將長影片剪成 Shorts 要多久?RemixViral 實測及其他工具的速度説法
How long does it take to turn a long video into Shorts? Our measured jobs, and what other tools claim
RemixViral 公開了 3 項長影片剪輯工作的數據,其中 2 項完成,並以一項 17 分 58 秒的影片工作展示候選片段排序及剪輯的耗時。這項工作在 222 秒後產生 10 個候選片段,首段成片再等 40 秒,全部成片則在 962 秒後完成。OpusClip、Klap 和 Vizard 公開了相對速度説法,但沒有提供每分鐘來源影片所需秒數;文章建議用相同影片測量各工具。
Every clipping tool says it is fast. None of the three clipping tools we looked at publishes how fast it is per minute of source video. So here is what we can measure ourselves, with the caveat up front: our database holds exactly three long-video jobs, two of which finished. Treat this as a worked example and a method, not a benchmark.
TL;DR
- One 18-minute upload (1,078 s) got a ranked shortlist of 10 moments in 3 min 42 s. The 10 finished 9:16 clips then arrived one at a time, the first 40 s after the shortlist and the last 16 min 2 s after. Start to last clip: 19 min 43 s, close to the video's own runtime.
- The sample is tiny. RemixViral has run 3 long-video jobs in production: 2 finished, 1 failed (the source was under a minute). We publish all three.
- OpusClip, Klap and Vizard publish relative claims ("5x faster", "10x faster than manual clipping", "a matter of minutes") and no seconds-per-source-minute figure. We did not run their tools, so we compare nothing against them except what their own pages say.
Every job we have measured
| Job | Source length | Clips asked / returned | How the video came in | Time to shortlist | Result |
|---|---|---|---|---|---|
| A (23 Sep 2026) | 9 min 57 s (597 s) | 3 / 1 | YouTube link, existing captions used (this path has since been removed) | 617 s | Finished; 1 clip, first file 74 s later |
| B (17 Oct 2026) | 17 min 58 s (1,078 s) | 10 / 10 | Uploaded file, transcribed with Whisper in 2 chunks | 222 s | Finished; 10 clips |
| C (23 Sep 2026) | under 60 s | 3 / 0 | Link import (no longer supported) | 42 s | Failed: "Source is under 1 minute, nothing to clip out of it" |
"Time to shortlist" is the wall-clock time from creating the job to the moment ranked clip candidates were saved. It includes any wait for our job scheduler, because that is what a user experiences.
Job A should not be compared with job B. It used the old link-import path, which read existing captions instead of transcribing, and it ran earlier in the pipeline's life. RemixViral now accepts uploads only. Job B is the only measurement of the current path: about 4.9 seconds of video processed per second of waiting to reach the shortlist.
Where job B's time went after the shortlist
The shortlist is not the end. Each clip is then cut, reframed to 9:16 and captioned by a render worker shared with other jobs, one clip per scheduler tick by default so a burst of jobs cannot starve everything else. For job B that looked like this:
| Clip rank (best first) | Clip length | Finished, seconds after shortlist |
|---|---|---|
| 1 | 21.8 s | 40 |
| 2 | 22.3 s | 85 |
| 3 | 39.1 s | 176 |
| 4 | 27.8 s | 212 |
| 5 | 57.2 s | 443 |
| 6 | 43.5 s | 436 |
| 7 | 24.4 s | 563 |
| 8 | 20.3 s | 688 |
| 9 | 32.7 s | 810 |
| 10 | 36.8 s | 962 |
(Ranks 5 and 6 finished in the opposite order, 443 s and 436 s.) The 10 clips total 326 s of footage, 30% of the source. Average spacing between finished clips was 96 s, three times the average clip length, so rendering is paced by the queue, not by clip duration. The upside is that the best clip is available first; the cost is that "all 10 done" is a long wait. Whether a clip's file is ready is the number to watch, not the shortlist.
What the free clip covers
Read from remixviral.ai on 2026-11-17 and from our code:
- One free clip project per account, no card. Sign in with Google, upload one video you own, up to 10 minutes, and get up to 3 ranked clips.
- Uploads only (MP4, MOV, MKV or WebM); there is no link import. The free upload limit in code is about 1.5 GB.
- The limits are enforced on the server. There is also a daily cap on new free projects, so on a busy day the offer can run out.
- After that: 150 credits for $2.99, one time, about 3 more clip projects; no subscription. Sources up to 120 minutes; sources over an hour cost more.
From job B's timings, its first three clips were ready 176 s after the shortlist. That suggests the free project, with a 10-minute source and 3 clips, should land in a few minutes, but that is an inference: no free-project job has run yet in our database (the offer started on 14 November 2026), so we have not measured it.
What other tools publish about speed
Claims, read from each tool's public page on 2026-11-17. We did not log in, run their tools or test any of these statements.
| Tool | What the public page says about speed or limits | Source |
|---|---|---|
| OpusClip | "Create and publish videos 5x faster with OpusClip's web app and API" (workflow section). A testimonial on the page says clips that took days now take minutes. FAQ: a free-forever plan has 60 minutes of video processing time a month; new users get a 7-day Pro trial with 90 minutes (about 30 downloadable clips) | https://www.opus.pro/ |
| Klap | "In a matter of minutes, we will give you multiple viral-worthy clips". Rule of thumb: a one-minute video produces about 5 clips. Free on 1 video, no card | https://klap.app/ |
| Vizard | "10x Faster than manual clipping"; "Turn long video into 30+ clips, in one click"; "Try it free. No sign up needed" | https://vizard.ai/ |
All three express speed relative to doing it by hand or in "minutes". None gives seconds per source minute or a failure rate. That is a gap anyone can fill for their own use with the protocol below.
Benchmark any clipper yourself in 20 minutes
- Use the same video file for every tool, in three lengths if you can (about 5, 15 and 45 minutes).
- Write down four timestamps: upload finished, shortlist shown, first finished clip downloadable, last finished clip downloadable.
- Note failures, retries and anything you had to click to continue.
- Report speed as seconds of waiting per minute of source for the shortlist, and separately for the last clip.
- Watch every clip with sound before posting. Fast and wrong costs more time than slow and right.
How to pick good clips from the shortlist
Job B's shortlist shows what a tool returns, not which clips will perform. Its 10 moments ran 20 to 57 s (median 30 s), together covered 30% of the source, and started anywhere from 0:18 to 17:03. Their internal scores ranged from 75 to 85, which is a ranking inside one video, not a prediction of views. A practical way to choose:
- Keep clips that make sense with no context. If you need the sentence before it, extend the start. See our guide on which part of a long video to clip.
- Check the first 3 seconds. A question, a number or a claim beats "so anyway".
- Check the last line resolves. Auto cuts sometimes land a beat early.
- Trim to the platform's limit. YouTube says Shorts can be up to 3 minutes long (YouTube Help); check each platform's current limit before cutting.
- Watch each clip once with sound. Tone and reactions do not show up in a transcript.
FAQ
How long does a 1-hour video take?
We have not measured one. Our code comments describe a 60-minute source as needing "minutes" of transcription and about 20 minutes of CPU, which is a design estimate, not a timing. The one current-path measurement is the 18-minute job above.
Is the free clip really free?
Yes: no card, one project per account, source up to 10 minutes, up to 3 clips. After that it is credits, one time, with no subscription.
Do other tools take longer than RemixViral?
We do not know. We have not run them, and their pages publish only relative claims.
Why do clips finish one at a time?
The render stage runs one clip per scheduler tick by default, on a worker shared with other jobs, so a burst of clip jobs cannot starve ordinary video renders.
Can I paste a YouTube link?
No. RemixViral takes uploads of videos you own. Link import was removed.
How many clips can one video produce?
Up to 20 ranked moments per source, each with timestamps and a one-line reason. The free project returns up to 3.
Methodology and data
- Data date: queried 2026-11-17 from RemixViral's production database, read-only. The jobs ran on 23 September and 17 October 2026.
- Included: every row in the long-video job table (3 rows, 2 accounts) and every clip row (11 clips, all finished).
- Excluded: nothing. No titles, links, transcripts or account details are shown.
-
Time to shortlist: job
updatedAtminuscreatedAtfor finished jobs. Clip time: clipupdatedAtminuscreatedAt, wherecreatedAtis when the shortlist was saved. Both include scheduler waits. - Limits: n = 2 finished jobs from 2 accounts; one job on a path that no longer exists; no free-project jobs yet; production load at those times unknown. Do not extrapolate to longer sources.
- Competitor text was read from each public page on 2026-11-17 and quoted as a claim.
Disclosure: we make RemixViral.
來源:DEV Community · dev.to