/alibabacloud-bailian-video-creator

name: alibabacloud-bailian-video-creator
description: AI video creation skill supporting text-to-video, image-to-video, reference-to-video, video editing, and video understanding. Uses wan2.6-t2v, wan2.6-i2v, wan2.6-r2v-flash, wanx2.1-vace-plus, qwen3.5-plus and other models. Use this skill when users need to generate videos, edit videos, or analyze video content. Note: on first run, it will auto-manage DashScope API Keys (create/recycle) and may auto-install the Alibaba Cloud CLI ModelStudio plugin.

AI Video Creator

Professional-grade AI video creation skill supporting text-to-video, image-to-video, video editing, and video understanding. Built on Alibaba Cloud DashScope API.

Prompt Design Guide

The core formula for high-quality video prompts: `[Scene Overview] + [Shot Design] + [Camera Movement] + [Lighting & Atmosphere] + [Visual Style]`

Key elements: Camera language (shot size, movement, angle), shot rhythm design, lighting and atmosphere control, visual style reference. For multi-shot videos, use the shot list format `Shot N [start-end time] Shot description`.

Negative prompt template: `blurry, low quality, distorted, glitchy, unnatural movement, jumpy cuts, inconsistent lighting, artifacts, watermark, text overlay, static, frozen frames, uncanny valley`

> For detailed camera language reference tables, shot list templates, lighting styles, and complete prompt examples, refer to [references/prompt-guide.md](references/prompt-guide.md)

Version

v2.1.0 (2026-03-10)

Feature List

| Feature | Model | Status | Description | |---------|-------|--------|-------------| | Video Understanding | qwen3.5-plus | ✅ | Analyze video content with configurable frame extraction rate | | Text-to-Video (Multi-shot) | wan2.6-t2v | ✅ | Generate multi-shot videos from text descriptions | | Image-to-Video | wan2.6-i2v | ✅ | Generate video from image + audio | | Reference-to-Video | wan2.6-r2v-flash | ✅ | Generate multi-character video from multiple reference materials (video/image) | | Video Editing (Repainting) | wanx2.1-vace-plus | ✅ | Video repainting while preserving original motion/structure | | Video Region Edit | wanx2.1-vace-plus | ✅ | Fine-grained editing of specific video regions |

**Note**: The DashScope qwen3.5-plus model's `image_url` type only supports static image analysis and does not support direct video file upload. To analyze video, extract key frames first and use the image analysis feature.

Prerequisites

Before executing any task, the following checks must be completed:

1. **Verify API Key availability**: Run `from api_key import get_api_key; api_key = get_api_key()`. If `get_api_key()` raises an exception (including "quota exceeded", `AuthenticationError`, `InvalidApiKey`, `401`, etc.), **you must immediately stop execution and display the error message to the user** -- do not skip API calls, do not continue with subsequent steps, do not fabricate execution results 2. **Confirm remote API availability**: This skill completes all video generation and content analysis tasks via the DashScope remote API, without relying on local GPU, PyTorch, Stable Diffusion, etc. As long as the API Key is valid, the skill has full capabilities. **Do not claim inability to complete tasks just because local tools are missing** 3. **API Key security**: Never write the full API Key in plain text in script files, logs, or terminal output. Must read via the `api_key.py` module or environment variable `os.environ.get("DASHSCOPE_API_KEY")`. Display in masked form when debugging (e.g., `sk-***xxx`)

Orchestration Logic

Products and Services Involved

All APIs in this skill are called through **Alibaba Cloud DashScope (Bailian)**; no other cloud products are involved.

| Component | Purpose | |-----------|---------| | `scripts/api_key.py` | Unified API Key management (reads from `~/.aliyun/config.json` or environment variable) | | DashScope Video Generation API | Text-to-video, image-to-video, reference-to-video, video editing (async) | | DashScope MultiModalConversation API | Video understanding (sync) | | FFmpeg (local tool) | Video format conversion, key frame extraction, and other pre/post-processing (optional) |

Feature Selection Decision Tree

Select the corresponding feature based on the user's **input materials** and **intent**:

User Request
│
├─ User has an existing video to analyze/understand?
│   └─ YES → Video Understanding (video_understanding.py, qwen3.5-plus)
│
├─ User has an existing video to modify?
│   ├─ Modify a local region (has mask)?
│   │   └─ YES → Video Region Edit (video_local_edit.py, wanx2.1-vace-plus)
│   └─ Overall style repainting?
│       └─ YES → Video Editing (video_edit.py, wanx2.1-vace-plus)
│
├─ User has multiple reference materials (character videos/images) to synthesize?
│   └─ YES → Reference-to-Video (reference_to_video.py, wan2.6-r2v-flash)
│
├─ User has a single image to generate video from?
│   └─ YES → Image-to-Video (image_to_video.py, wan2.6-i2v)
│
└─ User only has a text description?
    └─ YES → Text-to-Video (text_to_video.py, wan2.6-t2v)

Quick Reference Table

| User Input | Feature | Script | Model | |------------|---------|--------|-------| | Video URL + analysis question | Video Understanding | `video_understanding.py` | qwen3.5-plus | | Pure text description | Text-to-Video | `text_to_video.py` | wan2.6-t2v | | Image URL (± audio URL) | Image-to-Video | `image_to_video.py` | wan2.6-i2v | | Multiple reference materials (video/image mix) | Reference-to-Video | `reference_to_video.py` | wan2.6-r2v-flash | | Video URL + new style description | Video Editing | `video_edit.py` | wanx2.1-vace-plus | | Video URL + mask image + edit description | Video Region Edit | `video_local_edit.py` | wanx2.1-vace-plus |

Invocation Flow

**You must directly run existing scripts in the `scripts/` directory to complete tasks. Creating new scripts from scratch as replacements is forbidden.** Each feature has a corres