lala Models API — 開發者文檔

認證

所有請求帶 header x-lala-key: <你的 API key>（或標準 Authorization: Bearer <key>）。

申請 key：電郵 [email protected]或經 WhatsApp 聯絡我們。

模型列表

curl https://lala.buddylala.ai/v1/models -H "x-lala-key: $LALA_KEY"

對話 Chat Completions

OpenAI-compatible。支援 stream: true 串流輸出。

curl

curl https://lala.buddylala.ai/v1/chat/completions \
  -H "x-lala-key: $LALA_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lala-6",
    "messages": [{"role": "user", "content": "你好，介紹下自己"}],
    "max_tokens": 500
  }'

Python（openai SDK）

from openai import OpenAI

client = OpenAI(
    base_url="https://lala.buddylala.ai/v1",
    api_key="***",   # 你的 lala key
)

resp = client.chat.completions.create(
    model="lala-super",
    messages=[{"role": "user", "content": "寫一首關於秋天的詩"}],
)
print(resp.choices[0].message.content)

Node.js

const res = await fetch("https://lala.buddylala.ai/v1/chat/completions", {
  method: "POST",
  headers: {
    "x-lala-key": process.env.LALA_KEY,
    "content-type": "application/json",
  },
  body: JSON.stringify({
    model: "lala-ultrafast",
    messages: [{ role: "user", content: "1+1=?" }],
    stream: false,
  }),
});
const data = await res.json();
console.log(data.choices[0].message.content);

向量 Embeddings

curl https://lala.buddylala.ai/v1/embeddings \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{"model": "lala-rag", "input": "寵物食品零售自動化"}'

回傳標準 OpenAI embeddings 格式（data[].embedding 向量陣列），可直接入向量資料庫做 RAG。

圖像生成 Image

curl https://lala.buddylala.ai/v1/images/generations \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{"model": "lala-image", "prompt": "a cute corgi puppy, studio photo", "n": 1}'

回傳 data[].url（圖片 CDN 連結）。計費：每張固定價，見價格頁。

視頻生成 Video（任務式）

視頻生成是兩段式：先提交任務取得 task id，再輪詢取得結果。

步驟 1：提交任務

curl https://lala.buddylala.ai/v1/responses \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "lala-video",
    "input": [{"type": "text", "text": "a cat yawning in sunlight"}],
    "duration": 5,
    "resolution": "720p",
    "ratio": "16:9",
    "generate_audio": true
  }'
# → {"id": "cgt-xxxx", ...}   ← 保存此任務識別碼！

參數：duration 4–30 秒（或 -1 智能選擇）、resolution 480p/720p/1080p、ratio 16:9 / 4:3 / 1:1 / 3:4 / 9:16 / 21:9 / adaptive、generate_audio 默認 true（自動配人聲／音效／背景樂）。

步驟 2：輪詢結果

curl https://lala.buddylala.ai/v1/responses \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{"model": "lala-video-get", "input": "cgt-xxxx"}'
# status: succeeded 時 output[].content[].text = 視頻下載 URL

語音合成 Speech (TTS)

curl https://lala.buddylala.ai/v1/responses \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "lala-speech",
    "input": "你好，歡迎致電 Buddylala",
    "output_format": "hex",
    "voice_setting": {"voice_id": "male-qn-qingse", "speed": 1, "vol": 1, "pitch": 0}
  }'
# → {"data": {"audio": "<hex 編碼音頻>"}}

hex 轉 mp3：python3 -c "import binascii,sys; open('out.mp3','wb').write(binascii.unhexlify(sys.argv[1]))" "<hex>"。80+ 預置音色、9 種情緒。

音樂生成 Music

curl https://lala.buddylala.ai/v1/responses \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "lala-music",
    "input": "upbeat jingle for a pet shop",
    "lyrics": "la la la happy day with you"
  }'
# → {"data": {"audio": "<hex>"}}

lyrics 必填（可先用 lala-lyric 生成）。

歌詞生成 Lyric

curl https://lala.buddylala.ai/v1/responses \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "lala-lyric",
    "mode": "write_full_song",
    "input": "一首關於貓的輕快廣東歌"
  }'
# → {"song_title": "...", "style_tags": "...", "lyrics": "[Verse]..."}

mode：write_full_song（完整歌曲）或 edit（續寫／修改，需帶 lyrics 現有內容）。

結構化判研 Decision（非文本生成）

lala-decision不生成文本，只做結構化判斷：提供一段狀態（工單／對話／日誌）＋一組問題，並行回傳 Choice / Score / Noul 三類答案附校準概率。支援圖像判斷：可附最多 4 張圖（PNG/JPEG/WebP，每張 ≤4MiB，base64 內嵌，不收 remote URL）一併分析。典型回應 1–5 秒，適合工單分派、意圖識別、內容風控。

curl https://lala.buddylala.ai/typesafe/systemone \
  -H "x-lala-key: $LALA_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "lala-decision",
    "state": "Customer: my order is 5 days late and I want a refund!",
    "questions": {
      "urgency": {
        "type": "choice",
        "instructions": "判斷工單緊急程度",
        "criteria": {"低": "語氣平和", "中": "有不满但无期限", "高": "情绪激动或提退款"}
      },
      "refund_requested": {
        "type": "noul",
        "instructions": "客戶是否明確要求退款？"
      }
    }
  }'
# → {"answers": {"refund_requested": {"noul": 0.99, "type": "noul"}}, "model": "lala-decision", ...}

問題類型：choice（criteria 係 object：選項名→判定標準）、score（criteria 係 array：由低到高檔位）、noul（true/false 判斷，不需 criteria）。

圖像判斷（可選）：在 body 加"images": ["<base64 圖片>"]（最多 4 張；每張 ≤4MiB、16 megapixels；整個 request body ≤13MiB；只收 base64 內嵌，不收 URL）。示例：

{
  "model": "lala-decision",
  "images": ["iVBORw0KGgo...（base64 PNG）"],
  "state": "客人傳送了一張收據相片，要求我們查看",
  "questions": {
    "is_valid_receipt": {"type": "noul", "instructions": "圖片是否為一張有效收據？"}
  }
}

錯誤處理

Response 文字經 brand normalization（供應商相關字樣會統一為 lala/buddylala 品牌稱呼），JSON 結構及欄位完全標準，不影響解析。