Will a dubbed YouTube Short fit in 3 minutes? Measure before render

A dubbed Short fits if the finished video runs 3 minutes or less, which is YouTube's stated limit for Shorts creation tools, and the only way to know is to measure the dubbed audio before you render. Sume does not enforce that limit for you: its Timeline 1.0 render accepts up to 1,800 seconds, so a 4-minute dub renders happily and then fails the rule on YouTube's side.

YouTube's limit is from its own Shorts help page, read 2026-10-03, which also states a maximum resolution of 1080p. That page says nothing about dubbing or multi-language audio. The Sume limits come from the Timeline 1.0 docs.

Why does a dub run longer than the original?

A translated script is a different length from the source, and speech takes the time the words take. A source that sits at 2 minutes 50 seconds in one language can land past 3 minutes in another once the new narration is read at a natural pace. Sume publishes no speaking-rate figure and no per-language length ratio, so this post gives none; the number that matters is the duration your own TTS job returns.

That makes the check mechanical. Each finished TTS job reports the audio duration_seconds. Compare it with the length of the source video before you build anything on top of it.

What are the ways to make it fit?

Sume documents no time-stretch for video or audio. There is no setting that speeds up the picture to match longer narration, so the video side is changed by trimming slots, not by stretching them.

• Shorten the translated script. This is the best fix, because it removes words instead of squeezing them.
• Nudge the pace. TTS 1.0 accepts generation_config.speed from 0.6 to 1.5; keep the change small, because the extremes sound rushed or slow.
• Cut a scene from the video slots. Timeline 1.0 takes ordered video[] slots, each with its own start and duration, so you can drop or trim a segment.
• Split into two Shorts when the content allows, and render each on its own.

How do you check the render length for free?

Timeline 1.0 has an unbilled plan endpoint, POST /v1/timeline-1.0/plan. It validates the document, compiles it, and returns duration_seconds, billable_minutes and estimated_cost_usd_micros without creating a job or reserving credits. Put the dub's duration in the request, read the answer, and only then submit the paid render.

import os, requests

H = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
body = {
    "audio": {"url": os.environ["DUB_AUDIO_URL"], "duration_seconds": 172},
    "video": [{"source_url": os.environ["SOURCE_VIDEO_URL"], "start": 0,
               "duration": 172, "fit": "cover"}],
}
r = requests.post("https://api.sume.com/v1/timeline-1.0/plan",
                  json=body, headers=H, timeout=60)
r.raise_for_status()
plan = r.json()
plan = plan.get("data", plan)
print(plan.get("duration_seconds"), plan.get("billable_minutes"),
      plan.get("estimated_cost_usd_micros"))

Limits to check against each other

Which languages tend to run long?

Sume does not publish a ratio, and neither does this post, so there is no safe rule such as adding a fixed percentage. What you can rely on is the pattern of your own work: dub one clip per target language once, record each result against the source length, and keep that table. After three or four clips you will know which of your languages needs a tighter script and which fits as written.

Keep the source tight too. A source Short that already runs at 2 minutes 58 seconds leaves no room for a language that needs more time. If you plan to dub, cut the source to leave a margin, and decide the margin from your own measurements.

Can the same video carry several dubbed audio tracks?

Sume's render writes one MP4 with one audio spine, so each language is its own video file and its own upload. This post says nothing about how YouTube treats multiple audio tracks, because the Shorts help page read 2026-10-03 does not cover it. If you use a platform feature for extra audio tracks, read that platform's own page for its limits.

What does it cost to be wrong?

A render at 3 minutes costs $0.30 at the catalog rate, since the price rounds up to whole output minutes. A render that runs 3 minutes 5 seconds bills as 4 minutes, which is $0.40, and it still would not upload as a Short. Measure first, render once.

The same check works for a batch. Run every language's TTS job, list the durations, and only render the ones under your ceiling; send the others back for a shorter script. For the multi-language chain around this step, see dub one Short into eight languages.