● AI auto-clipper for Windows

Long videos in. Viral shorts out.

Paste a YouTube, Twitch or podcast link. Shortsmith finds the moments people replay, reframes them to 9:16 around whoever is talking, and burns in word-perfect animated captions. All on your own PC.

Windows 10 & 11Runs locallyNVIDIA GPU optional1080×1920 export
The Long Table Podcast · Ep. 21401:42:10 · 1080p
00:0025:0050:001:15:001:42:10
How it works

One link. Four steps. Zero timeline scrubbing.

Every clip goes through the same pipeline you would do by hand, minus the two hours in an editor.

STEP 1 · IMPORT

Paste a link or drop a file

YouTube, TikTok, Twitch, X, Vimeo or a local MP4. Up to 4K source, or several links at once to queue them.

STEP 2 · TRANSCRIBE

Every word, timestamped

Whisper speech recognition with word-level timing and proper punctuation, in 90+ languages.

STEP 3 · FIND MOMENTS

Scores the whole video

Ranks every possible clip by hook, payoff, energy and audience replays, then cuts on real sentence boundaries.

STEP 4 · CREATE CLIPS

Reframed, captioned, done

9:16 at 1080×1920 with animated captions, loudness normalized and ready to upload.

Moment detection

It knows which 40 seconds people actually replay.

Shortsmith reads the whole transcript and listens to the audio, then scores every candidate clip from 0 to 100. Clips start on a hook and end after the payoff, with intro music and dead air trimmed off.

HookQuestions, bold claims and "here's why" openers in the first sentence.
EnergyLoudness peaks, laughter and fast delivery compared to the rest of the video.
ReplaysYouTube's most-replayed data, when the video has it.
Claude AIOptional. Add an Anthropic API key and Claude picks moments and writes titles and hashtags.
Nobody tells you this about your first hire
Strong opening hook · Sparks curiosity · Lands a payoff
12:31 → 13:06 · 35s
98
"I was three weeks from losing the house"
High-energy delivery · Emotional language
41:02 → 41:49 · 47s
94
The 2am rule that fixed my sleep
Most-replayed moment · Fast-paced
1:07:44 → 1:08:12 · 28s
91
Why most advice about money is backwards
Strong opening hook · Complete thought
1:31:10 → 1:31:52 · 42s
86
Animated captions

Twelve caption styles. Every word on beat.

Captions are timed to each spoken word, so the highlight lands exactly when it's said. Pick a style below to see it. In the app you can also change the font, size, colors, outline, position and animation, or fix any word and re-render in seconds.

14 fontsword-level timingpop · bounce · fade · slidecolor · pill · karaoke highlight.srt export
Smart reframe

Landscape in, perfectly framed vertical out.

Face tracking follows whoever is on screen with a smooth virtual camera that only moves when it has to. Two people far apart get a split screen. Slides and b-roll get the full frame over a blurred background, so nothing is cut off.

The crop window glides to the active speaker and holds still between moves, so the shot never jitters.

Discover

Never wonder what to clip next.

Every time you open the app, Discover scans 60+ popular podcasts, streamers and creators plus trending YouTube searches. You get around 450 long-form videos with real momentum, ranked by clip potential. Hover one and press Clip it.

38%Momentum: views per day since upload
18%Freshness: new uploads ride the trend
17%Popularity: total views
15%Length sweet spot: 20 to 150 minutes
12%Category: talk-heavy content clips best
The details

Built for people who clip every day.

GPU acceleration

One click installs NVIDIA support. The most accurate speech model transcribes a 14-minute talk in about a minute on an RTX 3050.

Edit, then re-render

Fix any caption word, swap styles, nudge the start and end, or change the reframe mode. Re-renders take seconds.

Queue and walk away

Paste several links or press Clip it on a few Discover picks. They run one after another, with progress on your taskbar.

Platform-ready audio

Every clip is mastered to −14 LUFS, the loudness TikTok, Shorts and Reels play at.

Export your way

Save all clips to a folder, copy titles and hashtags, export captions as .srt, or jump to the original at the clip's timestamp.

Private by design

Video, audio and transcription stay on your PC. An optional API key is encrypted with Windows DPAPI.

Specs

What you need.

SystemWindows 10 or 11, 64-bit
Memory8 GB RAM, 16 GB recommended
GPUOptional. Any NVIDIA card with CUDA 12 for faster transcription and NVENC encoding
DiskAbout 500 MB for the app, plus a speech model of 145 MB to 1.6 GB
SourcesYouTube, TikTok, Twitch, X, Vimeo, Instagram, local .mp4 .mov .mkv .webm
Output1080×1920 or 720×1280 H.264, AAC 192 kbps, −14 LUFS
Languages90+ with automatic detection
FAQ

Questions.

Does it upload my videos anywhere?

No. Downloading, transcription, reframing and rendering all happen on your computer. The only optional exception is Claude AI mode, which sends the transcript text to Anthropic to pick moments.

Do I need a graphics card?

No. Everything runs on the CPU. With an NVIDIA GPU, transcription is several times faster and encoding uses NVENC.

How accurate are the captions?

You choose between four accuracy levels, up to Whisper large-v3-turbo. Words are individually timed, punctuated and capitalized, and any word can be corrected before re-rendering.

How long does a video take?

A 14-minute talk became four finished clips in about a minute and a half on a mid-range PC, including the download. Longer videos scale with their length.

Can I clip any video I find?

Only clip videos you have the rights to use, or that the creator allows. Many podcasts and streamers run official clipping programs, and those are a good place to start.

Shortsmith 1.0

Your next 10 shorts are already in that podcast.

Download for Windows Shortsmith-win64.zip · 195 MB · v1.0.0

Unzip anywhere and run Shortsmith.exe. The speech model downloads on first use.