Bria VGL — 对图像生成的完全控制

以结构化 JSON 明确定义每一项视觉属性，而非依赖自然语言并寄希望于其准确表达。VGL（Visual Generation Language，视觉生成语言）为 Bria 的 FIBO 模型提供显式、确定性的控制能力，涵盖对象、光照、相机设置、构图与风格等全部维度。

相关技能：使用 bria-ai 通过 Bria API 执行这些 VGL 提示。VGL 定义结构化控制格式；bria-ai 负责图像生成、编辑及背景移除。

核心概念

VGL 用确定性的 JSON 替代模糊的自然语言提示，显式声明所有视觉属性：对象、光照、相机设置、构图与风格。这确保了图像生成过程可复现、可精确控制。

运行模式

JSON Schema

输出一个合法的 JSON 对象，必须包含以下字段：

1. short_description（字符串）

图像内容的简明摘要，最多 200 词。需涵盖关键主体、动作、场景与氛围。

2. objects（数组，最多 5 项）

每个对象需包含以下字段：

{
  "description": "详细描述，最多 100 词",
  "location": "center | top-left | bottom-right foreground | etc.",
  "relative_size": "small | medium | large within frame",
  "shape_and_color": "基本形状与主色调",
  "texture": "smooth | rough | metallic | furry | fabric | etc.",
  "appearance_details": "显著的视觉细节",
  "relationship": "与其他对象的关系",
  "orientation": "upright | tilted 45 degrees | facing left | horizontal | etc."
}

人物主体 需额外补充以下字段：

{
  "pose": "身体姿态描述",
  "expression": "winking | joyful | serious | surprised | calm",
  "clothing": "着装描述",
  "action": "人物正在执行的动作",
  "gender": "性别描述",
  "skin_tone_and_texture": "皮肤外观"
}

对象簇 需额外补充以下字段：

{
  "number_of_objects": 3
}

尺寸建议：若人物为主角，应使用 "medium-to-large" 或 "large within frame"。

3. background_setting（字符串）

整体环境、场景及背景元素（不包含在 objects 中的对象）。

4. lighting（对象）

{
  "conditions": "bright daylight | dim indoor | studio lighting | golden hour | blue hour | overcast",
  "direction": "front-lit | backlit | side-lit from left | top-down",
  "shadows": "long, soft shadows | sharp, defined shadows | minimal shadows"
}

5. aesthetics（对象）

{
  "composition": "rule of thirds | symmetrical | centered | leading lines | medium shot | close-up",
  "color_scheme": "monochromatic blue | warm complementary | high contrast | pastel",
  "mood_atmosphere": "serene | energetic | mysterious | joyful | dramatic | peaceful"
}

若人物为主角，构图中须明确指定镜头类型："medium shot"、"close-up" 或 "portrait composition"。

6. photographic_characteristics（对象）

{
  "depth_of_field": "shallow | deep | bokeh background",
  "focus": "sharp focus on subject | soft focus | motion blur",
  "camera_angle": "eye-level | low angle | high angle | dutch angle | bird's-eye",
  "lens_focal_length": "wide-angle | 50mm standard | 85mm portrait | telephoto | macro"
}

针对人物：优先选用 "standard lens (35mm-50mm)" 或 "portrait lens (50mm-85mm)"；除非明确要求，否则避免使用广角镜头。

7. style_medium（字符串）

"photograph" | "oil painting" | "watercolor" | "3D render" | "digital illustration" | "pencil sketch"

默认值为 "photograph"，除非用户明确要求其他类型。

8. artistic_style（字符串）

若非摄影风格，用最多 3 个词描述其艺术特征，例如："impressionistic, vibrant, textured"

若为摄影风格，则使用 "realistic" 或类似表述。

9. context（字符串）

描述图像类型或用途：

• "High-fashion editorial photograph for magazine spread"
• "Concept art for fantasy video game"
• "Commercial product photography for e-commerce"

10. text_render（数组）

默认值：空数组 []

仅当用户明确提供确切文本内容时才填充该字段：

{
  "text": "用户提供的确切文本（绝不可使用占位符）",
  "location": "center | top-left | bottom",
  "size": "small | medium | large",
  "color": "white | red | blue",
  "font": "serif typeface | sans-serif | handwritten | bold impact",
  "appearance_details": "Metallic finish | 3D effect | etc."
}

例外情况：作为对象固有组成部分的通用文字（例如停车标志上的 “STOP”）。

11. edit_instruction（字符串）

单条祈使句指令，用于描述编辑或生成操作。

编辑指令格式

标准编辑（无掩码）

以动词开头，清晰描述变更内容，禁止引用 “original image”（原始图像）：

掩码区域编辑

目标区域统一表述为 “masked regions”（掩码区域）或 “masked area”（掩码区域）：

保真度规则

标准编辑模式

除非编辑指令明确要求变更，否则必须保留所有视觉属性：

• 主体身份、姿态、外观
• 对象的存在性、位置、尺寸、朝向
• 构图、相机角度、镜头特性
• 风格/媒介类型

仅严格按编辑指令所要求的内容进行更改。

掩码编辑模式

• 精确保留所有可见（未掩码）区域
• 填充灰色掩码区域，使其与未掩码区域无缝融合
• 匹配现有风格、光照与主题内容
• 禁止描述灰色掩码本身——应描述将填充掩码区域的内容

示例输出

{
  "short_description": "A professional businesswoman in a navy blazer stands confidently in a modern glass office, holding a tablet. Natural daylight streams through floor-to-ceiling windows, creating a warm, productive atmosphere.",
  "objects": [
    {
      "description": "A confident businesswoman in her 30s with shoulder-length dark hair, wearing a tailored navy blazer over a white blouse. She holds a tablet in her left hand while gesturing naturally with her right.",
      "location": "center-right",
      "relative_size": "large within frame",
      "shape_and_color": "Human figure, navy and white clothing",
      "texture": "smooth fabric, professional attire",
      "appearance_details": "Minimal jewelry, well-groomed professional appearance",
      "relationship": "Main subject, interacting with tablet",
      "orientation": "facing slightly left, three-quarter view",
      "pose": "Standing upright, relaxed professional stance",
      "expression": "confident, approachable smile",
      "clothing": "Tailored navy blazer, white silk blouse, dark trousers",
      "action": "Presenting or reviewing information on tablet",
      "gender": "female",
      "skin_tone_and_texture": "Medium warm skin tone, healthy smooth complexion"
    },
    {
      "description": "A modern tablet device with a bright display showing charts and graphs",
      "location": "center, held by subject",
      "relative_size": "small",
      "shape_and_color": "Rectangular, silver frame with illuminated screen",
      "texture": "smooth glass and metal",
      "appearance_details": "Thin profile, business application visible on screen",
      "relationship": "Held by businesswoman, focus of her attention",
      "orientation": "vertical, screen facing viewer at slight angle",
      "pose": null,
      "expression": null,
      "clothing": null,
      "action": null,
      "gender": null,
      "skin_tone_and_texture": null,
      "number_of_objects": null
    }
  ],
  "background_setting": "Modern corporate office interior with floor-to-ceiling windows overlooking a city skyline. Minimalist furniture in neutral tones, potted plants adding touches of green.",
  "lighting": {
    "conditions": "bright natural daylight",
    "direction": "side-lit from left through windows",
    "shadows": "soft, natural shadows"
  },
  "aesthetics": {
    "composition": "rule of thirds, medium shot",
    "color_scheme": "professional blues and neutral whites with warm accents",
    "mood_atmosphere": "confident, professional, welcoming"
  },
  "photographic_characteristics": {
    "depth_of_field": "shallow, background slightly soft",
    "focus": "sharp focus on subject's face and upper body",
    "camera_angle": "eye-level",
    "lens_focal_length": "portrait lens (85mm)"
  },
  "style_medium": "photograph",
  "artistic_style": "realistic",
  "context": "Corporate portrait photography for company website or LinkedIn professional profile.",
  "text_render": [],
  "edit_instruction": "Generate a professional businesswoman in a modern office environment holding a tablet."
}

常见错误

1. 不虚构文字 —— 仅当用户提供确切文本时才填写 text_render，否则保持为空
2. 不过度描述 —— 最多定义 5 个对象，优先描述最重要的对象
3. 匹配对应模式 —— 掩码编辑与标准编辑须使用各自对应的 edit_instruction 格式
4. 保持保真度 —— 仅修改指令中明确要求的部分
5. 保持具体性 —— 使用明确值（如 “85mm portrait lens”），避免模糊表述（如 “nice camera”）
6. 无关字段置为 null —— 非人物对象的专属字段（如 pose、expression 等）应设为 null

curl 示例

curl -X POST "https://engine.prod.bria-api.com/v2/image/generate" 
  -H "api_token: $BRIA_API_KEY" 
  -H "Content-Type: application/json" 
  -d '{
    "structured_prompt": "{"short_description": "...", ...}",
    "prompt": "Generate this scene",
    "aspect_ratio": "16:9"
  }'

参考资料

• Schema Reference —— 完整 JSON Schema，含全部参数可选值
• bria-ai —— 执行 VGL 提示的 API 客户端与端点文档