Three Months with WorkBuddy: I Put It Ahead of Final Cut Pro
Community Discussion · General

Three Months with WorkBuddy: I Put It Ahead of Final Cut Pro

Momo-chanMomo-chan2d ago2026/11/16 141 views

Last night at 11:20, I was still cutting a four-minute talking-head video. Seventeen screen-recording clips, five shots re-filmed on my phone, one voiceover track, all spread out on the timeline like a ball of yarn a cat just got at.

Halfway through cutting I scrolled past that Apple article on Final Cut Pro 11. Magnetic Mask, Transcribe to Captions, said to be updated across Mac, iPad, and iPhone, and you can edit spatial video too. The comments were all "finally."

I stared at the words "Transcribe to Captions" for two seconds, then switched back to my own window. Because the captions for this episode had already been sitting on my timeline three hours earlier, not a single character transcribed by Final Cut.

It's not that I don't use it. It's that I can't use it.

Transcribe to Captions solves the "voice to text to timeline" alignment problem. It recognizes what you said, splits it into sentences, and pins them to time points by waveform. Sounds convenient, but it has a precondition: what you say has to itself be usable copy.

In my case, one talking-head script is 2800 characters, and from topic selection to final draft it gets revised three times. First draft comes out at 3600 characters, too wordy. Second pass cuts to 3000, still has verbal tics. Third pass I read it aloud myself, delete all the "that" "I mean" "actually," and what's left is what can be recorded.

This work can't be done in Final Cut, can't be done in any editing software. It's text work.

So my workflow is reversed. First let WorkBuddy clean up the script, then go into the editing software. Over three-plus months, its place in my setup has gotten clearer and clearer: it handles text, not the timeline.

Let me get specific about how I use it.

Usually before recording, I toss the finalized talking-head script to WorkBuddy and have it split into a shot list by semantics, output a table: number, original script text, suggested visuals, estimated duration, caption text. This version it runs fast — from my testing, 2800 characters takes about twenty-some minutes to produce, thirty-some rows.

But the first time it ran I kind of wanted to close it. It split too evenly, almost mechanically by character count, each segment a hundred-plus characters. Talking-head doesn't talk like that. "I used this feature for two weeks" and "turns out it shifted all the dates in my table by a day" — similar character counts, but the former can be glossed over in one line, the latter needs a half-second pause, a close-up, let people react.

Later I learned. I manually add pause markers in the script, using slashes — wherever I feel I should breathe after a sentence, I draw a line. Toss it back, and the rhythm comes out 80% right. The remaining 20% I adjust myself, mainly merging two consecutive lines or splitting one.

The net time saved on this step, I've calculated. Doing the shot list manually, one episode takes about an hour and a half to two hours, with repeated re-reading of the script in between. Now twenty-some minutes for a first draft, forty minutes to fix, an hour total. Not much time saved, but it saves the most annoying, pure-transport time.

It also has a use after editing.

The srt exported from Final Cut I toss back to WorkBuddy and have it do two things: first, re-break the punctuation according to WeChat public account reading habits — editing software auto-generated sentence breaks often drag one sentence super long, which is a disaster on mobile; second, unify proper nouns — my product names often go all-caps one moment and lowercase the next, looks messy in captions.

I think this use is even more worth it than the previous one. Because caption work is most painful to fix in the editing software. You have to find that sentence on the timeline, double-click, change two characters, enter, then find the next. Twenty errors can eat forty minutes, and by the tenth fix you're numb and can't be bothered with the rest.

In WorkBuddy it's just a text table, go top to bottom, scan it in five minutes.

But its problems are always there too.

The most typical is table misalignment. I had it merge the shot list and caption table by number — row count came out right, but the columns shifted wholesale, the suggested visuals column ended up under caption text. At first I thought I'd messed up the column order on export, tried three times before confirming — it seems to hard-stitch by column position, not match by column name. I wrote a whole rant about this last week, still haven't found a particularly good workaround, can only scan the header after each run.

Another is it doesn't really understand duration. I have it estimate roughly how many seconds each talking-head segment is, and the numbers it gives are character count divided by something — actual recording often differs by a chunk. Chinese talking-head speed isn't constant; fast parts six characters a second, slow parts three and a half. It can't calculate this accurately, so now I just don't look at that column, I read it through myself with a stopwatch.

So I won't say it solved editing. It didn't. What it solves is the most tedious text organizing before and after editing; the part in the middle that really needs judgment is still me.

Back to Final Cut Pro 11. Both AI features I think are good things — Magnetic Mask should be very practical for people shooting products, no more drawing a mask ring to cut out the subject. Transcribe to Captions really lowers the barrier, especially for people shooting vlogs with no script, relying on improv — transcribe first then fix is way better than typing sentence by sentence.

But for me it's a bit different. I have a script, and the script is finalized before recording. Transcribing at that point is re-recognizing text that's already determined, with error correction in between — an extra step instead. And the recognized sentence breaks follow speech pauses, not reading rhythm, so I have to fix it again.

From my testing, for scripted talking-head, using the finalized text directly as captions is faster than transcribe-then-proofread. For unscripted improv content, transcription is a must. These two don't conflict, just apply to different people.

So my judgment is, tools like WorkBuddy and the AI in editing software each handle a segment, in a before-and-after process relationship. The stronger the editing software's AI, the more important the text quality of the earlier segment. No matter how accurately you transcribe, if the script itself is wordy, repetitive, logically jumpy, the transcription has the same problems, just more neatly wrong.

Conversely, if the earlier script and shot list are clean, you actually don't use those AI features in the editing software much. That's my current state — I turned on Final Cut's transcription, but basically don't use it.

It handles text, not the timeline. This sentence took me three-plus months to figure out, and once I did, the position I assigned it got fixed: one pass before editing, one pass after editing, don't touch the middle.

Who's it for? People with regular weekly output, a talking-head script, content leaning toward talking-head or explanation — this workflow saves real time. Who's it not for? People shooting vlogs, relying on live improv, whose material is itself voice — you need accurate transcription more, the shot list matters less, so WorkBuddy's value to you is limited.

As for me, once I finish cutting tonight's episode, tomorrow I have to record another. The script is already split into thirty-two rows in WorkBuddy; I haven't looked at the visual prompt column yet, probably need to fix seven or eight spots.

2 replies

?
Ctrl + Enter to reply
Weiwei's Shop

That bit about splitting the storyboard table evenly by character count is so real. After I added slash pause markers it only got about 80% right, and I still have to manually handle the remaining 20% of merges and splits.

Old Deng
Old Deng2d ago
Reply to Weiwei's Shop

Don't force-tune those 20%. OP's 2800-word script cut into thirty-some lines—80% alignment is already good enough. Merging and splitting the rest is worse than going back to the script and adding slashes.