Caption tool disrespects "multiline/single line" and "word frequency" settings

[Untold Entertainment Inc.]

1. From the Transcript tab in the Text panel, click the hamburger menu and choose Create Captions.
2. In the Create Captions dialog, click the blue arrow to expand "Captioning Preferences"
3. Slide the "Minimum length in characters" and "Minimum duration in seconds" sliders all the way to the left
4. Under "Lines," choose "Single"

Expected: 
The intent is to have a single-word caption for each word spoken.  No captions should be on multiple lines. No captions should contain more than one word.

Actual:
Captions frequently (and randomly?) occupy multiple lines, despite me explicitly telling PPro i want single lines. 
Captions frequently (and randomly?) contain more than one word, despite me dragging those sliders all the way to the left.

i read this response to someone else having this problem:
https://community.adobe.com/t5/premiere-pro-discussions/captions/m-p/13311192

... but the size of the text box is not the issue. Premiere Pro will frequently put two words in a single caption, and will frequently multiple-word captions on multiple lines. If i go in and delete the line break, the words can clearly fit on a single line. But there shouldn't even BE multiple words in a caption to begin with, because i have the sliders pulled all the way to the left, and the software should be smart enough to know that a SPACE character indicates a separate word.

The net result is that instead of "set it and forget it," i have to spend a lot of time combing through my entire video to make numerous annoying little tweaks to the captions to break up multiple words and multiple lines. i shouldn't have to.

[Stan Jones]

@Untold Entertainment Inc.,

Upvoted although I can argue pro/con on whether it is a bug per se. At a minimum it is frustrating that there is not more control. But captions are not designed for single word output - until you use the increasingly common one word/animated captions!

Thanks for focusing on this. I had not tested for some time. At the moment, I am in PR Beta 25.3.0.34.

I agree that the best option for getting one word caption segments is Max characters 7, Min duration 1.2, and Single lines.

In a 3:46 sample that I have used for testing in several versions with a horizontal/landscape frame, I get a very high percentage of single words and zero captions that are not single line. But there are many two word exceptions ("a software" or "at Adobe") and a few short phrases ("I had a bunch of them").

But there is a lot of clean-up for a strict one word style.

See my comments about "each option cannot be primary" here:

https://community.adobe.com/t5/premiere-pro-discussions/create-captions-minimum-duration-in-seconds-...

Stan

Edited 4/24/25 to correct "Min" characters to Max characters.

[Untold Entertainment Inc.]

Thanks, Stan. Speaking as both a PPro customer and a software developer, if an option presented to the user says "Don't make this red," and the user clicks it, and the response from the software is to make something red anyway, that's most assuredly a bug. There's really no other way to spin it.

[jamieclarke]

Hi @Untold Entertainment Inc. - Thanks for submitting your bug report. We need a few more details to try to help with the issue.
Can you let us know your computer specs and what version of Premiere you are on?

Sorry for the frustration.

[Untold Entertainment Inc.]

Hi, Jamie.

Windows 11 24H2
Premiere Pro 2025 v 25.2.0 Build 147
GeForce RTX 4090. Driver version 572.83

There's something programmatically interesting happening, i've noticed: certain specific phrases tend to be grouped together without fail. The words "in the comments" seem to reliably ignore the captions setting and group together on one line, along with "let me know." But it might just be that the speaker says them quickly, and the caption setting slider can't be adjusted below 1.2s? (Although with that setting, many/most words are still displayed singly, despite taking far less than 1.2s to utter.) So it seems to be picking and choosing which words and phrases to group?  No idea about the double-line issue tho.

- Ryan

[jamieclarke]

Hi @Untold Entertainment Inc. -   I’ve moved your post to the Ideas forum. Currently, captions are working as intended the optimizer will try to hit the targets if possible, but it’s not guaranteed. If you’re looking for stricter control, Submachine plugin can help with that.  Here's a tutorial Submachine - Full Walkthrough

[Untold Entertainment Inc.]

Just to clarify: when you choose the "single line" setting, and it results in some captions appearing on double lines, that's the feature working as intended?

[jamieclarke]

Hi @Untold Entertainment Inc. - I read this as you wanted to have a single-word caption for each word spoken.

Can you post a screenshot of what you are seeing, if you select single line then I would expect captions to be a single line.

[Untold Entertainment Inc.]

Sure. Here's a screenshot of what i'm seeing: after telling the tool to limit captions to a single line, it went ahead and made some captions double-line anyway. That's a bug, no matter how you slice it.

i know you're motivated to reclassifty these reports as "Ideas" and software suggestions wherever possible because fixing bugs costs Adobe time and money, but as a customer, i am spending money on the software, and time to fix the software's errors. Forgive me if "go buy a $150 plugin to work around the bug in the $800/year software you've already purchased from us" isn't an enticing solution. 

This report is about two problems (per the post subject): 1. the captioning system puts multiple words on a line, despite the user selecting the minimum possible word frequency settings, and 2. the system displays captions in double lines, despite the user selecting "Single" under the "Lines" setting.  One issue feeds into the other. i think the system disregards the selected style's font size (another bug), and so when it puts multiple words on a line (which it shouldn't - that's a bug), and when the Style's font is large enough, the text box pops to two lines.  

In the screenshot, you can see that although i selectet the minimum possible word length, duration, and gap, the system has decided to put four words in one caption. And the Style's font size is large enough to pop the text box to multiple lines to fit everything. If the system didn't put multiple words on a line, the captions would easily fit the Sequence width, with room to spare.

Why don't you buy the Submachine plugin and integrate it into the software if it does the job correctly, if the Adobe dev team can't quite seem to make their solution function as advertised?

- Ryan

[jamieclarke]

Hi @Untold Entertainment Inc. -  It’s always best to report one issue per bug report — it helps us track, prioritize, and resolve problems more effectively. We’re absolutely here to help and want to make sure your issues get addressed properly.

As I mentioned earlier, I didn’t realize you were running into multiple issues, so I initially focused on just one. That was my oversight thanks for your patience.

Regarding captions: at the moment, the optimizer will do its best to meet your targets, but it’s not guaranteed in every scenario. From your screenshot, it looks like the caption style you’re using may be too large and is running into the edge limits for closed captions. I’d recommend decreasing the font size if you’d prefer the text to stay on a single line.

Submachine was suggested as a potential workaround for your specific use case one that Premiere Pro doesn’t currently support natively. We’ve been hearing a lot of interest in TikTok-style subtitles lately, which is why your post was routed to the Ideas section.

Thanks again for sharing your feedback and helping us improve!

[Untold Entertainment Inc.]

But can you see how that's not a viable reaction?  If the tool was able to limit captions to a single word, as it should be, the font size would not be an issue. (Also: hey, user... just GUESS at the font size that will work). The words "this" and "is" and "the" and "biggest" could all easily be displayed on the screen at that font size. The software knows (or should know) that these are all individual words, because they are delineated by spaces. 

When you say something like "the optimizer will do its best to meet your targets," you're creating this image of "The Optimizer" as an adorable, plucky little robot that's doing its darndest to try to keep up with the demading user.  i reject your premise.

In reality, there are human software engineers whose job it was to deliver this feature bug-free, and which wass sold to me as part of a very expensive software service from Adobe. For me to point out that it's not working as intended is not me being unreasonable, or asking for functionality above and beyond the software's capabilities. It's me, the customer, outlining my very reasonable expectations for software that functions currently as advertised. 

i'm a developer myself, so i know a bug when i see one! The existing captions feature can obviously do what i'm asking it to do, and what it purports to do. It has all the information it requires in order to split those captions into single words. It even thinks it can do single-word, single-line captions, judging by its exposed parameters (and therefore i, the user, think it should do that too). The fact that it ignores those spaces between words and puts multiple words on a line is a bug.  If it can't separate by spaces (even though that's very straightforward, programmatically), it can even look at the currently selected style's font size and project out how wide the caption will be, and if it determines it's too wide for the sequence, it can break that caption up, using — you guessed it — the spaces it should have heeded to begin with. But it doesn't do that. And that because the human software developers who wrote the code for it made a mistake. The feature is bugged. 

- Ryan

[Francis-Crossman]

Hello all.  I'm a Product Manager on the Premiere Pro team, and I just became aware of this thread. But this is not the first time I've heard of this issue.  We can tell you that it's not a bug and it's working as designed, but if it's confusing and doesn't meet your needs, then we got the design wrong and it doesn't really matter what we call it, bug or otherwise.  I think you have a valid point.  I was able to reproduce the issue myself – it's not hard.

I know these issues have been discussed here, but to reiterate, you can get situations that appear to disrespect the rules if the following criteria are met:

1.  Dialog is being spoken very quickly, you can cover several words in 1.2 seconds.  We will include all those words in the caption even though it may be way more than 7 characters.
2.  If the text style is very large, it will automatically wrap to the next line seeming to disrespect the single line choice. 

I spoke with some engineers and have proposed the following changes. Let me know what you think?

1. Reduce the minimum number of characters all the way to 1 (but always show at least one whole word even if it's longer than one character)
2. Reduce min duration to go to 0.25 seconds (but always show one full word for the duration it was spoken even if it takes longer than 0.25 seconds)

Remember that these can't be brick wall limits.  If they are, and the rules conflict with each other, what should we do?  Just not display the caption at all?  I think not.  So these rules MUST be flexible.  Let me know what you think of my proposed solution.  Do you think it will work?

On a side note.  The ultimate request here, I think, is to have robust animated captions that can do word-by-word animation so you don't have to mess around with plugins.  We have heard that request loud and clear.

1.  

[Untold Entertainment Inc.]

Thanks so much for your attention on this, Francis. 

Yes, i think reducing the time spoken will help. But the reason this behaviour is jumping out to me loud and clear as a bug is that even when my speaker rattles off a litany of words very quickly, the tool seems to be able to keep up. It seems to be only certain phrases that combine words (for example, the words "in the comments"). But maybe that's just my observer bias?  

SETTINGS PRIORITY

i'm still unsure why the engineers believe that reducing the tool to recognize a maximum of 1 character would fix the issue though? The current lowest setting for "maximum length in characters" is 7 characters. 

In my mysterious example, the words "in the comments," which Premiere Pro regularly combines into a single caption, are each delineated by a space. The tool recognized each word just fine when it wrote them out. But it chose to put all three words in the same, single caption, despite clearly being able to determine that those are three distinct words that are separated by a space. So if we're investigating the tool's rules, it would appear that the "minimum duration in seconds" setting may be overriding, or taking precedence, over the "maximum length in characters" setting. 

i'm just scrubbing through my most recent project now, looking at the captions. In one moment, my speaker says "I got one word for you." The tool has divided these words up like so: "I got / one / word / for / you," deciding to put the two words" I got" into a single caption. 

What is the tool "thinking"?  You might say that "I got" is too short for it to detect as two separate words, so it put both of those words into a single caption. But "I got" has a space in the middle, and the tool knows that. Further, the words "got," "one," "for," and "you" are all identically sized 3-letter words, and only "got" is paired up with a second word. Those words are all under the tool's supposed minimum 7 character "max" limit — there should be no functional difference between any of them, from a computer's perspective.

In another project, the tool had no trouble isolating the words "I" and "a," both single-letter words, into their own caption. So the bug (and i'm still calling it a bug, though i suspect that doing so changes your development priorities?) seems to be that despite the user choosing a maximum character limit, the tool ignores that setting, and really only cares about the time it thinks the speaker is taking to utter a phrase.

If that's the case, perhaps in the case of a slurred phrase where the tool hears "in the comments" but can't quite tell where each word begins and ends, instead of just giving up and putting all the words into a single caption, it can split the difference and carve up that caption, dividing that space evenly so that "in," "the," and "comments" are evenly distributed?  i would much prefer very slightly inaccurate timing to the tool just ignoring my settings and putting multiple words in one caption.

As for the font size being too large to display on a single line, here's where i think that the software can be reasonably expected to do more heavy lifting than it is currently doing. 

FONT SIZE AND LINE OVERFLOW

Let's say the engineering adjusts the feature. The user can now choose "maximum characters: 1" and "minumum duration in seconds: 0," along with "Single lines" (instead of "Double lines") But then the user, from the same prompt, chooses a caption style with an enormous font. Those first three settings are what likely matters most to that user: one word at a time, each word on a single line.  So if the caption is the word "AND," and the user has chosen a gigantic font size that forces the tool to put each letter on a different line:

A
N
D

... a computer knows this. It knows the font is too large to contain that word on a single line like the user requested. The computer knows the word will overflow, because the computer is the one drawing the caption!

So if we're talking about priority override of certain settings, my expectation here would be that Permiere Pro's got my back: it will resize that caption for me, reducing the font size to the maximum size possible so that the word "AND" still fits on a single line, exactly like i requested (despite me being kind of a dumb user and choosing a font that's too large). This is not the tool's rules conflicting: this is the tool providing what the user has asked for from the tool's own settings prompt.

Here's how that plays out more practically, with a personal example: i like to use a big, chunky font for my single-word captions on my TikTok videos about board games. As long as everything is broken up into single words (which it's not, due to the bug), those captions all fit on a single line 99% of the time.  But very occasionally, the caption will say "BOARDGAMEGEEK.COM."  That single, 17-charcter long word is too wide for my settings. Because of that, i have to scrub carefully through the timeline and find all of the places where my speaker has spoken that url, and then adjust the font size down so that it fits on a single line. 

But i don't use Adobe software to make my life more difficult. i want things to be as easy and seamless aas possible. The computer knows ahead of time, while creating the captions, that certain captions have overflow issues. i would prefer the tool to correct those issues FOR me — even if it slows down the caption creation process. Because as slow as it is, it can't possibly be slower than me picking through the timeline and correcting each and every caption by hand. And if you're worried that some users won't want the tool to resize their caption font sizes like that, you simply put the power to decide in the hands of your users, by exposing a checkmark that says "Automatically resize overflowing captions." 

Otherwise, the net result is that i choose "Single line," and Premier Pro says "Forget you — i'm making it double lines." The software directly contradicts what the user is trying to do. And like i said before, if you expose a button that says "green," and the user clicks it and the thing turns red instead, we're not in "feature request" territory. That's a bug. 🙂

- Ryan

[Stan Jones]

@Francis-Crossman,

Thanks for giving this topic attention. The new caption workflow, speech to text, and text-based editing have moved ahead steadily since their introduction. Fabulous, work-saving features. And there's lots more to do.

I agree with the proposals as good next steps.

@Untold Entertainment Inc., I'll respond to your comments later; I need to do some tests first.

@Francis-Crossman, issues involved in Ryan's comments are limitations to the caption text boxes, wrapping, etc.

The proposed changes are more flexible, but I'll comment on an additional option of a "Single Word" setting for captions. It has value in being a more intuitive process than "if you want single word captions, use max characters 1, min duration 0.25, and single line." It differs when the single word length would be less than the minimum duration. I personally don't have a problem with "I do" versus "I" and "do," if they are spoken so quickly it won't fit the quarter of a second. But there are some creators doing some crazy stuff!

Also, plug-ins must currently use caption creation as part of their animated workflows. There is no scripting support to access the transcription itself, only captions. More scripting access may be on the roadmap, but it will not happen soon. As long as a caption is at least one FRAME in length, it may have value.

> The ultimate request here, I think, is to have robust animated captions that can do word-by-word animation so you don't have to mess around with plugins. We have heard that request loud and clear.

Excellent!

Stan

[alex_3717]

Hello @Untold Entertainment Inc,

I completely get your frustration, caption generation should be predictable, especially when you’ve precisely set all the preferences. From what I’ve seen (both in video editing workflows and even lightweight text rendering in mobile gaming apps like endless runners), the root issue often isn't the UI settings, but how Premiere internally clusters spoken phrases during transcription.

Even with the sliders at minimum and "Single" line enforced, Premiere's speech-to-text engine still does some auto-grouping based on timing and phonetic flow, not just character count. It tries to avoid captions feeling too choppy for viewers, but that definitely clashes with setups where you want hyper-granular control (like word-by-word breakdowns). I run into similar headaches when trying to auto-generate subtitles for in-game cutscenes, sometimes the "AI" decides two short words sound better grouped, even if technically they shouldn't.

There’s a workaround (though tedious): manually adjusting the segmentation during the transcript phase before creating captions gives you finer control. Still not ideal when you’re aiming for a true "set and forget" workflow, but it does help minimize post-generation cleanup.

Hoping Adobe improves this with a future update, because you’re right, it feels like a basic expectation when you dial in the settings so precisely.
- Alex

[Untold Entertainment Inc.]

Hi @alex_3717 . Yes, it's that "auto-grouping" you mention that's causing all the problems. Your workaround, unfortunately, isn't really a workaround: whether i'm fine-tuning the transcript word-by-word, or fine-tuning the captions word-by-word on the other end, the result is the same: i'm spending a bunch of time fussing over it, when robots are perfectly capable of doing it FOR me.