Unwatermark Video Watermark Remover
UnWatermark documents a manual video API that accepts a video plus a matching mask image. WaveSpeed’s Video Watermark Remover is a separate model for logos, captions and on-screen text. The first decision is whether you need to draw and maintain a precise mask or prefer a model-led cleanup path.
Try the tool: Open the video watermark remover model. Review the live charge before running a clip.
Try this workflow
Trim a permitted clip around the hardest moment of watermark motion.
Check the video model charge and submit the sample rather than a full project first.
Review the mark at entry, maximum overlap and exit; reject flicker or drifting edges.
Can one mask cover a moving watermark?
- UnWatermark's manual-video API requires a white-on-black mask image with exactly the source video's resolution. A fixed white region is only a suitable test if it contains the mark at the start, middle and end of the clip. If the mark leaves that region, revise the selection or use a different permitted workflow; the documentation does not establish a moving-mask feature.
Check mask requirements before choosing the API
UnWatermark's manual-video API reference requires both a source-video URL and a mask-image URL. The mask must exactly match the video resolution; its documented bounds are 100 MB, 1080p, five minutes and 30 fps. It also describes a job-creation request followed by status polling. WaveSpeed's video model is a separate submitted-video route with a ten-minute maximum per job. Those published limits are useful for fit assessment, not proof of relative output quality.
| Input situation | More relevant first test | Why |
|---|---|---|
| You can supply a precise mask for a stable region | UnWatermark manual API | You explicitly mark the intended area. |
| You cannot prepare a matching mask | WaveSpeed model playground | The proposed workflow starts from the video input. |
| Clip exceeds one service's documented limit | Trim or split a permitted sample | Reject unsupported input before paying. |
For a licensed 1920 × 1080 clip, create a 1920 × 1080 black-and-white mask and mark only the intended region white before testing UnWatermark. A resized preview mask is not equivalent to the source-resolution input. With a moving bug, inspect whether the mask still covers it at the start, middle and end; a static mask may miss a shifted mark. For either service, download and inspect adjacent frames and audio. Do not infer support for a moving mask or identical billing from these API descriptions. For a completed WaveSpeed result, inspect frames at watermark entry, maximum overlap and exit; this is an evaluation step, not a head-to-head result.
Can you preview the edges before exporting?
- The model result can be previewed after a run in WaveSpeed’s playground, but no frame-by-frame editing control is claimed here.
How to compare a mask-guided output with a model result
- Use a short authorized sample before a long export. Keep its original resolution and frame rate recorded, then compare the same frames after processing.
- Inspect at least one frame before a scene cut, one after it and one during rapid motion.
WaveSpeed AI options for repeat cleanup
- WaveSpeed documents clips up to 10 minutes per job and duration-based billing. Check the displayed estimate before submitting and download reviewed output for your own archive.
Next step: Submit a short authorized clip to the video model and compare motion across the same timestamps you tested with a mask.