Bailian Studio

Use DashScope for OCR、TTS、文生图、图生图。

Requirements

• Python 3
• dashscope (>=1.24.0)
• oss2
• requests
• ffmpeg（TTS 播放依赖，使用 ffplay）

Install:

pip install -r requirements.txt

Config

API Key (priority order):

1. DASHSCOPE_API_KEY env
2. secrets/bailian.env

OSS (priority order):

1. OSS_ACCESS_KEY, OSS_SECRET_KEY, OSS_BUCKET, OSS_ENDPOINT, OSS_REGION env
2. secrets/bailian.env

Example secrets/bailian.env:

DASHSCOPE_API_KEY=sk-xxx
DASHSCOPE_BASE_URL=https://dashscope.aliyuncs.com/api/v1

# TTS 可选配置（留空走默认）
BAILIAN_TTS_MODEL=qwen3-tts-flash
BAILIAN_TTS_VOICE=
BAILIAN_TTS_SAMPLE_RATE=16000

OSS_ACCESS_KEY=ak-xxx
OSS_SECRET_KEY=sk-xxx
OSS_BUCKET=your-bucket
OSS_ENDPOINT=oss-cn-beijing.aliyuncs.com
OSS_REGION=cn-beijing

Defaults:

• Region/base URL: Beijing (https://dashscope.aliyuncs.com/api/v1)
• Image model: qwen-image-2.0-pro
• Output dir: tmp/bailian-studio/
• Output format: PNG

Usage

TTS (speak)

python3 {baseDir}/scripts/tts_speak.py --text "你好"

OCR (text)

From local image (uploads to OSS):

python3 {baseDir}/scripts/ocr_text.py --image /path/to.png

From URL:

python3 {baseDir}/scripts/ocr_text.py --url https://example.com/image.png

Image generate (text-to-image)

python3 {baseDir}/scripts/image_generate.py \
  --prompt "一只坐在云端的橘猫" \
  --width 1024 \
  --height 1024

Image generate (image-to-image)

Local image:

python3 {baseDir}/scripts/image_generate.py \
  --prompt "改成赛博朋克风格" \
  --image /path/to/reference.png \
  --width 1024 \
  --height 1024

URL image:

python3 {baseDir}/scripts/image_generate.py \
  --prompt "改成水彩插画风格" \
  --image https://example.com/reference.png \
  --width 1024 \
  --height 1024

stdin prompt

echo "一只会发光的鲸鱼漂浮在夜空" | python3 {baseDir}/scripts/image_generate.py

Behavior

• 本地参考图：先上传 OSS，再传给 DashScope
• URL 参考图：直接透传给 DashScope
• 默认一次生成 1 张图
• 成功后 stdout 打印保存路径
• 若文件名已存在，自动重命名
• 失败时输出错误信息并返回非 0 退出码