysis = f.read()

# 添加视频生成指令
enhanced = f"""VIDEO GENERATION PROMPT
Duration: 5 seconds
Quality: High Definition
Frame Rate: 24fps

SCENE ANALYSIS:
{analysis}

MOTION GUIDELINES:
- Smooth, deliberate camera movement
- Enhance visual depth with elegant transitions
- Maintain consistent lighting throughout
- Cinematic color grading
- Focus on visual storytelling

TECHNICAL SPECS:
- Resolution: 1080p minimum
- Aspect Ratio: 16:9
- Format: MP4 (H.264)
"""

with open(WORKSPACE / "2604110411_enhanced_prompt.txt", "w") as f:
    f.write(enhanced)

第四步：调用 Veo API（最关键步骤）

import base64
import requests

# 编码图像
with open(IMAGE_PATH, "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode("utf-8")

# 读取增强提示词
with open(WORKSPACE / "2604110411_enhanced_prompt.txt") as f:
    prompt = f.read()

# ✅ 正确的载荷结构（v3.0.0）
payload = {
    "instances": [{
        "prompt": prompt,
        "image": {
            "bytesBase64Encoded": image_b64,    # ← 关键：不可用 "data" 或 "inline_data"
            "mimeType": "image/jpeg"            # ← 必须存在且拼写准确
        }
    }]
}

VEO_URL = f"https://generativelanguage.googleapis.com/v1beta/models/veo-3.0-generate-001:predictLongRunning?key={GOOGLE_API_KEY}"

response = requests.post(VEO_URL, json=payload, timeout=60)
result = response.json()

if response.status_code in [200, 202]:
    operation_name = result["name"]
    print(f"✓ 操作名称: {operation_name}")
else:
    print(f"✗ 错误: {result}")
    exit(1)

第五步：轮询操作状态

import time

POLL_URL = f"https://generativelanguage.googleapis.com/v1beta/{operation_name}?key={GOOGLE_API_KEY}"

for attempt in range(120):  # 最长 10 分钟
    time.sleep(5)
    
    poll_response = requests.get(POLL_URL, timeout=10)
    poll_result = poll_response.json()
    
    if poll_result.get("done"):
        print(f"✓ 在 {(attempt + 1) * 5} 秒内完成")
        
        # 从响应结构中提取视频 URL
        video_uri = poll_result["response"]["generateVideoResponse"]["generatedSamples"][0]["video"]["uri"]
        print(f"✓ 视频 URL: {video_uri}")
        break
    
    progress = poll_result.get("metadata", {}).get("progressPercentage", "?")
    print(f"  正在轮询 ({attempt + 1}/120)... {progress}%")

第六步：下载视频

# URL 需附加 API key
download_url = f"{video_uri}&key={GOOGLE_API_KEY}"

video_response = requests.get(download_url, timeout=120, stream=True)

output_path = WORKSPACE / "2604110411_video.mp4"
with open(output_path, "wb") as f:
    for chunk in video_response.iter_content(chunk_size=8192):
        if chunk:
            f.write(chunk)

size_mb = output_path.stat().st_size / (1024 * 1024)
print(f"✓ 视频已下载: {output_path} ({size_mb:.1f} MB)")

响应结构

初始响应（第四步）

{
  "name": "models/veo-3.0-generate-001/operations/uiw8bpjdiqbn"
}

最终响应（第五步，当 done=true 时）

{
  "name": "models/veo-3.0-generate-001/operations/uiw8bpjdiqbn",
  "done": true,
  "response": {
    "@type": "type.googleapis.com/google.ai.generativelanguage.v1beta.PredictLongRunningResponse",
    "generateVideoResponse": {
      "generatedSamples": [
        {
          "video": {
            "uri": "https://generativelanguage.googleapis.com/v1beta/files/txg4shogthoc:download?alt=media"
          }
        }
      ]
    }
  }
}

输出文件

~/.openclaw/workspace/tibetanProc/
├── 2604110411_input_image.jpg         # 原始图像
├── 2604110411_prompt.md               # Gemini 分析结果
├── 2604110411_enhanced_prompt.txt     # 增强后的提示词（含运镜指导）
├── 2604110411_veo_init_response.json  # API 初始响应（初始化阶段）
└── 2604110411_video.mp4               # ✅ 最终视频

常见错误及修复方法

性能指标

版本历史

测试验证

已成功验证以下配置：

• 图像: 金色藏式供品（6.3 MB JPG）
• 模型: veo-3.0-generate-001
• 时长: 5 秒
• 输出: 2.4 MB MP4 视频
• 状态: ✅ 端到端功能正常