There are two kinds of subtitle tools for short vertical video. The Reel specialists (Submagic, OpusClip, Captions, Kapwing) burn animated captions into the picture and are built for 9:16. The editors with subtitle functions (CapCut, Adobe Premiere Pro, DaVinci Resolve) give you a timeline, a file export and full styling control. Every vendor advertises a high accuracy figure, and none of them breaks it down by language. The practical difference sits elsewhere: whether a correction sticks, whether a caption style can be saved, whether the free tier stamps a watermark or caps the length, and whether you end up with a subtitle file or only a finished video. All prices here: as of September 2026, vendor figures in US dollars, varying by platform, region and tax.
I use none of these for my own Reels; at my end the subtitle is a fixed work step with a local speech model and a check against the script. I went through them anyway, because almost everyone starts there. Everything I collect on AI at work lives on the AI page. How automatic subtitles get made in the first place is in Auto-Generate Subtitles; here I compare the tools themselves.
Why recognition accuracy is the wrong question
Submagic advertises "99% caption accuracy" across "123+ supported languages", OpusClip claims "99% Transcription accuracy" across "20+ Languages", and Submagic's language page adds that its own system is "4.77% better than Amazon AI or AssemblyAI in all languages". Those are vendor claims, not verified measurements, and none of them is broken down by language or by dataset. For German a figure averaged over all languages says very little, because the errors almost always sit in the same places: proper nouns, technical terms, compound words.
CapCut puts it more plainly on its own resource page than any percentage does: review the generated text and "correct possible errors", especially proper names, technical terms and specialised vocabulary. How good the recognition really is cannot be read off vendor figures, and the difference appears afterwards anyway. Three questions, three sections:
- Does a correction stick, or do you retype the same proper noun in every video?
- Can the caption style be set once and reused?
- Does the text stay readable over moving footage?
Pricing, free tiers and watermarks
The free tiers differ far more than the paid ones: if you only want to try this out, you often land on CapCut or Captions. CapCut generally exports manually built projects with free assets cleanly, but template outros, premium or stock assets and the chosen export route can still trigger a watermark or a paywall. Submagic skips the free tier but, per its pricing page, applies no watermark on any tier. You pay from the first video and get a result you can publish from the start.
Length caps rarely make the front page and still decide the question. Submagic caps video length per tier: 2 minutes on Starter, 5 on Pro, 30 on Business, plus 15, 40 or 100 videos a month, with 4K only on Business. Kapwing caps free exports at four minutes and 720p. Both are plenty for a short Reel and not enough for a podcast recording you want to cut clips from. Meta's ads guide allows Reels up to 15 minutes in the ad placement; for organic Reels, check the current limit in the app. In either case, the tool may cap you sooner.
German, styles, branding and export
Export is the column people skip most often. Submagic and CapCut hand you the subtitle file, so you can publish the same video burned in on Instagram and toggleable on YouTube. OpusClip and Captions hand you only the finished video. For a Reel in the feed that makes no difference; on every reuse, the file saves you a full pass.
Proper nouns: correct once instead of every time
The function that makes the biggest difference week to week is called something different by every vendor. A permanent dictionary that holds spellings across videos is described by Submagic and OpusClip in their help pages.
Submagic keeps its own dictionary: you enter "unique terms, brand names, and industry-specific vocabulary" and the system is meant to hit those spellings from then on. The example in the help article: "Cap kut" becomes "CapCut".
OpusClip calls it Brand Vocabulary: you double-click a misheard word in the editor and tick "Add to brand vocabulary", or enter the terms up front in the Asset Library under "Add proper nouns". On the effect, the help page says: "All new clips going forward will transcribe the terms you added correctly." Clips already generated have to be reprocessed; within one clip, "Apply to all" corrects every occurrence at once.
For CapCut and Captions I found no permanent dictionary in the sources I checked; what is described there is the per-video route through the editor: tap the wrong word, type the right one. Fast, and the video in front of you is done.
Locally with Whisper you hand the spellings over via --initial_prompt; the route is covered in detail in the guide to automatic subtitles. The advantage: the list is a text file in your own project, not a record in a vendor's account.
The arithmetic behind this is simple: if the same five names appear in every video, a dictionary saves five corrections per video. At three videos a week that is roughly 780 corrections a year that never happen. Not a forecast, multiplication.
Setting the style once
A caption style is font, size, colour, outline, position and animation: six settings that should be identical in every video so your channel looks like one channel.
CapCut has "Apply to all captions" for this: one style on every automatically generated caption in the project, in the web, desktop and mobile versions. One detail with teeth sits in CapCut's own help article: captions pasted manually or imported as SRT do not respond to it. So anyone who generates the file elsewhere and only wants to style it in CapCut loses exactly the function they came for.
OpusClip stores the look in a brand template, including font, colour, position, animation, logo and call to action; custom fonts come in through the Asset Library. Submagic offers its brand kit from Pro up, not on Starter. Captions gives you exactly one template on the free tier.
Adobe Premiere Pro works with track styles for the whole caption track, DaVinci Resolve styles through track styles in the Inspector; both routes are in the questions below. In Resolve, automatic generation sits in the Studio version at $295 one off. What both can otherwise do, and cost, is in AI Video Editing Software Compared.
Locally the style is not a menu but a parameter: ffmpeg renders subtitles into the picture through libass, and force_style takes the ASS style fields. Your brand specification sits in one line of script instead of a click sequence, and it is automatically the same again on the next video.
Readable over moving footage
The point where most Reels fail, and the one that appears in no feature list. Two adjustments and one visual check.
Position. The only published percentage figure comes from Meta, and it sits in the Ads Guide for the Instagram Reels ad placement: leave at least 14 percent at the top, 35 percent at the bottom and 6 percent on each side free of text and logos. For organic Reels, Meta publishes no figures of its own, and how much the interface covers depends on app version and device. So I take the ad figures as a conservative template and measure once on my own device: upload the Reel, look at the preview, see what the interface, the description and the profile picture actually cover. That becomes your template, and the captions move up once inside the saved style. If you would rather not measure alone, bring the screenshot into the community; people there check it on other devices.
Contrast. White type on moving footage is readable sometimes and not at others, depending on what happens behind it. Two ways out: a strong outline, or an opaque box. In the ASS style fields that libass, and therefore ffmpeg, understands, that is a single value: per the libass project's format guide, BorderStyle is either 1 for an outline following the shape of the glyphs or 3 for rectangular boxes. Alongside it, Outline and Shadow for thickness, OutlineColour and BackColour for the colours, MarginV for the margin. In the Reel tools the same settings are called outline, background and position.
The test: watch the video on your phone, in daylight, with the sound off. If you have to read a line twice, go through seven points: position, contrast, type size, dwell time, line length, line break, animation. And read the word itself again, sometimes the recognition simply got it wrong.
When this comes around every week
Everything above is a tooling choice, and for one Reel a week that is the right level. From the third onwards it becomes a staffing question, and I answered it this way: not another app, but a fixed role. Eddi is my AI employee for video editing. He gets raw footage and a script and delivers a finished Reel: pauses and false starts cut out, subtitles on, overlays placed, 9:16.
What runs at his end is the local route from the tables above: a Whisper model on my own machine produces the text, ffmpeg cuts, Remotion builds the transitions and overlays, all on the command line. Then comes the step that sets him apart from a subscription: he checks the transcript against a second one before a word goes into the picture. Only then does he burn the subtitles in, in the brand style. After every encode a black frame scan runs, and that one came out of a complaint from me, not out of a manual.
The technology is the same as in a subscription. The difference: the check is built in and does not hang on whether anyone still feels like proofreading forty caption lines on a Friday evening. How a role like that gets built is in Hiring an AI Employee: The Process, and how the work is divided up at my end is in My AI Workforce. The idea, the hook and the script are a separate matter: Can AI Create Instagram Reels?.
Frequently asked questions
Which app is best for subtitles on Reels?
It depends on what should happen after recognition. For animated captions with a saved brand style, Submagic and OpusClip are built for it. For free use with a file export, CapCut is the obvious choice as long as you use free assets and a watermark-free export route. For full control without running costs, take the local route via Whisper and ffmpeg.
Can CapCut auto-generate subtitles and export them as SRT?
Yes, both. CapCut generates captions by speech recognition and outputs them separately from the video as an SRT or TXT file; in the export you pick the caption export rather than the video export. CapCut names one condition itself: only automatically recognised captions can be exported, not text typed in by hand.
Are there free subtitle tools for Reels without a watermark?
Yes. CapCut is free in its base form, and manually built projects with free assets generally export without a watermark. Check the preview when using templates, premium or stock assets, template outros or a different export route. Captions states on its own help page that it applies no watermark on any tier and gives the free tier 60 to 200 lifetime credits. The caption sticker inside Instagram costs nothing, and locally Whisper is free under the MIT licence.
Can DaVinci Resolve generate subtitles automatically?
Yes, but only in the Studio version at $295 one off. Styling runs through track styles in the Inspector, with font, colour, size and position; Resolve either renders the subtitles into the picture or exports them as TTML, SRT or VTT. The free version has no automatic generation.
How do I set one caption style for all captions in Premiere Pro?
Through track styles. You style a single caption, create a style via the plus in the track style area, and apply it to the entire caption track. Change it later and every caption updates in one go. "Save to Local styles" makes the same style available in further projects.
How to carry on
Take your last Reel and check three things before you switch tools. First: were the words you corrected by hand the same ones as in the Reel before? Then you need a dictionary, not better recognition. Second: does your caption sit in the bottom third, where the interface and the description sit on a phone? Then move it up and save the style. Third: can you get an SRT file out? If not, you are doing every reuse twice.
After that you will know whether your problem is the tooling or the repetition. If it is the repetition, the next step is a fixed role rather than the next subscription. The templates for that live in my community, together with the people who have already walked the same road.
