Upload a clip, mark the region that lost detail, and rebuild it with AI. It works best on soft blur, mosaic you added, and compressed uploads of your own footage.
Blur throws detail away, so nothing restores the exact original pixels. That is the honest answer to how to de-censor video: soft focus, a blur bar, or compression mush come back convincingly, while a solid censor block returns an invented patch. Knowing which one you have decides whether the result is usable, and a single blurred frame often only needs an Image Enhancer.
A sharpening pass adds contrast to mush; it cannot put back an edge that was never recorded. The model reads each frame, estimates what the region should contain, and repaints it, which is why a blur bar over a sign resolves into readable letters, at a size you can then lift with an AI Image Upscaler.
You draw the box, so nothing else gets repainted. Bars, mosaic patches and soft corners stay contained, faces outside the selection are untouched, and one selection carries across a shot instead of being redrawn every frame.
Only the selected area changes. Grain, lighting and lens character elsewhere stay as shot, so a repaired area does not read as a pasted patch; and when the whole frame is soft rather than one region, a Video Upscaler does the wider job.
Step 01
Use footage you own or are licensed to restore. Short, steady, high-bitrate clips rebuild better than a compressed screen recording, because more of the original edge information survives the upload.
Step 02
Box the blur bar, the mosaic patch or the soft corner. When several areas failed, handle them one at a time, so a weak rebuild on one does not drag down the rest of the frame.
Step 03
Scrub to the same frame and compare against the original. Accept it, lower the strength, or leave the region alone, then finish the cut in the same project without another upload.
Heavy censoring destroys the data, and the page says so instead of promising a miracle.
Only the box you draw changes, so the rest of the frame keeps its grain and lighting.
Restore family tapes, old demos and licensed clips, not footage you have no right to change.
Repaired footage stays in the project beside the brief, ready for the cut and its variants.
Open the workspace without an account and test one region before committing to a full pass.
The rebuild runs in the browser, on the clip you already have open in the project.
Partly. Blur discards information, so nothing can restore the exact original pixels. What AI does is rebuild a plausible, sharp version of the region: soft blur and compression mush come back convincingly, while a solid censor block is replaced by an invention rather than the truth.
Yes, with a limit worth knowing. The workspace opens without an account and covers a few free rebuilds, enough to test one region on one clip. Longer footage or repeated passes can depend on credits, so start with the segment you actually need and check a still with an Image Enhancer before committing.
On footage you own or are licensed to restore, the area behind a mosaic patch or a bar can be rebuilt, and that is the honest version of how to remove a blur bar from a video. Expect a plausible replacement, not recovered truth, and never use it on material you have no right to alter.
Faces are the hardest region because viewers notice when one is slightly wrong. A small, well-lit face inside the box rebuilds best; a wide blur over a whole face gives you a new person, and the tool does not pretend otherwise.
No. Only the marked region is repainted, so grain, lighting and the rest of the shot keep their look. That is what lets a repaired area sit inside an otherwise untouched clip during AI decensor work for videos that need one bad patch fixed.
No. The rebuild runs in the browser, so decensor video online without installing a desktop suite, and the result lands in the same project where a second pass never depends on which machine has the app open.
Upload the clip, box the region that failed, and compare one rebuild before running the whole shot. Keep it only if the replacement still looks like the footage you shot.
Unblur a clip
