Remove an Object or Person from a Video
To remove an object from a video, upload it to Valmera and describe what to take out and when: “remove the bin in the bottom-right corner from 0:12 to 0:30.” The AI agent places a rectangle over it and repaints those pixels, rebuilding the background from other frames where it was visible and from the surrounding picture where it never was.
Logos, watermarks, text and small objects on a steady shot disappear cleanly. A large object over a detailed background, or a person walking through a busy scene, usually leaves a soft patch, and a cut, a crop or a cover is the better edit there. Every erase can be undone with one sentence.
Take It Out of the Shot
Describe what to remove, judge the preview, export. Creating an account and uploading are free; editing requires a subscription.
Create an account →How to Remove an Object from a Video in Valmera
- 1Upload the footageMP4, MOV, M4V, MKV or WebM, up to 100 GB and 3 hours. Valmera indexes it once, so the agent can look at any frame.
- 2Say what to remove, and when“Remove the mic stand on the left from 0:05 to 0:40.” A start and end keep the repaint to the seconds that need it, which makes it faster. Several things can go in one request.
- 3Let the agent repaint and checkIt places the rectangle on real frames, repaints it, renders a preview and looks at the erased area before it reports back.
- 4Keep it, put it back, or exportJudge the patch at full size. If it looks worse than the object, ask to put it back and cut, crop or cover instead. When it looks right, export from Studio; the repaint is rebuilt at full resolution for the final file.
Remove the bin in the bottom-right corner from 0:12 to 0:30.Take the sticker off the laptop lid for the whole video.Remove the mic stand and the light stand in one pass.That patch looks soft. Put it back and crop it out of the vertical version instead.How Video Inpainting Works in Valmera
- Placement. The agent looks at real frames, each carrying a faint coordinate grid, and places a rectangle around the object. For burned-in text, Valmera measures the box from the pixels instead.
- Fill type. A text fill repaints only the letter strokes and keeps the picture between them. A box fill repaints the whole rectangle; that is what an object, a sticker or a solid logo gets.
- A clean plate from other frames. Valmera samples frames across the window and, for every pixel, takes the median of the frames in which that pixel was uncovered. The result is a real photograph of what was behind, which is why changing captions vanish so cleanly.
- Inpainting from the surroundings. Pixels that no frame shows uncovered, such as those behind an object that sits in the rectangle the whole time, are reconstructed inward from the pixels around the box. The edge is feathered so no outline shows, and on a steady shot the fill is computed once and held still.
- A patch on the source. The repaint covers only your window and is tied to the source footage, so later cuts, captions and timestamps don’t disturb it. Undo is instant, and the export rebuilds the patch at full resolution from your original.
No generative model invents pixels: everything in the patch comes from your own footage. The repaint never adds a new object to your shot, and it also can’t recreate detail that no frame ever showed.
What Removes Cleanly, and What Doesn’t
Three things decide the result: how much of the frame the rectangle covers, how detailed the background behind it is, and whether the shot is steady.
| What you want gone | Typical result | If the patch shows |
|---|---|---|
| Burned-in captions, watermarks, usernames | Usually clean: thin strokes are rebuilt from other frames or from the pixels right around them. | Widen the box to include outlines and shadows. |
| A logo, sticker or small object on a steady shot | Usually invisible, especially over a plain or softly textured background. | Tighten the box to the object. |
| A small object over brickwork, foliage or a pattern | Usable, but straight lines and repeating detail may not continue through the patch. | Crop it out or cover it. |
| A large object, or a moving, detailed background | Often a soft patch: the fill has to cover a lot of picture no frame shows. | Crop, push in with a zoom, or cover it with a cutaway. |
| A person standing still at the edge of a steady shot | Can work when they are small in frame. | Crop the frame so they fall outside it. |
| A person walking through a busy scene | Not convincing: the box has to cover their whole path. | Cut those seconds or cover them with b-roll. |
For subtitles and watermarks, remove burned-in text finds and erases every caption band in one request.
Give It a Time Window
Erase windows are set in source time, so they stay on the right footage when you cut the video later. They also set the cost: the repaint works frame by frame inside the window, so a window that covers only the seconds the object is visible finishes fastest. Each window can span up to about 10 minutes of footage, around 7 minutes at 1080p and under 2 minutes at 4K.
Several objects can go in one request. Erasing the same area again replaces the earlier repaint rather than stacking on top of it, so retries never pile up seams.
Erase or Blur?
| Erase | Blur, pixelate or black bar | |
|---|---|---|
| Result | The thing is gone; nothing marks the spot. | Visibly hidden; viewers see that something was covered. |
| Best for | Watermarks, logos, signs, clutter, stray objects. | Faces, license plates, documents, screens. |
| Background | Matters: steady, simple backgrounds repaint best. | Makes no difference. |
| Length | Up to about 10 minutes of footage per window. | Any length. |
When viewers should see that something was hidden on purpose, a censor is the right answer: see blur part of a video, or for people, blur faces and license plates.
When the Patch Shows: Four Better Edits
- Cut it. If the object is on screen for a few seconds, cut those seconds. A clean cut is less noticeable than a visible patch.
- Crop it out. Something at the edge of a 16:9 shot often falls outside a 9:16 or 1:1 crop aimed at your subject; auto reframe does that in one request.
- Push in. A zoom over those seconds pushes edge content off screen and reads as emphasis rather than a fix.
- Cover it. Lay a full-frame b-roll cutaway over the moment while your audio keeps playing, or use a censor.
For a single hero shot where a person walks through a detailed scene, a desktop compositing tool with tracked masks, or a dedicated video inpainting model, will do better than any fixed-rectangle repaint.
Limits
Rectangles only: there is no brush, mask or rotoscoping, so everything inside the box is repainted. No object tracking: the rectangle is fixed for each time window, so something that moves needs a box covering its whole path. No generative fill: large areas over detailed or moving backgrounds come back soft. Each window is limited to about 10 minutes of footage (around 7 at 1080p, under 2 at 4K). And it is picture only: someone erased from the frame can still be heard, so cut or mute that audio separately.
Related Workflows
Listing videos are a common use: a bin, a hose or a car in the driveway can go, as long as you don’t hide permanent defects. See real estate listing videos. Every erase and censor option is in the effects docs.
Creating an account and uploading are free. Editing requires a subscription — see plans.
Using Claude, ChatGPT or Cursor? The same erase_region tool is available through Valmera's MCP server.
Frequently Asked Questions
Made with Valmera: real exports
Remove an Object from Your Video
Describe what to take out, review the preview, export a full-quality MP4. Creating an account and uploading are free; editing requires a subscription.
Create an account →