Inspiration

How We Turned a Pile of Blog Posts into a Flood of Shorts: Automation for Speed, Humans for Quality!

This is a story from one of our VX service clients—a hotel based in Yeoju that runs a blog to promote their property.

They had already published over 200 blog posts, and wanted to turn all that content into videos. That’s when they came to our VX managed service.

Once you pass 200 posts, manually creating videos one by one is basically impossible. If a human had to do everything by hand, the production cost would skyrocket. 💸

So here’s what we do: we automate all the repetitive work using Claude, code, and the Videostew API. Then, in the final stage, a human steps in to polish the details and level up the quality.

This post is a recap of how far we’ve automated the entire pipeline—and what’s still better done by humans.

If we had to sum up the whole flow in one line, it would be this: we read the original blog post, turn it into a Shorts-style script, classify which images or video clips to use, and render a first draft of the video. After that, a human editor reviews it, swaps in more fitting visuals, fixes awkward lines in the script, and re-renders the final version.

The magic of our VX managed service comes from combining these two pieces: automation + human touch.

First, what the automation handles

The very first thing we did was create a Shorts template to use for automation. We set up a layout in the Videostew editor that defines where the title goes, where captions appear, where photos are placed, and the final branded outro.

Since the visual quality of every video ultimately depends on this template, this part is handled by a human designer. (Yes, this can be generated code-first and automated too—but when it comes to placing and animating elements on an actual canvas, doing it directly in the editor is usually faster and much more comfortable.)

Once the templates are ready, the heavy lifting for mass video creation is handed over to Claude Code. You chat with Claude, tell it what bulk work needs to be done, and here’s what happens step by step when you line up everything Claude does.

From one blog post to one finished video. Repetition is automated, but quality is handled by humans.

1) Reading the article and turning it into a script

No one is going to reread over 200 posts by hand. So Claude Code reads each article, figures out the angle on its own (business trip stay, one-month stay, near-hospital stay, etc.), and organizes it into a YouTube Shorts script. Of course, when we script it, we run it through our in-house agent so it’s shaped perfectly for Shorts, using all of our know-how.

Oh, and come to think of it, humans step in at this stage too. Sometimes Claude Code breaks the rules a little and writes the script its own way, so we compare it with the original text and review whether the unique characteristics of the original article are really coming through.

2) Sorting photos and video assets

Next comes organizing the visual assets to be used. The client’s resource folder was a glorious chaos of hundreds of mixed images: rooms, kitchens, appliances, parks, cafés—and on top of that, video files too.

Claude Code sorted everything into categories, filtered out anything that looked obviously AI-generated, low quality, or duplicated, then picked the representative shots and even cleaned up the file names. Hundreds of files were trimmed down to a neat set of 117. Since we’d be swapping in even better-fitting images later anyway, this step was all about token optimization: selecting only the most reusable visuals that we could comfortably use across many videos.

3) Cloud upload and subtitle matching

To fully automate video creation, we need publicly accessible image URLs. So once the images are selected, Claude’s code uploads them to Google Cloud Storage (GCS) and generates public links. Temporary free hosting might look tempting, but one random broken link can wipe out your entire video source. That’s why we use proper, reliable storage. (For the record, Videostew actually runs on S3, but since our Claude setup was already wired into Google Cloud Console, it naturally suggested GCS—so we just rolled with it.)

Next, we send the script and images into the Videostew Data API to assemble the video.

For each subtitle line, a suitable image is automatically matched by category, motion-heavy scenes get video clips instead of stills, and finally, a template-based outro is batch-applied to every video.

One thing to note: if the element labels defined in the template don’t match the field names you send through the API, your images simply won’t appear. Normally, this is the kind of small mismatch that can waste hours— but Claude runs its own test runs, so once we get a single successful pass, we can mass-produce videos at scale with virtually zero glitches. No more “why is this slide blank?” surprises. 😅

4) Video generation and validation

Once everything was wired up, the actual generation was unbelievably fast. (We went out for lunch and came back to find 57 videos finished…)

In this batch, we created 57 videos in one go. We pre-calculated how many slides each video should contain, then automatically cross-checked that against the final output as a form of validation. All 57 passed the check perfectly, and their durations landed neatly between 31 and 36 seconds—ideal for Shorts-style content. (By the way, here’s a bit of our internal playbook: if the content feels informational, we try to stretch it as close to 60 seconds as possible; if it feels promotional, we keep it under 40 seconds. That’s one of our little performance secrets.)

Now it’s time for humans to step in!

What you see here is the fully automated output. The format is nicely standardized and batch-produced. (Blurred for client privacy)

First, we carefully watch the rendered videos from the beginning. As we go through them, we spot patterns: a certain word that keeps sounding off, or the wrong image popping up again and again.

That’s when we make a revision plan. We decide which extra assets to bring in, how to tune the custom dictionary, and sketch out the big picture so we can minimize repetitive work.

What humans fix at the very end

More concretely, here’s what we do.

First, matching images and video clips. Claude Code does auto-attach photos to subtitles, but it’s still not quite on par with a human’s eye for selection.

So our editors manually upload key scene sources to the library, then pick the most fitting images and clips for each subtitle, one by one. This is what shapes the “first impression” of the video, and this is where quality can jump dramatically.

Second, polishing the script. Auto-split subtitles sometimes break lines in awkward places or sound just a bit unnatural. A human editor watches each piece, fixes clumsy line breaks, and rewrites stiff sentences so they sound natural when spoken out loud.

Third, correcting AI voice pronunciation. The narration is done with AI voice, but some specific words get mispronounced or sound slightly off. We add those words to a custom pronunciation dictionary and tweak them until they sound natural to the ear.

If you handle these three steps in-house, automation can cover you “well enough” to a point. But because our clients outsource this to us, we go the extra mile: this is the stage where humans add the final, crucial touch. That last bit of hand-crafted detail is what separates a video that “looks like it was cranked out by AI” from one that feels like a human really cared.

Ultimately, the most important factor is the quality of the original text. If the source content has real value, that value carries through even when you scale production. You can mass-produce videos, and as long as the original writing is strong, its meaning and impact won’t get diluted in the video format.

You absolutely can do it yourself. It’s just a lot easier if you let us handle it.

Technically, you can run this whole process in-house if you’re determined enough. With well-defined automation rules, your team can definitely produce meaningful content on its own.

But there’s one important condition: the original text itself has to be worth turning into a video. If the content is empty, no amount of clever automation will magically turn it into a great video…

And if you care about the quality of every single video, humans eventually have to step in. Having your internal team obsess over each piece, every week, can quickly become a burden. It really depends on the nature of your organization, but everyone already has a “real job” to do, right?

You want to upgrade your content quality, but hiring or assigning someone just for that feels overkill? That’s exactly when you can hand it off to us.

In fact, most of our clients simply send us a few article links each week—or just one PDF of their magazine—and they’re done.

We might go back and forth a bit in the beginning to align on content direction, but once that’s set, it becomes almost mechanical. For example, you send materials every Monday, we produce and deliver the finished videos… and that’s basically the extent of the communication.

FAQ

Q. Is everything created fully automatically?

If you want full automation, you can just use the Videostew API directly. When you work with us, repetitive tasks are automated, but the final polish is done by humans—and that’s the real advantage. The things that truly decide quality—image matching, copy tweaks, AI voice pronunciation fixes—are all reviewed and finalized one video at a time by a dedicated specialist.

Q. Does it get more expensive as I make more videos?

We work with a fixed per-minute rate (starting from 7,000 KRW per minute). That means you can plan your weekly video publishing and set your budget accordingly—no unpleasant surprises later.

Q. Can you match our brand tone?

Absolutely. Since design quality is driven by templates, our client and a Videostew design expert work together at the start to pick templates that fit your brand. We create these templates with your CI and BI in mind, and for long-term partnerships we continuously refine them over time based on performance data.

Turning a pile of text into videos no longer has to be a slow, manual grind. Repetition can be automated at high speed, while humans focus on polishing the final quality.

If you’re sitting on tons of blog posts, documents, or internal materials that really should be videos by now, our VX managed service can take over the entire process for you—from start to publish. Reach out anytime, and let’s make those videos work for you! 🚀