Threads' five-minute video limit
Threads takes five-minute video. Shorts caps at three; Reels stops recommending past three. Formats that need the extra minutes, and a padding test.
Threads has accepted video up to five minutes since launch. Meta's July 2023 introduction of Threads said posts could include "videos up to 5 minutes in length," and that ceiling is still the one you upload against. YouTube Shorts still hard-caps at three minutes. Instagram Reels is the comparison people get wrong: Instagram's Reels recording help currently lets you record and edit up to 20 minutes, but the same page states that Reels over three minutes will not be recommended to new audiences.
The useful window is three to five minutes. That length is too long to be a Short, past the Reels discovery line, and legal on Threads — in a feed whose baseline unit is still a paragraph of text. The obvious mistake is to treat five minutes as three minutes with padding. The useful move is to find the things that genuinely do not fit under three and were being cut to death everywhere else.
The length landscape
| Surface | Maximum length | What the cap actually does | Source |
|---|---|---|---|
| Threads | 5 minutes | Hard upload ceiling | Meta, July 2023 |
| Instagram Reels | 20 minutes | Over 3 minutes not recommended to new audiences | Instagram Help |
| YouTube Shorts | 3 minutes | Hard upload ceiling | YouTube Help |
| 15 minutes | Not a short-form feed | LinkedIn Help |
LinkedIn is longer, but LinkedIn is not a short-form feed and its audience arrives in a different posture. Against Reels and Shorts, Threads is the surface that will take a four-minute file without hitting a hard cap or a published discovery cutoff.
In a feed of short text posts, a five-minute video is not competing against other five-minute videos. It is competing against the decision to keep scrolling.
The three-minute line is where formats go to die
If you have produced for Reels or Shorts, you know the specific damage a three-minute line does. On Shorts it is a hard ceiling. On Reels it is a discovery cutoff: the file will post, and it will not be recommended to new audiences. Either way you restructure everything upstream to survive the line: compress the setup, cut the second example, skip the caveat, drop the part where you show the thing failing before it works.
Three formats lose their point under that compression.
1. The unbroken demonstration. A process shown end to end, without cuts hiding the boring middle. The value is precisely that nothing was removed, because the removed parts are where the audience suspects the trick is. A two-minute version with jump cuts proves nothing; a four-minute version with none proves the claim. This is the format where length is the evidence rather than the cost.
Works for: physical processes, repairs, cooking, assembly, a software task performed at real speed, anything where "but you cut out the hard part" is the natural objection.
2. The recorded conversation segment. Not a clipped podcast highlight, which is a three-minute format and belongs on Reels. A single exchange with the setup intact: the question, the pause, the actual answer, the follow-up that complicates it. Podcast clipping tools exist because the interesting bit is usually 40 seconds long, but the interesting bit is frequently only interesting because of the 90 seconds before it, and that is a five-minute asset, not a three-minute one. Our guide to clipping a podcast with AI covers the short version; this is the case for sometimes not clipping.
Works for: expert interviews, customer conversations, disagreements between two people who both know the subject.
3. The annotated teardown. Take one artefact, go through it in order, with commentary. A competitor's landing page, a piece of hardware, a contract, a dataset, a UI. The structure is inherently serial and the value scales with how many items you get through. Three minutes is a handful of observations. Five minutes is a list, and a list is where a teardown stops being a hot take and starts being a reference.
Works for: analysis, criticism, education, anything where the audience wants the full list rather than the highlight.
What these share: cutting them shorter removes information rather than removing fat. That is the test.
The padding test
Before you post anything over three minutes, run it.
- Cut it to 2:59 and watch both. If the short version says the same thing, the short version is the video and the long one is vanity. If the short version leaves out something a viewer would need, keep the length.
- Name what the extra minutes carry. In one sentence. "The extra 90 seconds is the second failure attempt, which is the proof." If you cannot name it, it is padding.
- Find the point where nothing new arrives. Scrub through and mark every moment where the next 15 seconds introduces no new information. If you have three such marks, you have a shorter video with intervals in it.
- Check the opening against the whole. A long video needs its first ten seconds to promise a specific payoff, not a vague one. "Here's what happened when we ran it a third time" survives; "let's talk about testing" does not. The opening fails earlier on long assets than short ones, because the viewer is being asked for more.
- Assume no sound on the first pass. Design so the video is legible muted, then let audio add rather than carry. Burned-in captions are not an accessibility afterthought at this length, they are the retention mechanism.
Producing five minutes without a five-minute budget
Generated video does not arrive in five-minute pieces. Models produce short clips, and a long asset is an assembly problem rather than a generation problem. That changes the shape of the work.
Build the spine from real material. For all three formats above, the long element is usually something captured rather than generated: the demonstration, the conversation, the artefact on screen. Generated assets do the connective work, not the substance: cutaways between sections, title cards, establishing shots, transitions. The AI b-roll generator covers that layer without a shoot.
Assemble on one timeline. The editor is EDL-based, which means the five-minute cut is a single re-renderable timeline rather than a sequence of exports you stitch. Change one clip, re-render the timeline. The 480p preview pass is free and carries a short per-user cooldown, so you can check pacing repeatedly before committing; the final export is charged once regardless of how many clips are on the timeline. That economics matters specifically at this length, because a five-minute cut is twenty times the duration of a fifteen-second one and would otherwise multiply iteration cost by the clip count.
Keep the frame rate consistent. Default output is 25 fps. Mixing sources shot at different rates into a long timeline produces judder that nobody notices in a fifteen-second clip and everybody notices in four minutes. Fix the frame rate at the top of the project, not at export.
Caption the whole thing. Five minutes of unpaced text is worse than none. Auto-timed captions from the audio track are the baseline; the AI caption generator handles the timing, and the styling decision is a brand one you make once.
Write to the length before you produce. Long-form generated dialogue drifts if the script is not sized to the segments. The approach in scene direction sized to clip length applies directly: decide the segment boundaries first, then write into them.
If you are generating the primary footage rather than capturing it, model choice narrows quickly at longer durations. Start with the best models for long videos rather than extending a short-clip model past its comfortable range.
Where the long cut goes afterwards
A five-minute Threads video is also the source asset for everything shorter. The teardown becomes a set of standalone observations. The conversation becomes a handful of clips. The demonstration becomes a fifteen-second result shot with the full version linked.
That ordering is the right one: produce long, cut down, distribute wide. Producing short and trying to expand is how padding happens. The mechanics of the down-cut are in repurposing long-form content into shorts.
One caution on cross-posting: Instagram's recommendations and originality guidance treats visible watermarks as a reason content is not eligible for recommendations. If your down-cuts are heading to Reels, export them clean from the timeline rather than pulling them back off another platform. Nothing you export carries a Versely watermark on any plan, so the only watermark risk is one you introduce yourself.
FAQ
Should every Threads video be five minutes?
No. Most should not be. The limit is an option, and the three formats above are the ones that need it. A fifteen-second visual punchline attached to a strong text post is still the highest-hit-rate thing on this platform, and the format work in Threads for brands covers that end.
Is five minutes long enough to move a video off YouTube?
Not usually. Five minutes is a feed format, not a destination format, and the viewing posture in a scroll feed is different from a viewer who chose your video from a results page. Treat the Threads cut as the version someone encounters and the longer YouTube cut as the version someone seeks out.
How do I know if the length is costing me completion?
Compare completion against your own shorter posts rather than against a benchmark. If a four-minute video holds a similar share of viewers to your two-minute videos, the length is earning itself. If it drops off sharply at a consistent timestamp, you have found the point where nothing new arrives. Retention-curve analysis is the same discipline here as on YouTube, just at a smaller scale.
Can I generate a full five minutes rather than assembling it?
Not from a single generation. Models produce clips, and long assets are built on a timeline. Practically that is an advantage, because a timeline is editable and a single long generation is not; you can swap one segment without re-running the whole thing. The video editor is where that assembly happens.