必需参数:

• prompt (string): 期望图像的文本描述。可以包含 @modelname 来使用训练好的模型。

可选参数:

• baseModel: 要使用的 AI 模型
  • "gemini-3-pro-image-preview" - Nano Banana Pro（推荐）
  • "flux2-max" - Flux2 Max
  • "seedream-4-5-251128" - SeeDream 4.5
• mode: 分辨率模式（因模型而异）
  • Nano Banana Pro: "default", "2k", "4k"
  • Flux2 Max: "1k", "hd"
  • SeeDream: "2k", "4k"
• width / height: 图像尺寸（520-2160px）

工作流程:

1. 使用参数 POST 到 /images
2. 在响应中接收 id
3. 每 3 秒轮询一次 GET /images/{id}
4. 当 status === "ready" 时，访问 imageVersions.original

代码示例请参见 examples/image_generation.js

2. 视频生成

端点: POST /videos

必需参数:

• prompt (string): 期望视频的文本描述
• 源图像（以下之一）:
  • imageUrl: 源图像的 URL
  • originalImageCompletionId: 先前图像生成的 ID

可选参数:

• settings.mode: 视频模型
  • "default" - 默认模型
  • "veo31" - VEO 3.1
  • "kling26" - Kling 2.6
  • "wan25" - Wan 2.5
• settings.duration: 视频长度，单位秒（2-12 秒，取决于模型）

工作流程:

1. 确保您有源图像（如果需要，先生成一个）
2. 使用参数 POST 到 /videos
3. 在响应中接收 id
4. 每 2-3 秒轮询一次 GET /videos/{id}
5. 当 status === "ready" 时，访问 videoPaths

代码示例请参见 examples/video_generation.py

3. 图像编辑（上下文编辑）

端点: POST /image-edits

必需参数:

• mode: 编辑模式
  • "context" - AI 编辑（主要模式）
  • "skin" - 皮肤修复
• prompt: 编辑指令（例如，"change background to beach"）
• 源图像（以下之一）:
  • imageUrl: 源图像的 URL
  • inputImageUrls[]: 源图像 URL 数组（最多 9 个）
  • originalImageCompletionId: 先前生成的 LetzAI 图像的 ID

可选参数:

• settings.model: "gemini-3-pro-image-preview", "flux2-max", "seedream-4-5-251128"
• settings.resolution: "2k" (HD) 或 "4k" (Ultra HD)
• settings.aspect_ratio: "1:1", "16:9", "9:16", "4:3", "3:4", "21:9", "9:21"
• baseModel: settings.model 的替代项
• webhookUrl: 可选的回调 URL
• organizationId: 用于计费的可选组织 ID

工作流程:

1. 使用参数 POST 到 /image-edits
2. 在响应中接收 id
3. 每 3 秒轮询一次 GET /image-edits/{id}
4. 当 status === "ready" 时，访问 generatedImageCompletion.imageVersions.original

注意: 局部重绘 (mode: "in") 和扩展画布 (mode: "out") 已弃用 - 请改用上下文编辑。

4. 图像放大

端点: POST /upscales

必需参数:

• 源图像（以下之一）:
  • imageUrl: 源图像的 URL
  • imageCompletionId: 先前图像生成的 ID

可选参数:

• strength: 放大系数（1-3）

工作流程:

1. 使用参数 POST 到 /upscales
2. 在响应中接收 id
3. 每 3 秒轮询一次 GET /upscales/{id}
4. 当 status === "ready" 时，访问放大后的图像

5. 自定义 AI 模型（训练好的模型）

LetzAI 用户可以通过 Web 界面训练关于人物、对象或风格的自定义 AI 模型。这些训练好的模型可以通过 @modelname 语法在提示词中使用。

列出模型端点: GET /models

查询参数:

• page: int (默认: 1)
• limit: int (默认: 10)
• sortBy: "createdAt" | "usages"
• sortOrder: "ASC" | "DESC"
• class: "person" | "object" | "style"

获取模型详情: GET /models/{id}

模型类别:

• person: 基于特定人物的照片训练
• object: 基于产品/对象图像训练
• style: 基于艺术风格示例训练

在提示词中使用模型: 使用 @modelname 语法标记模型:

• @john_doe on the beach at sunset - 使用人物模型
• A product photo featuring @my_product - 使用对象模型
• Portrait in @vintage_style aesthetic - 使用风格模型

注意: 模型训练通过 LetzAI Web 界面 (letz.ai) 完成，而非通过 API。

工作流程决策树

用户想要创建图像:

1. 根据质量/成本需求确定合适的模型
2. 使用 POST /images 并指定合适的 baseModel
3. 如果使用训练好的模型，在提示词中包含 @modelname
4. 每 3 秒轮询一次 GET /images/{id} 直到就绪
5. 返回 imageVersions.original URL

用户想要使用自定义训练好的模型:

1. 使用 GET /models 列出可用的训练好的模型（如果需要，按类别过滤）
2. 在生成图像时，在提示词中包含 @modelname
3. 正常使用 POST /images 生成图像

用户想要编辑现有图像:

1. 获取源图像 URL、inputImageUrls 数组或 originalImageCompletionId
2. 使用 POST /image-edits 并设置 mode="context"
3. 根据需要包含分辨率、宽高比和模型的设置
4. 每 3 秒轮询一次 GET /image-edits/{id} 直到就绪
5. 返回 generatedImageCompletion.imageVersions.original

用户想要创建视频:

1. 确保他们有源图像（URL 或 imageCompletionId）
2. 如果没有源图像，先使用 /images 生成一个
3. 使用 POST /videos 并指定所需的设置
4. 每 2-3 秒轮询一次 GET /videos/{id} 直到就绪
5. 从 videoPaths 返回视频 URL

用户想要放大图像:

1. 获取源图像 URL 或 imageCompletionId
2. 使用 POST /upscales 并指定所需的 strength
3. 每 3 秒轮询一次 GET /upscales/{id} 直到就绪
4. 返回放大后的图像 URL

状态轮询模式

LetzAI 使用异步生成。在任何 POST 请求之后，您必须轮询相应的 GET 端点，直到作业完成。

状态值

轮询间隔

• 图像: 每 3 秒
• 视频: 每 2-3 秒
• 图像编辑: 每 3 秒
• 放大: 每 3 秒

详细的轮询实现请参见 examples/polling_pattern.md

定价参考

错误处理

常见 HTTP 状态码

错误响应格式

{
  "error": "错误描述",
  "code": "ERROR_CODE"
}

限制

• 异步生成: 所有生成都是异步的 - 必须轮询结果
• 视频源: 视频生成需要源图像
• 参考图像: 图像编辑最多 9 张参考图像
• 模型训练: 无法通过 API 训练自定义 AI 模型 - 使用 letz.ai Web 界面
• 需要 API 密钥: API 访问需要付费订阅

快速参考：API 端点

关键响应字段

• 图像/放大: imageVersions.original, imageVersions["1920x1920"], imageVersions["640x640"]
• 编辑: generatedImageCompletion.imageVersions.original
• 视频: videoPaths 对象, videoVersions 数组
• 状态值: new, in progress/generating, ready, failed

其他资源

• API 文档: api.letz.ai/doc
• 开发者文档: letz.ai/docs/api
• 详细 API 参考: api_reference.md
• 代码示例: examples/

每周安装数

877

代码仓库

letz-ai/letzai-skill

首次出现

2026年1月27日

安全审计

Gen Agent Trust HubFailSocketPassSnykWarn

安装于

opencode868

gemini-cli867

kimi-cli866

amp866

github-copilot866

codex866

Required Parameters:

• prompt (string): Text description of the desired image. Can include @modelname to use trained models.

Optional Parameters:

• baseModel: AI model to use
  • "gemini-3-pro-image-preview" - Nano Banana Pro (recommended)
  • "flux2-max" - Flux2 Max
  • "seedream-4-5-251128" - SeeDream 4.5
• mode: Resolution mode (varies by model)
  • Nano Banana Pro: "default", "2k", "4k"
  • Flux2 Max: "1k", "hd"
  • SeeDream: "2k", "4k"
• width / height: Image dimensions (520-2160px)

Workflow:

1. POST to /images with parameters
2. Receive id in response
3. Poll GET /images/{id} every 3 seconds
4. When status === "ready", access imageVersions.original

For code examples, see examples/image_generation.js

2. Video Generation

Endpoint: POST /videos

Required Parameters:

• prompt (string): Text description of the desired video
• Source image (one of):
  • imageUrl: URL of source image
  • originalImageCompletionId: ID from previous image generation

Optional Parameters:

• settings.mode: Video model
  • "default" - Default model
  • "veo31" - VEO 3.1
  • "kling26" - Kling 2.6
  • "wan25" - Wan 2.5
• settings.duration: Video length in seconds (2-12 depending on model)

Workflow:

1. Ensure you have a source image (generate one first if needed)
2. POST to /videos with parameters
3. Receive id in response
4. Poll GET /videos/{id} every 2-3 seconds
5. When status === "ready", access videoPaths

For code examples, see examples/video_generation.py

3. Image Editing (Context Editing)

Endpoint: POST /image-edits

Required Parameters:

• mode: Edit mode
  • "context" - AI editing (primary mode)
  • "skin" - Skin fix
• prompt: Edit instruction (e.g., "change background to beach")
• Source image (one of):
  • imageUrl: URL of source image
  • inputImageUrls[]: Array of source image URLs (max 9)
  • originalImageCompletionId: ID of previously generated LetzAI image

Optional Parameters:

• settings.model: "gemini-3-pro-image-preview", "flux2-max", "seedream-4-5-251128"
• settings.resolution: "2k" (HD) or "4k" (Ultra HD)
• settings.aspect_ratio: "1:1", "16:9", "9:16", "4:3", "3:4", "21:9", "9:21"
• baseModel: Alternative to settings.model
• webhookUrl: Optional callback URL
• organizationId: Optional org ID for billing

Workflow:

1. POST to /image-edits with parameters
2. Receive id in response
3. Poll GET /image-edits/{id} every 3 seconds
4. When status === "ready", access generatedImageCompletion.imageVersions.original

Note: Inpainting (mode: "in") and Outpainting (mode: "out") are deprecated - use Context Editing instead.

4. Image Upscaling

Endpoint: POST /upscales

Required Parameters:

• Source image (one of):
  • imageUrl: URL of source image
  • imageCompletionId: ID from previous image generation

Optional Parameters:

• strength: Upscale factor (1-3)

Workflow:

1. POST to /upscales with parameters
2. Receive id in response
3. Poll GET /upscales/{id} every 3 seconds
4. When status === "ready", access upscaled image

5. Custom AI Models (Trained Models)

LetzAI users can train custom AI models on persons, objects, or styles via the web interface. These trained models can be used in prompts via the @modelname syntax.

List Models Endpoint: GET /models

Query Parameters:

• page: int (default: 1)
• limit: int (default: 10)
• sortBy: "createdAt" | "usages"
• sortOrder: "ASC" | "DESC"
• class: "person" | "object" | "style"

Get Model Details: GET /models/{id}

Model Classes:

• person: Trained on photos of a specific person
• object: Trained on product/object images
• style: Trained on artistic style examples

Using Models in Prompts: Tag models with @modelname syntax:

• @john_doe on the beach at sunset - Use a person model
• A product photo featuring @my_product - Use an object model
• Portrait in @vintage_style aesthetic - Use a style model

Note: Model training is done via the LetzAI web interface (letz.ai), not via API.

Workflow Decision Tree

User wants to create an image:

1. Determine appropriate model based on quality/cost needs
2. Use POST /images with appropriate baseModel
3. If using a trained model, include @modelname in the prompt
4. Poll GET /images/{id} every 3s until ready
5. Return imageVersions.original URL

User wants to use a custom trained model:

1. Use GET /models to list available trained models (filter by class if needed)
2. Include @modelname in the prompt when generating images
3. Generate image normally with POST /images

User wants to edit an existing image:

1. Obtain source image URL, inputImageUrls array, or originalImageCompletionId
2. Use POST /image-edits with mode="context"
3. Include settings for resolution, aspect_ratio, and model as needed
4. Poll GET /image-edits/{id} every 3s until ready
5. Return generatedImageCompletion.imageVersions.original

User wants to create a video:

1. Ensure they have a source image (URL or imageCompletionId)
2. If no source image, generate one first using /images
3. Use POST /videos with desired settings
4. Poll GET /videos/{id} every 2-3s until ready
5. Return video URL from videoPaths

User wants to upscale an image:

1. Obtain source image URL or imageCompletionId
2. Use POST /upscales with desired strength
3. Poll GET /upscales/{id} every 3s until ready
4. Return upscaled image URL

Status Polling Pattern

LetzAI uses asynchronous generation. After any POST request, you must poll the corresponding GET endpoint until the job completes.

Status Values

Polling Intervals

• Images: Every 3 seconds
• Videos: Every 2-3 seconds
• Image Edits: Every 3 seconds
• Upscales: Every 3 seconds

For detailed polling implementation, see examples/polling_pattern.md

Pricing Reference

Error Handling

Common HTTP Status Codes

Error Response Format

{
  "error": "Error description",
  "code": "ERROR_CODE"
}

Limitations

• Async Generation: All generation is asynchronous - must poll for results
• Video Source: Video generation requires a source image
• Reference Images: Maximum 9 reference images for image editing
• Model Training: Cannot train custom AI models via API - use letz.ai web interface
• API Key Required: Paid subscription required for API access

Quick Reference: API Endpoints

Key Response Fields

• Images/Upscales: imageVersions.original, imageVersions["1920x1920"], imageVersions["640x640"]
• Edits: generatedImageCompletion.imageVersions.original
• Videos: videoPaths object, videoVersions array
• Status values: new, in progress/generating, ready, failed

Additional Resources

• API Documentation: api.letz.ai/doc
• Developer Docs: letz.ai/docs/api
• Detailed API Reference: api_reference.md
• Code Examples: examples/

Weekly Installs

877

Repository

letz-ai/letzai-skill

First Seen

Jan 27, 2026

Security Audits

Gen Agent Trust HubFailSocketPassSnykWarn

Installed on

opencode868

gemini-cli867

kimi-cli866

amp866

github-copilot866

codex866