What problem does it solve? Integrating Alibaba Cloud DashScope (Bailian) media models is error-prone because image generation, TTS, and ASR use DashScope-native endpoints rather than the OpenAI-compatible path, each with its own quirks like asterisk size formats and async polling. This Skill documents the exact endpoints, request bodies, and pitfalls so you can call Qwen-Image, Qwen-TTS, and Qwen-ASR correctly on the first attempt. ## Core Features & Use Cases - Image Generation: Call qwen-image-2.0-pro and related models with correct size formatting ("1024*1024"), prompt extension, negative prompts, and seed control. - Text-to-Speech: Synthesize narration with qwen3-tts-flash using selectable voices and language types, retrieving the WAV URL from the response. - ASR with Word Timestamps: Transcribe publicly hosted audio via the async qwen3-asr-flash-filetrans model, poll the task until completion, and obtain per-word begin/end times in seconds for subtitle generation. - Use Case: Build a video narration pipeline — generate a voiceover with TTS, transcribe it with ASR to get word-level timestamps, and group words into subtitle cues for a video project. ## Quick Start Set DASHSCOPE_API_KEY in your .env file, then ask the agent to generate a sample image with the dashscope provider and a short TTS narration clip to verify both endpoints work.
