AI UNDERDOGSDAILY PICK
AI UNDERDOGS
AI 剪视频，全程不离浏览器
AI Video Clipping, All In-Browser
imgly/videoclipper
大多数 AI 剪辑工具都要你上传视频到他们服务器
VideoClipper 直接把模型塞进浏览器里跑
你的素材一步都不用出本地
Most AI clippers make you upload to their
server. VideoClipper bundles the model right in the
browser — your footage never leaves your machine
它到底是什么
What Exactly Is This
imgly 本身就在做编辑 SDK
这次他们搞了个完全本地化的 AI 视频剪辑器
模型文件放在 public/models 目录
整个推理过程在浏览器内完成
不依赖任何后端服务
上传、等待、下载这三步直接砍掉了
imgly already makes editing SDKs — this time
they built a fully local AI video clipper
The model files sit in public/models, inference runs
entirely in the browser, no backend needed. Upload
wait, download — all three steps eliminated
★ SIGNAL 1
字幕不只转文字
Captions Do More Than Transcribe
他们的提交记录里有一条写的是「改进字幕质量：说话人边界和句子检测」
多人视频不再按时间窗口硬切
而是真的能识别谁在说话、在哪里换人
这个细节，99% 的同类工具都不会做
One commit literally reads 'Improve caption quality: speaker
boundaries and sentence detection.' Multi-speaker videos get properly
segmented captions — not just blunt time-based cuts
99% of similar tools skip this entirely
★ SIGNAL 2
模型全塞进浏览器
Models Packed Into The Browser
AI 模型文件放在项目的 public/models 目录里
直接由浏览器加载运行
整个推理链路完全客户端化
没有一行数据会离开你的电脑
imgly 这次是真的把 local-first 理念贯彻到底了
The AI models live right in public/models —
loaded and run by the browser. The entire
inference pipeline is client-side, not a single byte
leaves your computer. imgly went all-in on local-first
with this one
作者在 Hacker News 上发帖
标题就直接写了「vibe-coded」
意思是边摸鱼边写的
没有产品发布会那种包装话术
就是一个开发者觉得现有方案不够好
自己动手搞了一个
The author posted it on HN with 'vibe-coded'
right in the title — built it while
vibing, no launch-day polish. Just a developer who
thought existing options weren't good enough, so he
made his own
AI UNDERDOGS
开源自部署，去翻翻代码
Open source — go poke around the code
imgly/videoclipper
关注 · 每天发现更多 AI 神作
github.com/imgly/videoclipper