Install any skill in seconds. Free to start, no credit card required.
Get Started Free →基于ByteDance agentkit-samples多媒体用例的小省导购员数字人带货视频生成技能,整合多模态内容生成能力(图像、视频、音频),支持AI绘画、语音合成、视频生成,与小省导购员人设融合,9:16竖屏适配,直接对接带货视频生成流程
.claude/skills/anbeime-agentkit-multimedia-shopping/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 76% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 183% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -4% | 0% |
本技能依赖以下Python包:
Pillow>=10.0.0
requests>=2.28.0
numpy>=1.24.0使用AI绘画生成小省导购员角色形象:
输入信息:
生成流程:
输出:角色参考图(9:16竖屏)
使用AI绘画生成场景背景:
输入信息:
生成流程:
输出:场景参考图(9:16竖屏)
使用TTS生成导购员语音:
输入信息:
生成流程:
输出:语音文件(16kHz单声道wav)
使用音乐生成工具生成背景音乐:
输入信息:
生成流程:
输出:音乐文件(16kHz单声道wav)
使用多模态组合生成视频:
输入信息:
生成流程:
输出:完整视频(25秒,9:16竖屏)
功能说明:生成小省导购员数字人带货视频的完整流程
执行方式:
关键参数:
输出:完整带货视频(25秒,9:16竖屏)
功能说明:仅生成小省导购员角色参考图
执行方式:调用generate_character.py
关键参数:
输出:角色参考图(9:16竖屏)
角色固定特征描述
↓
AI绘画生成(角色形象+场景背景)
↓
语音合成(导购员语音+背景音乐)
↓
InfiniteTalk提示词生成(使用infinitetalk-shopping-avatar Skill)
↓
多模态视频生成(图像+音频+提示词)
↓
完整带货视频(25秒,9:16竖屏)本技能与infinitetalk-shopping-avatar Skill协同工作:
协同流程:
本技能生成角色参考图 → infinitetalk-shopping-avatar生成提示词 → InfiniteTalk生成视频
本技能生成语音文件 → chinese-wav2vec2-base编码 → InfiniteTalk使用
本技能生成音乐文件 → chinese-wav2vec2-base编码 → InfiniteTalk使用| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 22,814 | 13,250 | -42% | 1 | 1 | 0% | 2,984 | 4,682 | +57% | 0 | 0 | — |
case-02 | fail→fail | 13,596 | 11,429 | -16% | 1 | 1 | 0% | 2,090 | 3,802 | +82% | 0 | 0 | — |
case-03 | fail→pass | 15,337 | 11,492 | -25% | 1 | 1 | 0% | 2,386 | 4,200 | +76% | 0 | 0 | — |
case-04 | pass→fail | 44,742 | 48,764 | +9% | 1 | 1 | 0% | 3,900 | 3,220 | -17% | 0 | 0 | — |
case-05 | pass→pass | 28,632 | 22,227 | -22% | 1 | 1 | 0% | 5,389 | 7,151 | +33% | 0 | 0 | — |
case-06 | pass→pass | 23,370 | 14,264 | -39% | 1 | 1 | 0% | 4,252 | 4,461 | +5% | 0 | 0 | — |
case-07 | fail→pass | 46,541 | 11,726 | -75% | 1 | 1 | 0% | 1,580 | 4,468 | +183% | 0 | 0 | — |
case-08 | fail→pass | 15,516 | 9,864 | -36% | 1 | 1 | 0% | 2,568 | 3,440 | +34% | 0 | 0 | — |
case-09 | fail→fail | 13,661 | 15,924 | +17% | 1 | 1 | 0% | 2,195 | 4,266 | +94% | 0 | 0 | — |
case-10 | fail→pass | 15,544 | 5,169 | -67% | 1 | 1 | 0% | 2,895 | 2,770 | -4% | 0 | 0 | — |
case-11 | fail→fail | 14,585 | 9,561 | -34% | 1 | 1 | 0% | 2,242 | 3,596 | +60% | 0 | 0 | — |
case-12 | fail→fail | 17,722 | 9,257 | -48% | 1 | 1 | 0% | 2,854 | 3,623 | +27% | 0 | 0 | — |
case-13 | fail→pass | 16,579 | 9,493 | -43% | 1 | 1 | 0% | 2,641 | 3,689 | +40% | 0 | 0 | — |
case-14 | fail→pass | 21,140 | 9,299 | -56% | 1 | 1 | 0% | 3,445 | 3,893 | +13% | 0 | 0 | — |
case-15 | fail→pass | 16,159 | 6,644 | -59% | 1 | 1 | 0% | 2,487 | 3,247 | +31% | 0 | 0 | — |
case-16 | fail→pass | 14,975 | 3,172 | -79% | 1 | 1 | 0% | 2,459 | 2,564 | +4% | 0 | 0 | — |
case-17 | fail→pass | 18,508 | 10,079 | -46% | 1 | 1 | 0% | 2,884 | 3,849 | +33% | 0 | 0 | — |
case-18 | fail→pass | 13,718 | 5,753 | -58% | 1 | 1 | 0% | 2,523 | 2,906 | +15% | 0 | 0 | — |
case-19 | fail→pass | 13,759 | 8,406 | -39% | 1 | 1 | 0% | 2,064 | 3,167 | +53% | 0 | 0 | — |
case-20 | pass→pass | 15,402 | 6,644 | -57% | 1 | 1 | 0% | 2,615 | 3,174 | +21% | 0 | 0 | — |
case-21 | fail→pass | 10,958 | 5,057 | -54% | 1 | 1 | 0% | 2,045 | 3,016 | +47% | 0 | 0 | — |
case-22 | fail→fail | 18,379 | 8,112 | -56% | 1 | 1 | 0% | 2,576 | 3,319 | +29% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +55 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.