Automating High-Volume Reel Creation with Custom On-Screen Text (50–70 Videos/Day)

Hello,

I currently work in OFM and I’m looking to industrialize the creation of reels with large on-screen text (hooks), at a volume of around 50 to 70 videos per day. At the moment, everything is done manually and, even though the strategy works, the time spent editing has become a major bottleneck. The goal is to set up an automation where I provide, on one side, a folder containing the source videos (short clips of 6–8 seconds), and on the other side a Google Sheet or CSV containing all the texts to be displayed, often split into one or two parts with specific timing. The automation would then automatically match each video with its text and output the final videos in bulk, without having to open and edit each video individually.

When I look for resources online, I mostly find tutorials about AI-based automatic subtitles, which do not fit my use case at all, since this is not about transcribing audio but about overlaying deliberately written, structured, and controlled marketing text. I’m therefore looking for technical leads, feedback, or tutorials about solutions that are actually suited to this kind of workflow (After Effects with external data, CapCut, scripting, automation tools like n8n, or similar) and that allow producing this volume of videos reliably and quickly.

Thanks.

I don't have a ready‑made solution, but I have some ideas that can help you build one:

The simplest way is to use an AI like ChatGPT (or any other you like) to create a script that uses ffmpeg.

I've done similar things before. On Linux I used a Bash script; on Windows—or any OS—you can use a Python script.

The big unknown for me is how the script will read the CSV/Google Sheet. In my case each video to be processed was in a subfolder that contained the text file(s) to insert into the video, but I'm sure it's possible.

I suggest you open a first chat with the LLM of your choice and tell it to play the role of a prompt engineer, creating the most accurate prompt possible for a second chat. That second chat will act as an expert, guiding you through the script creation.

Explain your situation in the first chat. You can simply say, “I asked this question on a forum [your starter thread]” and paste this response: “[my message you’re reading].”

The prompt engineer will then assign the correct role and instructions so the AI can play the appropriate expert in the second chat.

End both prompts by telling them to ask clarifying questions before starting, including your level of knowledge and technical expertise. That way they’ll adapt and teach you things you might not know but need.

I'm not a developer, but I’ve been able to create automations like this since 2023, using older versions of ChatGPT (3 and 3.5).

What are you using for editing currently?

Also, just so that I understand the objective; right now you have a folder of short video clips that you want captions on, but the captions need to come from the (sales copy) script you have instead of a transcript of the video and you're also trying to automate it, correct?

Primarily edit in Premiere, so that's where I've got the deepest knowledge. Normally to do subtitles you'd have it run the transcription and build the subtitles from there. However, you can also upload your own files at that step instead. Haven't done that, so not exactly sure how it would function in this case, but you should be able to go in and adjust your timing at that point. Don't have much to tell you on automation, but if you can standardize things it should be doable with that process, you just might have to set up your file management and documents in a particular way.

Would love to hear about what you figure out and what solutions you find!

Did you figure this?

Yeah this is usually done with AE + CSV/JSON data-driven templates or a ffmpeg/script pipeline, anything manual like CapCut won’t scale reliably to 50–70/day.

It can be done easily with ffmpeg and some custom python/ruby script. Come on its 2026, script like this could be made by AI in one prompt or PM me if you still need help.

ffmpeg is fast but styling text to look like those high-converting OFM reels is a massive pain in the ass with raw drawtext filters. if you want them to actually look decent and not cheap, aerender is the way. you make one solid template in after effects with your fonts, animations and responsive text boxes, then use a python script to swap the text from your csv and trigger the aerender CLI in the background. another solid option if you know some javascript is remotion... it was basically built for this exact programmatic video stuff and handles csv inputs easily.

Yeah @SMM Junction is right about ffmpeg drawtext looking like garbage. i tried that route last year for some tiktok campaigns and the text styling was just too painful to look at, especially for OFM where you need that clean, high-converting aesthetic.

if you want to avoid the learning curve of remotion or after effects scripting, python with moviepy is probably your best middle ground. it uses ffmpeg under the hood but handles text wrapping, custom fonts, and drop shadows way easier.

had a similar bottleneck a few months back and wrote a quick python script that reads a csv, grabs the random 8-second background clip, overlays the styled hooks, and renders them. takes maybe 10 seconds per video to spit out. i can share the basic structure if you go that route, but honestly just feed the moviepy docs to claude and it'll write the script for you in five minutes.

Honestly remotion is the move if you already touch any js, the csv handling is dead simple and you get proper react components so your hook styling stays consistent across all 70. moviepy works too but it gets slow once you stack drop shadows + custom fonts and you're doing volume daily.

One thing nobody really touched... you said the text is split into one or two parts with specific timing. That's where most of these scripts fall apart, not the rendering itself. Whatever route you pick make sure your csv has explicit columns for part1_text, part1_start, part1_end etc instead of trying to be clever and auto split. Keep the data dumb and the script stays simple.

@SignalMint moviepy doing 10s per clip sounds right for simple overlays but if he wants the animated/keyframed hooks that aerender route SMM mentioned is gonna look cleaner, just slower to render. kinda depends if you need movement on the text or static is fine.

Also for the matching part, easiest is just name your clips 001.mp4, 002.mp4 and match row index. don't overthink the folder logic.