This video was shot while traveling, on an iPhone 13 Pro with no external mic. The sound is pretty typical for that. Even a video like this can be cleaned up with AI noise removal so that only the voice is left. In video editing, you can also layer the voice-only audio over the original sound to make it easier to hear while still sounding natural.
Save the video to Files (recommended)
Videos you shoot on your iPhone are stored in Photos. Shizuka can load videos from “Photo Library” too, but when you choose one from “Photo Library” in an iPhone browser, the iPhone re-compresses the video before handing it to the browser. This is how iPhone (iOS) works, and it's the same for any browser-based tool, not just Shizuka. If you want to keep the original quality, it's best to save the video to the Files app first.
- Open the video in the Photos app and tap “Share”
- Choose “Save to Files”
- Pick a location (such as “On My iPhone”) and tap “Save”
| Choose from Files (recommended) | Choose from Photo Library | |
|---|---|---|
| Frame rate | 60fps | 30fps |
| HDR | Yes | No (narrower color range) |
| Format | HEVC (10-bit) | H.264 (8-bit) |
| Size (15 sec) | 151MB | 45MB |
Remove the noise with Shizuka
- Open Shizuka in your browser and tap the “Drag and drop an audio or video file… choose one” box
- Tap “Choose File” and select the video you saved in STEP 1 (if you skipped STEP 1, choose it from “Photo Library”)
- Under “Select AI model,” choose DPDFNet (High Quality)
- Press “Start Noise Removal”
Save it and put it back in Photos
- Switch between “Original” and “Processed” to compare how they sound
- Press “Download video (MP4)” to save it (it goes into “Downloads” in the Files app)
- Open the video in the Files app and tap “Share” → “Save Video” to put it back in the Photos app
The picture is left untouched
Shizuka uses AI to rebuild only the sound in the video and keeps the picture exactly as it is. Here's what we found when we chose a video shot in 4K at 60fps with HDR from Files and compared the result:
There's no feature for removing grain in the picture (noise in dark areas).
Videos where the noise is hard to remove
Shizuka is good at removing noise playing behind voices: wind, street chatter, air conditioners and ventilation fans. It struggles with videos like these:
- Videos with music: background music and singing get turned down as “noise” too
- Videos with strong reverb or echo: the echo of a large hall or a gym stays in
- Large videos: in the browser, the limit is 1GB. A video shot in 4K at 60fps goes over 1GB in less than 2 minutes (our 15-second test video was 151MB)
If you want to layer the voice-only audio over the original sound in video editing, you can pull just the audio out of the processed video with Ototori. To remove all the sound from a video, you can use Otokeshi.
FAQ
Can't I just choose the video straight from “Photo Library”?
Will the recording date, time and location be kept?
Does it also remove grain (noise) in the picture?
Is my video uploaded to a server?
What if I want to extract the voice-only audio, or remove all the sound?
Note: The example videos were shot by the author on an iPhone. They were processed with the same AI model as the public version of Shizuka.
About the author
Remove the noise from your video, keep the picture quality
Shizuka is free, with no sign-up. Your videos are never uploaded.
