videogencn - Chinese Video Generation Skill

Overview

Generate short video clips using Chinese video models across four providers — Alibaba Cloud Bailian (Wan/PixVerse/Kling/Vidu/HappyHorse), Volcengine Ark (Jimeng/即梦), MiniMax (海螺 AI), and Tencent Hunyuan (混元).

Four modes, auto-selected from inputs:

Mode	Inputs	Default model
t2v 文生视频	prompt only	wan2.7-t2v-2026-04-25 (Bailian)
i2v 图生视频	prompt + --image	wan2.6-i2v-flash (Bailian)
kf2v 首尾帧	prompt + --image + --last-frame	pixverse/pixverse-c1-kf2v (Bailian)
r2v 参考生视频	prompt + --ref (1-7 images)	pixverse/pixverse-c1-r2v (Bailian)

Video generation is asynchronous: submit → poll every 10s → download MP4. Result URLs expire after 24h, so the script always downloads immediately.

Local images: Wan/HappyHorse accept base64 data URIs directly; PixVerse/Kling/Vidu auto-upload to DashScope OSS (48h); Jimeng/Hunyuan use base64; MiniMax uploads via its file API.

When to Use This Skill

• User asks to 生成视频 / 文生视频 / 图生视频 / 首尾帧 / 参考生视频
• User names a Chinese video model: 万相/Wan, 爱诗/PixVerse, 可灵/Kling, Vidu, HappyHorse, 即梦/Jimeng, 海螺/MiniMax, 混元/Hunyuan
• User needs B-roll, animated stills, character-consistent clips, or frame transitions
• User asks about model pricing, features, or wants to compare models → open docs/models.html

Workflow

Step 0: Prompt Refinement (interactive)

Run for t2v/i2v unless skipped (see §0.4). Claude polishes the user's prompt before generation.

0.1 Analyze the raw input

Dimension	Check
Subject	Who/what? Appearance, action, expression
Scene	Where? Background, environment, atmosphere
Lighting	Time of day? Light quality? (golden hour, neon, soft diffused, backlit)
Camera	Shot type? (close-up, wide, aerial, tracking). Movement? (push-in, pan, orbit)
Mood/Style	Emotional tone? Visual style? (cinematic, anime, documentary, surreal)
Motion	What moves? How? Speed, direction, dynamics
Temporal	Any sequence? Beginning→middle→end?

0.2 Generate three refined variants

Present 3 variants in a table:

| # | 风格方向 | 优化后提示词 | 建议参数 |
|---|---------|-------------|---------|
| 1 | [风格名]  | [完整中文提示词] | 5s / 16:9 / 1080P |
| 2 | [风格名]  | [完整中文提示词] | 8s / 16:9 / 1080P |
| 3 | [风格名]  | [完整中文提示词] | 5s / 9:16 / 1080P |

• Variant 1: 忠于原意 — preserve core idea, add cinematic detail
• Variant 2: 创造性发散 — different artistic interpretation
• Variant 3: 实用主义 — optimized for vertical short-video

Prompt writing rules:

• Write in Chinese; front-load subject + action (first 20 chars matter most)
• Concrete visual nouns ("金色麦田") not abstract concepts ("丰收的感觉")
• Describe motion explicitly ("缓缓推近", "随风飘动")
• Add camera/lighting cues at the end ("电影感镜头", "逆光剪影")
• Keep within 150 characters
• Wan 2.7 multi-shot: 第N个镜头[N-Ns]: 描述 format

0.3 User feedback loop

User says	Action
"用第N个" / "N"	Use variant N as-is
"更诗意" / "更浪漫"	Regenerate with poetic tone
"更简洁"	Strip to essentials
"加动态元素"	Add more motion/action
"改为夜景" / "下雪" / etc	Apply scene change to all variants
"混合1和3"	Combine subject of 1 with style of 3
"直接用" / "不改了"	Skip refinement
Custom feedback	Apply and regenerate

Iterate until the user explicitly approves ("好", "可以", "用这个", "生成吧").

0.4 Skip conditions

Skip refinement when user says "直接生成" / "不用优化" / "skip", the prompt is already detailed (>80 chars), or mode is kf2v/r2v.

Step 1: Decide mode, provider, and model

• Mode: auto-detected from inputs (t2v / i2v / kf2v / r2v)
• Provider: --provider {bailian,jimeng,minimax,hunyuan} or auto-detect from model name
• Model: --model flag, or provider default for the mode
• Parameters: duration, resolution, ratio from variant suggestion or user override

Step 2: Confirm and generate

Show the final command and confirm with the user. Run the script; it blocks until the task finishes and saves the MP4.

Step 3: Deliver

Report output path, file size, and generation time. Save to cwd if no path given.

Cost note: video APIs bill per second of output. Confirm with user for long/many clips.

Providers

Alibaba Bailian 百炼

One API key (DASHSCOPE_API_KEY) covers 5 model families. Third-party models (PixVerse/Kling/Vidu/HappyHorse) are cn region only. Models: Wan (t2v/i2v, up to 15s), PixVerse (all 4 modes, 1-15s), Kling (t2v/i2v/kf2v + r2v on omni), Vidu (q3: 1-16s with audio; q2: 1-10s), HappyHorse (t2v/i2v, 3-15s).

Volcengine Ark (Jimeng 即梦)

ARK_API_KEY via https://ark.cn-beijing.volces.com/api/v3. Seedance 2.0: t2v/i2v up to 15s/2K, with audio, lip-sync, and camera motion. Ratios: 16:9, 9:16, 1:1, 21:9.

MiniMax 海螺 AI

MINIMAX_API_KEY via https://api.minimax.chat. MiniMax-H3 (Hailuo 3.0): t2v/i2v, 4-15s, 768P/1080P (2K via direct API), 24fps with native stereo audio. Legacy video-01: 6s at 720P, prompt optimizer on by default (--no-prompt-optimizer to disable). Host override: MINIMAX_API_BASE (e.g. https://api.minimaxi.com CN / https://api.minimax.io intl).

Tencent Hunyuan 混元

HUNYUAN_API_KEY via TokenHub. hy-video-1.5: t2v/i2v (5-10s, 720P, supports --duration/--seed). Experimental i2v-only: yt-video-2.0, yt-video-fx, yt-video-humanactor. Flags --resolution, --ratio, --audio, --camera-motion are not yet supported.

Model Selection Guide

Use case	Model	Provider
Best quality t2v, multi-shot	wan2.7-t2v-2026-04-25	Bailian
Fast action / combat	pixverse/pixverse-c1-t2v	Bailian
Smart storyboard + audio	kling/kling-v3-video-generation	Bailian
Long clips up to 16s + audio	vidu/viduq3-pro_text2video	Bailian
Douyin/XHS short-video	doubao-seedance-2-0-260128	Jimeng
Smooth motion, natural physics	MiniMax-H3	MiniMax
Animate an image (default)	wan2.6-i2v-flash	Bailian
Transition between two frames	pixverse/pixverse-c1-kf2v	Bailian
Character/subject consistency	pixverse/pixverse-c1-r2v	Bailian
Cheap drafts	wanx2.1-t2v-turbo, happyhorse-1.0-t2v	Bailian
Chinese t2v/i2v on Tencent	hy-video-1.5	Hunyuan
Portrait animation (experimental)	yt-video-humanactor	Hunyuan

Run python scripts/generate_video.py --list-models for the full model catalog.

Model Comparison Page

When the user wants to compare models, browse pricing, filter by features, or pick a model:

Open docs/models.html in the browser. It's a self-contained static page — no server needed. Use open (macOS) or xdg-open (Linux):

open docs/models.html

The page supports filtering by provider, mode (文生/图生/首尾帧/参考生), and features (audio/camera/multi-shot), with pricing and capability comparison across all 24 models.

Usage

# Text-to-Video (Bailian default)
python scripts/generate_video.py "一只柴犬在樱花树下奔跑,花瓣随风飘落,电影感镜头" shiba.mp4 \
  --duration 5 --resolution 1080P --ratio 16:9

# Image-to-Video
python scripts/generate_video.py "镜头缓缓推近,人物微笑" out.mp4 --image portrait.png

# First+Last Frame (kf2v)
python scripts/generate_video.py "花苞缓缓绽放成盛开的牡丹" bloom.mp4 \
  --image bud.png --last-frame bloom.png

# Reference-to-Video (r2v)
python scripts/generate_video.py "@girl 在 @cafe 里弹吉他" out.mp4 \
  --ref girl=girl.png --ref cafe=cafe.jpg

# Other providers
python scripts/generate_video.py "城市日落延时摄影" sunset.mp4 --provider jimeng --duration 10
python scripts/generate_video.py "海浪拍打礁石" ocean.mp4 --provider minimax --duration 6
python scripts/generate_video.py "金黄色的麦田在秋风中起伏" field.mp4 --provider hunyuan --duration 5

# Dry-run: preview request + cost estimate without submitting
python scripts/generate_video.py "一只柴犬在樱花树下奔跑" --dry-run

# Schema introspection (for agents)
python scripts/generate_video.py schema providers   # list all providers
python scripts/generate_video.py schema bailian     # list bailian models with capabilities

# JSON output (auto-detected when stdout is not a TTY; force with --format)
python scripts/generate_video.py --list-models --format json
python scripts/generate_video.py "海边的日落" out.mp4 --format json --provider jimeng

# Resume a task
python scripts/generate_video.py --task-id <task-id> out.mp4

# List all models
python scripts/generate_video.py --list-models

Options

Flag	Meaning	Default
--provider	bailian / jimeng / minimax / hunyuan	auto-detect
-m/--model	model name	auto by mode
-i/--image	first-frame image (path/URL) → i2v	—
--last-frame	last-frame image → kf2v (requires -i)	—
--ref	reference image name=path_or_url, repeatable → r2v	—
-d/--duration	seconds	5
-r/--resolution	360P/480P/540P/720P/1080P	1080P
--ratio	16:9 / 9:16 / 1:1 / 3:4 / 4:3 / 21:9	16:9
-s/--size	exact W*H for size-based models	from resolution+ratio
-n/--negative	negative prompt (Wan only)	—
--no-prompt-extend	disable prompt rewriting (Wan only)	extend on
--no-prompt-optimizer	disable built-in prompt optimizer (MiniMax only)	optimizer on
--audio	enable audio on PixVerse/Kling/Vidu/Jimeng	off
--no-audio	silent output on Wan audio models	audio on
--camera-motion	camera motion (Jimeng Seedance 2.0 only)	—
--seed	reproducibility	random
--task-id	resume polling an existing task	—
--dry-run	preview request body + cost estimate, no submit	—
--format	json or table (default: table in TTY, json otherwise)	auto
--list-models	list models and exit	—
schema <resource>	introspection: providers or a provider id	—

Requirements

pip install requests

Environment Variables

Variable	Required	Provider	Purpose
DASHSCOPE_API_KEY	yes (Bailian)	Bailian	https://bailian.console.aliyun.com/
DASHSCOPE_API_BASE	no	Bailian	cn (default) / sg / us
DASHSCOPE_VIDEO_MODEL	no	Bailian	default model override
ARK_API_KEY	yes (Jimeng)	Jimeng	https://console.volcengine.com/ark/
MINIMAX_API_KEY	yes (MiniMax)	MiniMax	https://platform.minimax.io
MINIMAX_API_BASE	no	MiniMax	API host override (CN: https://api.minimaxi.com, intl: https://api.minimax.io)
HUNYUAN_API_KEY	yes (Hunyuan)	Hunyuan	https://console.cloud.tencent.com/hunyuan

Third-party Bailian models are cn region only.