Inspiration
A good video can still have one distracting detail: a passerby in the background, an unwanted object, or something that draws attention away from the subject. Reshooting is not always possible, and editing masks frame by frame takes time.
I built RemoveObjects.video to make this kind of cleanup more accessible. The idea is to let people describe what they want to remove, check the selection, and process a short clip without installing desktop editing software.
What it does
RemoveObjects.video is an online AI video object remover with a guided workflow:
- Upload a video and select the section you want to edit.
- Choose a frame and describe the target, such as “the person wearing a red jacket.”
- Preview the detected mask to check that the right subject has been selected.
- Run the removal and download the processed video.
The service builds a mask across the clip and uses video inpainting to fill the selected area. A free daily trial supports clips up to 6 seconds with watermarked exports. Credit-based standard processing supports object-removal clips up to 50 seconds and exports without a watermark.
A separate AI video repainting tool offers another way to change video content with a text description. The interface is available in 20 languages.
How I built it
The frontend uses TypeScript, React, and React Router. Cloudflare Workers handles requests, R2 stores media, D1 stores task and billing records, and Workflows coordinates processing across services. FFmpeg runs in Cloudflare Containers to prepare clips, align media, and assemble the final output.
For object removal, SAM3 generates the video masks and ProPainter performs inpainting through Replicate. I also built a hybrid processing path that reuses background detail from the source footage to reduce softness in larger removed areas. The repainting tool uses models accessed through KIE.
The product combines these steps into one editing flow, with progress updates, automatic retries for specific recoverable failures, and credit settlement after a paid task succeeds.
Challenges I ran into
The main challenge was keeping the whole video pipeline consistent. A mask can look correct on one frame but miss the target later. Frame counts, dimensions, and timing also need to match between preprocessing, segmentation, and inpainting.
Another challenge was balancing quality, waiting time, and processing cost. I worked on frame alignment, background reconstruction, queue handling, and automatic resolution adjustments. Idle containers stop when they are no longer needed to keep infrastructure costs under control.
Accomplishments that I am proud of
- Launched a working video editing service with a free daily trial and optional credit-based processing.
- Added mask previews so users can check the target before starting removal.
- Built a hybrid background reconstruction path to improve detail in difficult removal regions.
- Connected uploads, processing, progress reporting, downloads, and billing in one multilingual interface.
What I learned
The final result depends on more than choosing an AI model. Video preparation, mask coverage, frame alignment, and the editing interface all affect the outcome.
A useful prompt identifies the target by appearance, clothing, or color. Descriptions based only on “left” or “right” can become ambiguous when the subject moves. Giving users a preview helps them make a better selection before processing.
Try it out
Try the free AI video object remover in English, العربية, Español, or 日本語 with a short clip. Describe the subject you want to remove, review the mask preview, and see how it works on your own footage.
Log in or sign up for Devpost to join the conversation.