How it works
Two tools, one honest page. The Metadata Cleaner makes a file say nothing about where it came from. The Spoofer makes a copy that is a new file to a duplicate matcher and, when you want it, describes itself as a photo or video from a phone you pick. Neither changes what a fan sees.
01Where it runs
Every tool runs locally with ffmpeg and exiftool bundled in the app. Nothing is uploaded, nothing waits in a queue, a folder can run the whole pipeline by itself. Multi-gigabyte files are fine.
The file goes to our worker over an encrypted upload, is processed, and is deleted after the retention window shown on the tab. Handy from a phone; Studio is the one to use when the file should never leave your machine.
02What gets removed
Every metadata block is discarded, not edited. Editing leaves fragments; starting from an empty container does not.
- EXIF: camera, lens, serial numbers, settings, orientation history
- GPS coordinates, altitude and the GPS clock
- XMP: editing history from Lightroom, Photoshop, CapCut, Canva
- IPTC: author, copyright, captions, keywords
- Embedded preview thumbnails and second images (MPF)
- C2PA / JUMBF content credentials and AI provenance tags (SynthID, "trainedAlgorithmicMedia", generator names)
- Comments and every other application segment
- Make, model, software and creation date atoms in the MP4 / MOV header
- The muxer signature that says a file came out of ffmpeg, HandBrake, CapCut or an editor
- Location (ISO 6709) written by phones and drones
- XMP blocks and
uuidboxes - Per-track user data and generic track names
- Creation and modification times stored in every track
In the Cleaner the compressed picture and the audio and video streams are copied across byte for byte, so the file hash changes while the content stays bit-identical. Turn on Reshuffle pixels and the image is re-encoded with an invisible colour and grain jitter, which also removes anything hidden inside the compressed data itself.
AI strip goes one step further on generated media: provenance blocks are dropped and a small corner is cropped away, so a watermark tile placed by the generator leaves the frame.
03What gets written
A spoofed photo carries the full set of fields a phone writes, not four tags and a gap. A partial set is itself a signal: real cameras do not stop halfway.
04Strength levels
The Spoofer re-encodes every copy with its own crop, rotation, speed, colour, grain and audio drawn from a band. The level sets the width of the band. Each variant draws its own values, so a batch of twenty lands twenty different distances from the original.
Crop under 1 %, rotation under 0.15°, colour and grain you cannot see. Survives a side-by-side look. Enough to produce a new hash and a new file; not aimed at duplicate detection.
Crop 1-2 %, a hair of rotation and speed, light grain. The usual choice for reposting on the same account.
Crop 3-5 %, rotation up to half a degree, speed ±2 %, colour and tone shifts across the whole frame. With Mirror and Scramble audio this is what most creators run across accounts.
Crop up to 9 %, rotation over a degree, visible warmth and contrast changes, heavier grain. Aimed at the downscaled fingerprint matchers use, at the cost of a copy you can tell apart next to the original.
Everything, wide. Only for clips that keep getting caught; the drift is obvious.
Why geometry and tone and not a few pixels: duplicate detection never compares file hashes. It compares a heavily downscaled fingerprint of the picture. Changing a handful of pixels does nothing to that fingerprint; changing the framing and the tone of every frame does.
05Time and consistency
Plausible values matter less than values that agree with each other. A file is rarely caught on one wrong field; it is caught on two fields that contradict.
- Exposure follows the clock. A frame timed at two in the morning gets a night exposure; a midday frame gets a bright one. Shutter, ISO and brightness are derived from one another so the three always add up.
- Two clocks, one moment. The capture time is local, the GPS clock is UTC, and the offset between them is the declared time zone - exactly how a phone records it, daylight saving included for that date.
- A batch looks like a session. Files spoofed together are spaced minutes apart and dated a little back, not stamped with one identical second.
- The video header agrees with the streams. Creation and modification times, track durations and the declared make and model are written together, never half of them.
06Location
Three settings. Off writes no GPS at all - that is what a phone with location access denied to the camera produces, a common and unremarkable state. USA, random city gives every file a different real city with its coordinates, altitude and time zone. Custom takes your own coordinates.
Each variant is nudged a few hundred metres at random, so a batch never repeats one exact point and files from different sessions do not share coordinates. Upscaler and Spoofer write location the same way; the Cleaner never inherits it - clean means clean.
07Video and audio
The Cleaner rewrites the header and copies the audio and video streams byte for byte. No re-encoding, no quality loss, no waiting - a large file is processed about as fast as it can be read. Only the header is held in memory.
The Spoofer re-encodes. Every variant gets its own crop, speed and colour on the picture and its own high-pass, low-pass, gain, slow stereo pan and pitch on the sound; Scramble audio adds a frequency shift on top, which is what defeats audio fingerprinting. The result is written with a fresh container, a plausible app-export handler and no muxer signature, then the chosen device identity goes on top.
08Formats
| Tool | Takes | Gives back |
|---|---|---|
| Studio Spoofer | Any photo or video ffmpeg can read - JPEG, PNG, WebP, HEIC, MP4, MOV, WebM, MKV, AVI | JPEG for photos; MP4 or MOV for video, sized for the platform you pick (9:16, 1:1, 4:5…) |
| Studio Cleaner | The same | Same format in, same format out; pixels bit-identical unless you reshuffle |
09What it cannot do
Being straight about the boundaries is more useful than promising everything.
- A cleaned video keeps its visual fingerprint. Because the Cleaner copies the streams untouched, metadata work alone will not defeat a system that matches videos by their frames. That is what the Spoofer's levels are for.
- A re-encode is still a re-encode. The compressed data of a spoofed file comes from an encoder, not a sensor. The header says app export rather than camera recording for exactly that reason - a file that claims to be a raw iPhone recording with x264 inside contradicts itself.
- The frame still has to make sense. A screenshot given camera metadata is still shaped like a screenshot, and a picture of a screen still looks like one. Metadata describes a file; it does not change what the picture is.
- No setting is a guarantee. Platforms change their matching without notice. Subtle tiers survive a side-by-side look, high tiers are aimed at duplicate detection, and you stay responsible for what you repost and where.