Luma AI生成服务器
使用Luma AI创建视频和图像,此MCP服务器处理来自Claude Desktop的Luma Dream Machine的所有API功能。
服务介绍
Luma AI MCP 服务器 🎥
Luma AI 的 Dream Machine API 的模型上下文协议服务器。
概述
此 MCP 服务器与 Luma AI 的 Dream Machine API(v1)集成,通过大型语言模型提供生成、管理和操作由 AI 生成的视频和图像的工具。它实现了模型上下文协议(MCP),以实现 AI 助手与 Luma 创意工具之间的无缝交互。
特性 ✨
- 文本到视频生成
- 带关键帧的高级视频生成
- 图像到视频转换
- 视频扩展和插值
- 带参考图像的图像生成
- 向视频添加音频
- 视频放大
- 信用管理
- 生成跟踪和状态检查
工具 🛠️
-
ping- Check if the Luma API is running
- No parameters required
-
create_generation- Creates a new video generation
- Input:
prompt(string, required): Text description of the video to generatemodel(string, optional): Model to use (default: "ray-2")- Available models: "ray-1-6", "ray-2", "ray-flash-2"
resolution(string, optional): Video resolution (choices: "540p", "720p", "1080p", "4k")duration(string, optional): Video duration (only "5s" and "9s" are currently supported)aspect_ratio(string, optional): Video aspect ratio (e.g., "16:9", "1:1", "9:16", "4:3", "3:4", "21:9", "9:21")loop(boolean, optional): Whether to make the video loopkeyframes(object, optional): Start and end frames for advanced video generation:frame0and/orframe1with either:{"type": "image", "url": "image_url"}for image keyframes{"type": "generation", "id": "generation_id"}for video keyframes
-
get_generation- Gets the status of a generation
- Input:
generation_id(string, required): ID of the generation to check
- Output includes:
- Generation ID
- State (queued, dreaming, completed, failed)
- Failure reason (if failed)
- Video URL (if completed)
-
list_generations- Lists all generations
- Input:
limit(number, optional): Maximum number of generations to return (default: 10)offset(number, optional): Number of generations to skip
-
delete_generation- Deletes a generation
- Input:
generation_id(string, required): ID of the generation to delete
-
upscale_generation- Upscales a video generation to higher resolution
- Input:
generation_id(string, required): ID of the generation to upscaleresolution(string, required): Target resolution for the upscaled video (one of "540p", "720p", "1080p", or "4k")
- Note:
- The generation must be in a completed state to be upscaled
- The target resolution must be higher than the original generation's resolution
- Each generation can only be upscaled once
-
add_audio- Adds AI-generated audio to a video generation
- Input:
generation_id(required): The ID of the generation to add audio toprompt(required): The prompt for the audio generationnegative_prompt(optional): The negative prompt for the audio generationcallback_url(optional): URL to notify when the audio processing is complete
-
generate_image- Generates an image from a text prompt with optional reference images
- Input:
prompt(string, required): Text description of the image to generatemodel(string, optional): Model to use for image generation (default: "photon-1")- Available models: "photon-1", "photon-flash-1"
aspect_ratio(string, optional): Image aspect ratio (same options as video)image_ref(array, optional): Reference images to guide generation- Each ref:
{"url": "image_url", "weight": optional_float}
- Each ref:
style_ref(array, optional): Style reference images- Each ref:
{"url": "image_url", "weight": optional_float}
- Each ref:
character_ref(object, optional): Character reference images- Format:
{"identity_name": {"images": ["url1", "url2", ...]}}
- Format:
modify_image_ref(object, optional): Image to modify- Format:
{"url": "image_url", "weight": optional_float}
- Format:
-
get_credits- Gets credit information for the current user
- No parameters required
- Returns available credit balance in USD cents
-
get_camera_motions- Gets all supported camera motions
- No parameters required
- Returns: List of available camera motion strings
Claude Desktop 设置 🖥️
-
从 Luma AI 获取您的 Luma API 密钥(注册或登录以获取您的 API 密钥)
-
将以下内容添加到您的 Claude Desktop 配置文件中:
- 在 macOS 上:
~/Library/Application Support/Claude/claude_desktop_config.json - 在 Windows 上:
%APPDATA%\Claude\claude_desktop_config.json
{ "mcpServers": { "luma": { "command": "uv", "args": [ "run", "--project", "/path/to/your/luma-ai-mcp-server", "-m", "luma_ai_mcp_server" ], "env": { "LUMA_API_KEY": "your-luma-api-key-here" } } } }替换:
/path/to/your/luma-ai-mcp-server为您的服务器目录的实际路径your-luma-api-key-here为您的实际 Luma API 密钥
- 在 macOS 上:
-
重启 Claude Desktop
-
完成!您现在可以在 Claude Desktop 对话中直接使用 Luma AI 工具。
快速故障排除 🛠️
如果您遇到问题:
- 检查您的 API 密钥是否正确
- 确保服务器路径正确
- 查看日志:
tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
高级视频生成类型 🎬
Luma API 支持通过关键帧进行多种类型的高级视频生成:
- 从图像开始:提供
frame0,设置type: "image"并提供一个图像 URL - 以图像结束:提供
frame1,设置type: "image"并提供一个图像 URL - 扩展视频:提供
frame0,设置type: "generation"并提供一个生成 ID - 反向扩展视频:提供
frame1,设置type: "generation"并提供一个生成 ID - 在视频之间插值:同时提供
frame0和frame1,设置type: "generation"并提供生成 ID
API 限制和注意事项 📝
- 持续时间:目前 API 仅支持 "5s" 或 "9s" 的持续时间
- 分辨率:有效值为 "540p", "720p", "1080p", 和 "4k"
- 模型:
- 视频生成:
- "ray-2"(默认)- 最佳质量,但较慢
- "ray-flash-2" - 生成速度更快
- "ray-1-6" - 旧版模型
- 图像生成:
- "photon-1"(默认)- 最佳质量,但较慢
- "photon-flash-1" - 生成速度更快
- 视频生成:
- 生成类型:视频、图像及高级(带关键帧)
- 纵横比:"1:1"(方形),"16:9"(横向),"9:16"(纵向),"4:3"(标准),"3:4"(标准纵向),"21:9"(超宽),"9:21"(超宽纵向)
- 状态:"queued"(排队中),"dreaming"(处理中),"completed"(完成),"failed"(失败)
- 放大:
- 只有当视频生成处于 "complete"(完成)状态时才能放大
- 目标分辨率必须高于原始生成的分辨率
- 每个生成只能放大一次
- API 密钥:需要在环境变量中设置
- API 版本:使用 Dream Machine API v1
许可证 📄
MIT