Caption workflows for agencies
Lock the cut, preview the style on this clip, batch 1× vs 2×, then send the transcript for review—not the pixels.
An agency caption job fails in the handoff, not in the model. The cut is still moving, the style is picked from a thumbnail, the whole batch is burned at 2× because Glass is the default, and the client marks up pixels that can only be fixed by rendering again. The workflow that holds is four gates in order: lock the cut, preview the style on this file, batch the 1× / 2× decision on purpose, then review the transcript as text.
This is not a brand-kit essay. Kit choices — type family, colour, a house look — can live in a deck. The production problem is sequence. Burned-in captions are pixels. Every step you skip becomes a second bill and a second round of notes.
Lock the cut before anyone captions
Captions follow the edit you actually ship. If the picture still has an extra beat, an intro bumper you might drop, or a CTA the brand has not signed, do not transcribe it. A later trim does not ripple a burned file. It orphans every cue after the cut and forces a recaption of the whole clip.
The lock is a named file, not a feeling. Version the picture: client_brand_hook-v3_locked.mp4. That is the only input the caption step is allowed to see. Editors who caption v2 while the suite is still arguing about v3 will caption v3 again. The time you save by "getting a head start" is the time you spend reconciling two transcripts.
Audio is part of the lock. A VO swap, a music dip, a line rerecorded in pickup — any of those change the speech the captioner hears. Lock picture and the track you will publish. If legal is still rewriting the claim, wait. The captioner will typeset whatever is on the tape, including the sentence you are about to kill.
For clip-from-long-form jobs, trim first. Captioning the source interview and then cutting highlights is how you pay to typeset twenty minutes of throat-clearing. The clip is the deliverable. Add captions to a video should hear the clip.
Write the lock into the SOW as a gate, not a courtesy. "Captioning starts after picture lock" is a sentence a producer can enforce. "We'll caption as we go" is how the invoice grows a recaption line nobody budgeted.
Preview the style on this clip, not a sample
A picker thumbnail is a still of a font on a grey card. The client does not watch grey cards. They watch this talking head, this kitchen, this screen recording, on a phone, often muted.
Preview means a few seconds of the locked file in the candidate preset. On Versely that is preview_caption_style against the real clip, not a studio demo. Contrast against this background. Check whether the animation fights the face. Check whether the first line sits in the safe area the platform chrome will cover. Then pick.
Two previews are enough to decide a look. Ten full renders are a shopping trip billed as production. Caption styles is the catalogue; the preview is the fitting room. Do not send the client a mosaic of ten finished videos and ask which vibe they prefer. They will pick two, ask for a mix, and you will recaption.
Name the preset in the job ticket before the batch starts. "Corpo, bottom, 1×" is a spec. "Make it pop" is not. If the brand already has a house caption look, reuse it. The agency's job is consistency across a campaign, not a new personality per cut. Changing preset mid-flight is a new treatment and a new comparison; it is not a free restyle.
Animated families (the Dynamic tier: Glass, Whisper, Glide, Fusion, Terminal, Handwritten, Backdrop, and the rest of the nine) are a design choice with a cost multiplier. Preview them when the caption is part of the performance — a listicle, a karaoke hook, a line you want the eye to chase. Preview a Basic family when the caption only has to be readable. The preview is where you make that call, once, on one clip, then apply it to the batch.
Batch 1× vs 2× on purpose
Versely bills captioning as a render multiplied by the preset tier: 21 Basic looks at 1×, 9 Dynamic looks at 2×. Transcription work is the same. The multiplier is the look. What captions cost across a month of clips is that arithmetic at cadence. An agency that leaves the agent default on Glass will 2× a whole retainer by accident.
Decide the tier at the campaign level, not per editor mood.
- 1× default. Talking-head ads, testimonials, screen recordings, anything whose job is "read the words." Pin a Basic preset. Quote the batch on that multiplier.
- 2× by exception. The hero hook, the listicle, the line that is the edit. Mark those files. Do not silently upgrade the other eighteen.
Batch the files that share a preset and a language. One language code per render: the captioner transcribes what is spoken; it does not translate. A Spanish VO with en-US left on the ticket is a bad English transcript, not a bilingual deliverable. A dubbed cut is a new caption job after the dub.
Do not mix overlay work into the caption batch. A hook line nobody said is a text overlay, authored copy, not a transcript. Running the captioner and hoping the headline appears is how you get a phonetic mess of the VO and no headline. Keep the two jobs on two lines of the ticket.
Quote recaptions. Picture lock slips. Clients change a product name. Both are real. Both are a new render. If the SOW says "captions included" with no recaption cap, you will eat the second burn. One included recaption after transcript approval is a sane default. Open-ended pixel notes are not.
Client review of the transcript, not the pixels
Send words as text. That is the whole review design.
Pull a transcript from the locked audio — get a video transcript is the plain-text job — and put it in a document the client can comment. Mark the strings that have to be right: product names, prices, the claim the hook depends on, the CTA URL if it is spoken. Ask for a yes on those strings. Do not ask "how do the captions feel."
Why text first: a wrong word on a muted feed is the whole message, and a wrong word is cheap to fix in a file. A wrong word discovered after burn-in is a recaption. Legal and brand teams can read a transcript on a train. They cannot usefully redline a 9:16 MP4 with a finger.
After approval, burn. Then send the MP4 as a confirmation pass, not as a writing pass. The confirmation questions are placement and timing: is the block off the chrome, does the first cue land with the mouth, is a cue flashing. If they start rewriting sentences on the MP4, stop and take it back to the document. You are otherwise captioning a moving target.
Name a single reviewer. Three stakeholders on a pixel file will give you three fonts and a recaption each. The transcript round is where marketing, legal, and the producer collide; the MP4 round is where one producer signs placement.
The agent path for the burn, once the text is signed, is transcribe and caption my video: speech in, styled timed captions out, 165 language codes, preview before the full render. Use it on the locked cut with the preset you already previewed. Do not use it as a place to discover the script.
Agencies that run this sequence caption once. Agencies that skip a gate caption the same clip until the retainer is the recaption line.
Caption the locked batch in Versely.
FAQ
Should the client pick the caption preset?
They can approve one look from two previews on the real clip. Do not send a catalogue. Pin the winner for the campaign so the next twelve files match. A new preset per cut is a new brand.
When is a recaption billable?
When picture or audio changes after lock, or when the approved transcript is rewritten. Placement tweaks you missed on the confirmation pass are on you. Write that split in the SOW before the first file.
Can we caption while we wait for legal?
You can transcribe to text and send that. Do not burn. A burned file of unapproved claims is a deliverable you will have to destroy and a credit you will not get back.