Automatic captioning is one of the few AI capabilities where the free tier is genuinely sufficient. Speech recognition got good enough that the bottleneck moved: the hard part is no longer transcription accuracy, it is caption craft, and that is where nearly every captioned video falls down.
Where we fit, stated plainly: 8frame generates video, not captions. We do not have a transcription or subtitle tool, and this article will not pretend otherwise. What follows is the honest map, because captioning is a step you do after the video exists and it matters more than most production decisions.
TL;DR
- Free is enough. Platform-native auto-captions and open-source speech recognition both produce usable transcripts
- Accuracy is not the problem. Line breaks, timing and placement are
- Two lines maximum, roughly 32 to 42 characters each, split at clause boundaries
- Position in the lower-middle third, never the true bottom: platform UI covers it
- Always read them back once. Names, jargon and numbers are where recognition still fails, and those are the words that matter
What free tiers actually give you
Platform auto-captions. Every major short-form platform generates them on upload, free, and the quality is now reasonable for clear speech. The catch is that you get the platform's styling and its line breaking, and the toggle can be off by default for some viewers.
Open-source speech recognition. Free to run, no upload limits, and it produces clean timestamped output you control. This is the route if you caption at volume or care about the styling.
Editor built-ins. Most video editors now include auto-captioning, free in the base tier, with the styling controls you actually need.
None of these is the constraint. Pick the one already in your workflow.
The craft rules, which are the real work
Two lines, never three. Three lines cover a third of a vertical frame and viewers stop reading.
32 to 42 characters per line. Longer and the eye has to track; shorter and the captions flicker.
Break at clause boundaries, not at character counts. Automatic tools break when they hit a width limit, which splits phrases in the middle and measurably hurts comprehension. "The thing nobody tells you / about pricing" reads. "The thing nobody tells / you about pricing" does not. This single fix improves captioned video more than any styling choice.
Lower-middle third, not the bottom. Platform interface elements (captions toggles, handles, progress bars, engagement buttons) occupy the bottom of the frame and the sides. Anything in the true bottom 15 percent is at risk.
One idea on screen at a time. Captions that run continuously become wallpaper. Let them breathe with the speech.
High contrast, with a plate or stroke. Generated and filmed backgrounds both shift in brightness mid-clip; white text alone will disappear over something bright. A subtle dark plate behind the text survives every background.
More on the surrounding decisions in how to caption video for silent viewing.
The proofread nobody does
Speech recognition fails predictably, and it fails on exactly the words that carry your meaning:
- Proper nouns. Product names, company names, people's names.
- Numbers and units. "$19" becomes "nineteen dollars" or "19" becomes "90".
- Domain jargon. Any term the model has not heard often.
- Accents and fast speech, still.
Reading a 60-second caption track takes 30 seconds. Shipping a video where your product name is misspelled in burned-in text is a self-inflicted wound. Do the 30 seconds.
Where generation does come in
If the video itself is generated, captions still come afterwards in the editor. The one place our side of the work touches captioning is negative: leave room for them. A generated shot composed tight to the bottom of the frame has nowhere for text to sit, and cropping later loses the composition.
So when you prompt a vertical clip that will carry captions, compose for it: subject in the upper two thirds, visual quiet in the lower middle. That costs nothing and saves the reframe. A Wan 3.0 480p draft at 24 credits is the cheap way to check the framing leaves room before committing to a finished take.
FAQ
What is the best free subtitle generator? The one already in your editor or platform. All the free options are accurate enough; none of them fix line breaking, which is the part that matters.
Does 8frame generate subtitles? No. We generate video, images and audio. Captioning happens in your editor.
Do captions really change performance? Short-form autoplays muted, so an uncaptioned talking clip is a silent film. The gain is not subtle.
Should captions be burned in or platform-native? Burned in gives you control and guarantees they appear. Native is accessible and searchable. Many creators do both, native for accessibility and burned-in for styling.
Compose the shot with room for the text. The 8frame canvas is free and unlimited, and generation is paid from $19/month.