TemaShortTurn your long videos — podcasts, streams, courses, talks — into vertical shorts. The 9:16 frame follows the speaker's face, and captions are made for you.

The vertical frame follows the speaker's face

Play your video and press “From here (I)” and “To here (O)” to mark each part. Each part becomes one short, and you can add as many as you like. TemaShort analyzes only those parts, finds the faces, and moves the 9:16 frame with the person — by default, the one shown longest and largest.

If the frame picks the wrong person or loses a face, switch person or hold the frame from any point in the timeline, and drag the frame where you want it. Zoom can be Normal, 1.25× or 1.5×.

Profiles, dark scenes and very small faces may be missed. Cropping a 9:16 frame from a Full HD (1920×1080) video enlarges a 608×1080 area to 1080×1920, so the result is a little softer; 4K sources stay sharp.

Two people? Stack them top and bottom

Choose the layout for each short: Follow 1 (crop to one face), 2 stacked (one person on top, one below — good for interviews and co-hosted shows), or Whole frame (keep the full 16:9 picture with a blurred background above and below).

Switch to whoever is speaking

Press “Switch to the speaker automatically” and TemaShort tells the speakers apart by voice, matches each voice to a face, and adds a mark wherever the speaker changes, so the frame cuts to whoever is talking. You can change the voice-to-face matching and edit or delete any mark.

Under Trim, you can also cut silences (trim the pauses in speech) and cut the parts where the person you're following isn't visible. Every cut can be restored one by one.

Similar voices, or two people talking at once, can switch at the wrong moment. Fix those spots with the marks.

Captions made from the speech, plus a title bar

Press “Add subtitles (transcribe)” and TemaShort transcribes the speech in each short and turns it into captions. Set the spoken language to English, Japanese or Chinese (shown in Simplified Chinese), or let it detect the language. The first time, it downloads the transcription model (about 574 MB).

Edit the text and timing in the list, shift everything at once, split or join lines. Pick one of four templates and three sizes, and drag the captions up or down in the preview — “Show areas that may be covered” marks where platform buttons tend to sit. A title bar of up to two lines can go at the top. You can also export the captions as an .srt file.

We measured the captions on test recordings made with synthetic voices: an average word error rate of 2.3% across four English recordings, 0% on a two-person English meeting (both speakers correctly separated), and a 0% character error rate on one Chinese recording; automatic language detection picked the right language each time.

Those figures come from clean synthetic test audio. Real recordings — with different voices, background noise and speaking styles — will have more errors, so check the captions before you export. There's no translation feature.

Tell Claude the timestamps, get captioned vertical shorts

The full version can be driven from Claude Code or the Claude desktop app on your PC. Ask “Turn 0:05–0:15 and 0:30–0:40 of this episode into vertical shorts with captions” and TemaShort opens (even if it was closed) and does the work: it follows the speaker's face, makes the captions and exports the shorts. You can also ask for a layout (follow, stack or fit), switching to the speaker, cutting silences, or .srt files. It uses MCP, the standard way to connect AI assistants to apps.

In our development test, we asked the real Claude Code to “make captioned vertical shorts from 0:05–0:15 and 0:30–0:40”, and it exported two shorts (1080×1920, 10 seconds each) in 107 seconds. Times depend on your PC and video.

Claude can use five tools: make_shorts (make shorts from time ranges), transcribe_video (write a timestamped .srt), export_shorts (export shorts you saved in the app), inspect_video and license_status. Not sure which parts to clip? Ask Claude to “transcribe minutes 30 to 60 and suggest three parts that would make good shorts”.

While it works, a bar at the top says “AI at work”, with a Stop button, and you can watch it move through Framing, Subtitles and Export. The shorts stay in your project, so you can fine-tune the frame or captions and export again. Shorts are written to a temporary .part file and renamed at the end, so if you press Stop, nothing is left behind — not even the shorts already finished. If a file with the same name exists, it stops before starting. Your original video is never changed.

By default, TemaShort doesn't send the captions or video frames to Claude. It returns only the exported file locations, time ranges, lengths and number of caption lines. Caption text is returned only when you ask for it (return_text) — for example, when you want Claude to suggest clips. And if Claude opens an exported video or .srt file, its content is sent to Claude (an AI service). For private recordings, give Claude the timestamps yourself and don't ask it to read the content.

You set it up once. In TemaShort, choose Help > “Use from AI (register with Claude)…” to copy a one-line command for Claude Code, or the config JSON for the desktop app's Chat tab, tailored to your PC.

claude mcp add --scope user temashort -- "<install folder>\TemaShort.exe" --mcp

For scripts, the same is available from the command line (TemaShort.exe --cli or temashort.cmd make, full version).

Making shorts, transcribing and exporting from AI (Claude) or the command line is a full-version feature. With the free version, TemaShort reads nothing, doesn't even open its window, and Claude is told: “Making, transcribing and exporting shorts from AI is a full-version feature. In the TemaShort window, the free version lets you try every feature (exported shorts get a watermark).” It may also work with other AI tools that support MCP (we've tested it with Claude). Claude Code and the Claude desktop app are separate products, and Claude is not included in the price of TemaShort. See our guide How to have Claude turn a long video into vertical shorts and the “Use from AI” chapter of the user guide (PDF).