Animated captions and lower thirds in Premiere Pro
Two different things go by the name "captions" in an edit. One is subtitles: what a person said, transcribed and put on screen in time with their voice. The other is the designed plate that carries a name, a chapter number, a location, a handle. Both get typed into the same search box. Only one of them is what this plugin makes.
Butter Motion does not transcribe anything
There is no speech recognition in it. It never opens your audio, it does not know what anyone said, and it cannot sync a word to a waveform. Word-by-word karaoke subtitles are outside what it does, and that is a design decision rather than a gap on a roadmap.
If subtitles are the job, Premiere already does it. The Text panel transcribes a sequence and turns that transcript into captions, no plugin needed, and several plugins built specifically around transcription will style those captions much further than the built-in styles go. Any of them will serve you better than we will. Go use them.
What is left after you take subtitles off the table is still a large pile of work: every graphic in a cut that carries text and is not a transcript.
The plates this actually covers
The panel groups these under one genre called lower thirds and sections, and in a real edit they turn out to be the same design problem wearing different hats:
- Name plates. Two lines, a person and their job, under the left third of the frame.
- Chapter cards. "PART TWO", a rule that draws itself, a subtitle under it.
- Location and date stamps. "Reykjavik, March 2026" arriving in the corner and leaving quietly.
- Handles and bugs. A channel name or
@handlethat sits in the corner for the length of a segment. - Section dividers. The full-frame card that separates two blocks of a long video.
- Callout labels. A short line of text that points at something in the shot.
None of those come from the audio. All of them come from a decision you already made in the edit, which is why typing them is not the slow part. Building them is.
Writing one
The pattern that works is the one in Writing prompts: name the object, name the motion, name the mood, and put any on-screen text inside quotes so it comes back exactly as you wrote it.
Lower third with "Anna Petrova" on the first line and "Head of Production" on the second. A thin rule under the name draws left to right. Slides in from the left, holds, slides out to the left. Dark, calm, one accent colour.
A chapter card is the same shape of sentence with a different object:
Full-frame chapter card. Large "PART TWO" in the centre, small "The return" underneath. Letters rise into place one after another, a hairline underlines the number. Dark background, warm accent.
And a corner stamp is shorter still:
Small location stamp in the bottom-left corner: "Reykjavik" on top, "March 2026" below in lighter type. Fades up, holds, fades out. Minimal, no box.
Three sentences is usually the ceiling. Longer prompts do not buy accuracy, they buy noise. What buys accuracy is the quoted text and one clear verb for the motion.
If the plate is already designed
Sometimes the design is not yours to invent. The client sent a brand book, or last season's plate exists as a still and the job is to make that still move.
Attach it. In the composer press the + button, paste an image with Ctrl/⌘ + V, or drag the file straight onto the field, and the AI animates what is in the picture. Leave the prompt empty and it picks the motion itself. Add a line to steer it: "bring the name in from the left and draw the rule after it". Images go up to 25 MB. Writing prompts covers the flow.
Typefaces deserve one more sentence. The animation is assembled on your own machine, so the font control offers the faces installed on it in a separate "Your fonts" group, next to the built-in set. Pick the client's licensed brand face there and it goes into the plate with no conversion step and no font file uploaded anywhere.
Three settings that decide whether the clip is usable
The prompt describes the design. The chips under the prompt field decide whether the file will drop onto your picture at all, and two of them are unforgiving.
Background: transparent. This is the one that matters most. A transparent generation comes back as a MOV in ProRes 4444 with a real alpha channel, which is what Premiere needs to key a plate over a shot. A generation with a background comes back as an MP4 with that background baked in, and it will cover your footage. Transparency is chosen before you generate, not after. Background and transparency has the exceptions.
Aspect ratio and frame rate. Match the sequence. A 9:16 plate for a Reels cut and a 16:9 plate for the same film are two generations, not one clip you can squeeze. Frame rate matching means the clip lands without conversion.
Length. A name plate that a viewer can read runs about five seconds. The range is 3 to 20. Long is trimmable, short is not fixable, so err long.
One sentence, then the whole project
This is the part that is worth the plugin, and it has nothing to do with any single clip.
Get one plate right. Then save it as a preset, and for every other speaker in the film you load that preset and change only the two lines of text in the customization controls. Same typeface, same accent, same entrance, same timing, different name. The controls are free for subscribers and they never spend credits, so a nine-person documentary costs one generation and eight bits of typing.
The same trick carries a series. If episode one has a chapter card you like, episode four does not need a new design, it needs the preset and a different number. Over a season that is the difference between a house style and a pile of slightly different cards nobody meant to make.
One honest caveat: aspect ratio is chosen at generation time, so a vertical version of the same set means new generations rather than a switch on a finished clip. The design does not have to be rebuilt, the preset carries over and only the aspect chip changes.
Name the preset after the project, not the person. "Doc interview plate" survives the edit. "Anna" does not.
If a plate is close but not right, the first tweak on every generation is free: type "remove the box behind the text" or "slow the entrance down" and it edits what you have rather than rebuilding it. Details in Editing a generation.
Getting it onto the picture
A transparent clip needs an empty video track above your footage. Interview on V1, plate on V2, and V2 has to exist and be free at the playhead before you press Insert to timeline. With no free track above, the overlay has nowhere to land, and the most common report of "the lower third vanished" is exactly this. Rendering to the timeline covers where the file goes and how it gets imported.
The render itself happens on your machine, so nothing about your sequence is uploaded to produce a plate. Only the text of your prompt leaves the building. That is covered in Why the render happens on your own machine.
What a set of plates costs
Generation cost depends only on the quality tier: 7 credits on Flash, 10 on Balance, 15 on Ultra. Nothing else moves the price, so a 4K plate and a 720p plate cost the same. Plates are simple geometry and text, which is exactly what Flash is for, and most of them do not need Ultra at all.
Everything after the generation is free: the first tweak, the customization controls, and re-rendering as many times as you like in as many formats as you like. On a Pro plan (500 credits a month) a full set of interview plates for a documentary is a rounding error against the month. Plans are in Pro and Studio, credit rules in How credits work.
Where the line is
If you need a transcript on screen, this is the wrong tool and no setting will change that. Premiere's own captions, or a plugin built around speech, is the honest answer.
If you need the twelve text graphics that are not the transcript, and you need them to look like they came from the same designer, that is the job this was built for. The full workflow for a single plate, start to finish, is in A lower third for an interview.