Hi DEV community! 👋
I recently transitioned from being a video director and editor to an indie developer. During my years in video production, one of the most frustrating and tedious tasks was repurposing archival footage or localizing videos.
Whenever we needed to reuse a clip, we had to manually mask and remove hardcoded subtitles, TV broadcast logos, and old watermarks frame by frame. It was a nightmare.
To scratch my own itch, I decided to build TextFromVideoRemover.
It’s an AI-powered web tool that lets users manually select the exact area of the text or watermark, and it uses AI inpainting to reconstruct the background smoothly.
🛠️ The Tech Stack
Coming from a non-traditional CS background, modern web tools have been a lifesaver. Here is what I used to build it:
- Frontend: Next.js, React, Tailwind CSS for a clean, responsive UI.
- AI Processing: Integrated with backend inpainting models optimized for temporal consistency across video frames.
- AI Coding Assistant: I've been heavily using CodeX for my full-stack vibe coding workflow. It significantly accelerated the development process.
🤔 Key Product Decisions
Building this wasn't just about integrating an AI model; it was about solving real user pain points.
1. Preserving Original Quality
Most online video tools aggressively compress your uploads to save server costs. I made sure my tool exports MP4 and MOV files at their original resolution. If you upload 1080p or 4k, you shouldn't get a pixelated mess back.
2. Fair Pricing (No Subscriptions)
As a creator, I really dislike being locked into $20/month subscriptions for tools I only use occasionally. So, I opted for a pay-as-you-go (credits) model. You only get deducted for the successfully processed video duration. If a task fails, it costs you nothing.
🚀 What's Next?
I’m currently refining the background repair capabilities for highly complex, dynamic scenes. Video inpainting is incredibly tricky because you have to maintain consistency without causing "flickering" between frames.
I would love for the DEV community to check it out and give me feedback! Whether it's about the UX, the tech stack, or general indie hacking advice, I'm all ears.
🔗 Link: textfromvideoremover.com
Let me know what you think in the comments! 👇
Top comments (1)
The "no re-compression" decision is the one that'll win you actual editors — most online tools quietly destroy the source and call it a feature. Temporal consistency is also the real hard part of video inpainting, so it's good that's where you're putting the effort rather than on the model showcase.
One question: how do you handle a moving watermark (like a broadcast logo that drifts or fades)? Does the mask track, or does the user re-select per segment?