Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
半调纸拼贴 B-roll 生成 skill:三闸门审批,Gemini Omni Flash 首尾帧组装动画 | Editorial halftone paper-collage B-roll agent skill
| Date | Stars |
|---|---|
| 2026-07-31 | 897 |
| 2026-08-10 | 1026 |
| 2026-08-18 | 1145 |
| 2026-08-19 | 1154 |
| 2026-08-20 | 1167 |
| 2026-08-21 | 1174 |
| 2026-08-22 | 1176 |
| 2026-08-23 | 1183 |
| 2026-08-24 | 1188 |
| 2026-08-25 | 1194 |
| 2026-08-26 | 1202 |
| 2026-08-27 | 1208 |
| 2026-08-28 | 1213 |
| 2026-08-29 | 1219 |
| 2026-08-30 | 1220 |
| 2026-08-31 | 1225 |
| 2026-09-01 | 1232 |
| 2026-09-02 | 1238 |
| 2026-09-03 | 1240 |
| 2026-09-04 | 1242 |
| 2026-09-05 | 1250 |
| 2026-09-06 | 1251 |
| 2026-09-07 | 1260 |
| 2026-09-08 | 1265 |
| 2026-09-09 | 1268 |
| 2026-09-10 | 1269 |
| 2026-09-11 | 1270 |
| 2026-09-12 | 1271 |
| 2026-09-13 | 1270 |
| 2026-09-14 | 1273 |
| 2026-09-15 | 1275 |
| 2026-09-16 | 1280 |
| 2026-09-17 | 1283 |
| 2026-09-18 | 1285 |
| 2026-09-19 | 1289 |
| 2026-09-20 | 1293 |
| 2026-09-21 | 1296 |
| 2026-09-22 | 1299 |
| 2026-09-23 | 1302 |
| 2026-09-24 | 1305 |
| 2026-09-25 | 1308 |
| 2026-09-27 | 1310 |
| 2026-09-28 | 1314 |
| 2026-09-29 | 1312 |
| 2026-09-30 | 1315 |
| 2026-10-01 | 1315 |
| 2026-10-02 | 1319 |
| 2026-10-03 | 1315 |
| 2026-10-04 | 1314 |
| 2026-10-05 | 1317 |
| 2026-10-06 | 1319 |
| 2026-10-07 | 1322 |
| 2026-10-08 | 1324 |
| 2026-10-09 | 1329 |
| 2026-10-10 | 1333 |
| 2026-10-11 | 1337 |
| 2026-10-12 | 1338 |
| 2026-10-13 | 1355 |
| 2026-10-14 | 1379 |
| 2026-10-15 | 1396 |
| 2026-10-16 | 1404 |
| 2026-10-17 | 1416 |
| 2026-10-18 | 1428 |
| 2026-10-19 | 1439 |
| 2026-10-20 | 1451 |
| 2026-10-21 | 1456 |
| 2026-10-22 | 1471 |
| 2026-10-23 | 1478 |
| 2026-10-24 | 1484 |
| 2026-10-25 | 1506 |
| 2026-10-26 | 1521 |
| 2026-10-27 | 1538 |
| 2026-10-28 | 1550 |
| 2026-10-29 | 1567 |
| 2026-10-30 | 1574 |
| 2026-10-31 | 1581 |
| 2026-11-01 | 1594 |
| 2026-11-02 | 1605 |
| 2026-11-03 | 1620 |
| 2026-11-04 | 1639 |
Today
+19 stars today
This week
+89 stars this week
This month
+322 stars this month
Momentum
115.0
growth rate 5.74%/day
# gbro-collage-broll
<p align="center">
<img src="assets/demo-purple.gif" width="180" alt="深紫底:多人协作压出科幻胶片">
<img src="assets/demo-yellow.gif" width="180" alt="芥末黄底:错误被印刷机批量放大">
<img src="assets/demo-red.gif" width="180" alt="红底:导演之手摆放棋盘走位">
<img src="assets/demo-teal.gif" width="180" alt="青绿底:剪刀裁开镜头轨道">
</p>
把一句约 5 秒的口播文稿,压成一个 sharp visual idea,再生成高级编辑风**半调纸拼贴(halftone paper-collage)组装动画** B-roll。
Turn a ~5s voiceover line into a premium editorial paper-collage assemble-from-empty B-roll clip, powered by Gemini Omni Flash first/last-frame video generation.
## 效果
- 强烈平坦的纯色纸面色场 + 黑白 halftone 照片剪贴 + 彩色卡纸点缀
- 元素从空场逐件滑入、卡位、组装(stop-motion 质感),不是淡入或慢 zoom
- 默认交付 9:16、5 秒、720×1280、24fps、无声 MP4,可直接垫在口播下面
## 工作流:三闸门审批
这个 skill 的核心不是 prompt 模板,而是强制的三阶段审批,让你把注意力花在审美判断上,而不是烧生成费用:
1. **Gate 1 · 隐喻确认** — 只输出视觉隐喻方案(核心意思 / 关键物件 / 底色 / 组装顺序),不生成任何图片视频
2. **Gate 2 · 静帧确认** — 确认后才生成彩色拼贴静帧 + contact sheet,再次等你确认
3. **Gate 3 · 视频生成** — 静帧通过后自动用 `gemini-omni-flash-preview` 做首尾帧组装动画,附完整 QA(逐秒抽帧、首帧空场验证、尾帧对照)
批量模式下支持部分通过:只有确认过的条目进入下一阶段。
## 环境要求
首次触发时 skill 会自动运行 `scripts/check_setup.sh` 自检,并针对缺失项给出配置指引。需要:
| 依赖 | 说明 |
|------|------|
| Codex 环境 | Gate 2 静帧生成依赖内置 `image_gen` 工具 |
| `GEMINI_API_KEY` | [Google AI Studio](https://aistudio.google.com/apikey) 创建,视频生成按量计费 |
| Python >= 3.10 | 用于视频生成脚本 |
| `google-genai >= 2.10.0` | skill 会引导创建共享 venv `~/hyperframes-projects/.omni-venv/` |
| ffmpeg / ffprobe | 首尾帧处理、去音轨、contact sheet |
视频生成脚本(`scripts/generate_video.py` + `scripts/upload_file.py`)已随 skill 自带,无需额外安装其他 skill。
## 安装
把整个目录放进你的 agent skills 目录(例如 `~/.agents/skills/` 或 `~/.claude/skills/`):
```bash
git clone https://github.com/pyang5166/gbro-collage-broll.git ~/.agents/skills/gbro-collage-broll
```
## 使用
对你的 agent 说:
```text
collage b-roll:很多人以为 AI 是来替你思考的,其实它更像一面镜子,会把你问题里的漏洞照出来。
```
触发词:`collage b-roll`、`纸拼贴 b-roll`、`半调拼贴`、`拼贴风格配画面`、`gbro-collage-broll`。
然后按 Gate 1 → Gate 2 → Gate 3 逐步确认即可。批量给多句文稿也可以,每句一个隐喻一条成片。
## 目录结构
```text
gbro-collage-broll/
├── SKILL.md # skill 主文档(三闸门协议 + prompt 模板 + QA 标准)
├── agents/openai.yaml # Codex interface 配置
├── evals/evals.json # 四条闸门行为评测
└── scripts/
├── check_setup.sh # 首次使用环境自检
├── generate_video.py # Gemini Omni Flash 批量视频生成
├── upload_file.py # Files API 上传辅助
└── generate_veo_first_last.py # 旧 Veo 链路(仅兼容保留,默认不用)
```
## FAQ
**为什么强制两次人工确认?**
错误的隐喻或静帧直接进视频生成,浪费的是真金白银的 API 费用。Gate 1 改文字是免费的,Gate 2 重生一张图远比重跑一条视频便宜。
**成片首帧边缘露出一点纸片?**
轻微的可以接受;严格空场需求建议用可编辑时间线的动画工具补前段。
**能换视频模型吗?**
默认固定 `gemini-omni-flash-preview`,明确指定其他模型时才切换。
Excerpt of 2,620 characters
Read on GitHub3
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:67951a0b20a66d10, topic:video-generation, readme:video generation