Recover your photos and videos from STOP/Djvu ransomware — no ransom, no key.
STOP/Djvu is one of the most widespread ransomware families aimed at home users. It
renames files with extensions like .coot, .djvu, .boop, .mzlq and leaves a
_readme.txt asking for about $490–$980. Files hit by its newer variants with an online
key cannot be decrypted by anyone except the attackers.
But STOP/Djvu only encrypts the first 150 KB of each file. In a typical 1–4 MB phone photo, 85–97% of the image data is still there, untouched. PhotoPhoenix rebuilds a viewable JPEG from that surviving data, and plays videos again (see Videos). It works entirely offline, on your own computer.
| What you get | When |
|---|---|
| 100% of the photo | an exact, untouched copy of the file is on a drive you point PhotoPhoenix to — rare; see below |
| Nearly 100% | the photo's own JPEG header happens to lie beyond the encrypted part |
| Most of the photo, exact colours | the camera used JPEG restart markers (common on older phones) |
| Most of the photo, estimated colours | everything else with enough data — the top strip that was encrypted stays grey |
| Nothing | files smaller than ~150 KB were encrypted completely |
PhotoPhoenix does not decrypt anything and never contacts the attackers or any server. If you are a victim: don't pay, keep the encrypted files (keys sometimes leak and the free Emsisoft STOP/Djvu decryptor may help with offline keys), and scan your PC — cracked software "activators" are a common infection route.
- Install Python 3.11+ (tick "Add python.exe to PATH").
- Download or clone this repository, then in its folder run:
For videos, also install ffmpeg (on your PATH) and, for their sound,
pip install -r requirements.txtpip install -r requirements-optional.txt. - Double-click
Run PhotoPhoenix.bat. - Choose the folder with the encrypted files and an empty folder for the results, pick Recover: Photos and videos / Photos only / Videos only, and press Start.
The first run learns your cameras' settings from healthy photos and videos (see below), then
recovers every *.jpg.coot / *.jpeg.coot photo and *.mp4.coot / *.3gp.coot / *.mov.coot
video. The Working on list shows each file in progress and its current step. Pause, Resume
and Stop at any time; starting again with the same folders picks up where it left off.
Linux / macOS: pip install -r requirements.txt, then PYTHONPATH=src python -m photophoenix.
- The encrypted files. Work on a copy if you can; PhotoPhoenix never modifies them anyway.
- Some healthy, unencrypted JPEGs, ideally from the same phones or cameras (photos you sent to others, cloud backups, an old SD card, a folder the ransomware missed). PhotoPhoenix learns each camera's JPEG settings from them. Without any, it can still infer most settings from the data itself, just less reliably.
- For videos whose index was encrypted: one original video from the same phone model (yours, a friend's, or a review site's camera sample — see below).
- A computer with Python 3.11+. No GPU, no accounts, no API keys, no paid services. About 200 MB of RAM per worker; 4 workers by default.
The grey band at the top is the part of the picture that was inside the encrypted 150 KB. It can't be recovered: it is scrambled with a key only the attackers have. Only an untouched copy (or the key leaking one day) can bring it back.
How big it is depends on the file size, because the encrypted part is always 150 KB:
| Photo file size | Share of the picture lost | What you see |
|---|---|---|
| ~3–4 MB (recent phones) | ~3–6% | a thin strip, easy to miss |
| ~1–1.5 MB (phones around 2014–2016) | ~10–20% | a clear band |
| ~300–500 KB (old phones) | ~30–75% | most of the top half |
| under ~150 KB | 100% | nothing to recover (reported as failed) |
Part of the 150 KB is camera metadata, so photos with a lot of it lose even less.
No grey area at all when the photo was restored from an untouched copy, or rebuilt from its own header (both complete). Also, when PhotoPhoenix can't tell a camera's full image height, it saves only the rows that survived: such a photo has no grey band but is shorter than the original — its top is missing.
Your choice in the app (Grey area): Keep it (default — original size, the band shows
honestly what is missing), Crop it off (grey rows removed; the photo is shorter), or Save
both (the full photo plus … (cropped).jpg). Cropping is lossless: the picture data is copied,
not re-compressed. report.csv marks cropped photos. Changing the choice later redoes the
affected photos. Setting: grey_area in config.toml; command line: recover --grey crop.
Videos are large, so the encrypted 150 KB is usually less than 1% of the file. What comes back depends on where the video's index (the table of where every frame is) was stored:
| Index stored | Typical source | Result |
|---|---|---|
| At the end — survived | most Android phone camera apps | The whole video plays, full length, with sound. Only the first moments (whose data was encrypted) are glitchy — report.csv gives an estimate, usually 1–3 seconds. |
| At the start — encrypted | many phone camera apps, iPhones, chat apps, editors | Picture and sound — if a healthy video from the same phone exists. PhotoPhoenix finds every H.264 frame in the surviving data, measures every AAC sound frame with FFmpeg's decoder, and writes a new index using a healthy reference video's settings, timed by the sound so lip-sync holds. Needs ffmpeg installed; sound needs pip install -r requirements-optional.txt (PyAV), otherwise picture only. Without a matching reference the video is listed as failed with that reason; add healthy videos from that phone via Learn cameras from…. |
| — | videos under ~150 KB | nothing to recover |
On one real 250 GB drive, 196 of 440 encrypted videos had their index at the end. Of the rest,
a sample rebuilt only where a healthy video from the same phone existed (e.g. 45 s and 141 s
1080p videos); the healthy "reference" from one phone turned out to be broken itself, so its
videos couldn't be rebuilt — until a camera sample from a phone review site was used instead
(see below). Rebuilt videos had sound with no audio decode errors. Untouched
copies (see below) are restored complete for any video. Choose what to recover in the app
(Recover: Photos and videos / Photos only / Videos only) or with recover = "photos".
A rebuilt video only needs the settings of another video from the same phone model, not its content. If you have none, someone else's original works just as well:
- Find an original, unedited video from the same phone model: your own on the phone, an SD card or a backup; a friend's (sent as a file); or a phone review site's camera samples. GSMArena reviews link the phone's own sample videos on the review's Camera page (download the file itself, not a YouTube version).
- Put it in a folder of its own and add that folder with Learn cameras from….
- Press Start again. Videos that failed for lack of a reference are retried.
Videos shared through chat apps, social media or YouTube are re-encoded and won't work. Every reference is tested on a short clip first and used only if the video decodes cleanly, so a wrong sample can't make anything worse. Example: no healthy Nokia 5 video existed on the drive above; GSMArena's 2017 Nokia 5 review sample rebuilt three 2019 Nokia 5 videos (2–7 minutes, 1080p, with sound, 0.2–2.3% of frames glitchy). The app shows these steps when it finishes a run with such videos.
A JPEG starts with a header: the "instructions" for reading the picture (size, compression tables, how colour is stored). STOP/Djvu destroyed that header in every photo, but each phone or camera model writes the same kind of header on all its photos. So PhotoPhoenix borrows the header of a healthy photo from the same camera model to read what survived. No part of the healthy photo's picture is used.
- First Start: it learns automatically from healthy JPEGs in
donor_dirs(default: the encrypted-photos folder). - Later, more sources: press Learn cameras from…, add folders, or Add whole PC for
every drive, and press Start learning. New cameras are added to the existing library.
Command line:
python -m photophoenix build-library D:\ E:\Backup. - What is read: only the first ~128 KB of each JPEG (the header). System, program and app-cache folders, tiny images and encrypted files are skipped. It runs at low priority and can be stopped at any time; whatever was learned is kept.
- Videos too: the same learning also notes the settings of healthy MP4/MOV videos it finds
(
local/video_refs.json), used to rebuild videos whose index was encrypted. - What it helps with: a camera PhotoPhoenix has a healthy photo from gets the most reliable repair. Without one it estimates the settings from the data, which works for most photos but occasionally gets colours wrong. It does not bring back the encrypted top part of a photo.
The Find untouched copies button checks whether an exact, unencrypted copy of any of your encrypted photos sits in folders you choose. A copy gives the whole photo back. It is worth one try, but for most people it finds few or no copies. Please read this before relying on it.
How it decides a file is a copy. STOP/Djvu keeps a file's size (plus 334 bytes it adds at the end) and leaves everything after the first 150 KB untouched. So a file that is exactly 334 bytes smaller and whose bytes after 150 KB are identical is the same file, even if it was renamed. Files smaller than about 160 KB can't be compared this way and are matched by name and size only (less certain).
What it will find
- The exact same file saved twice: a folder you copied to another drive, a backup, a USB stick or SD card, an old phone's folder you copied to the PC, a file transfer app's folder (SHAREit, Xender, Nearby Share), photos you received in WhatsApp and saved elsewhere.
What it will NOT find
- Copies that were re-saved or compressed. Photos you sent through WhatsApp, Messenger, Facebook, Instagram or e-mail apps that shrink images are different files. Their pixels may look the same, but their bytes are not, so they don't count as copies. (They can still help as "healthy photos" for learning your camera's settings.)
- Anything in folders you didn't choose. It doesn't search your whole computer, your phone,
or the cloud. To use Google Photos, OneDrive, iCloud or a phone, first download/copy those
photos into a folder, then add that folder to
copy_search_dirs. - Copies on drives the ransomware also encrypted. STOP/Djvu encrypts every drive connected at the time, including USB drives that were plugged in.
What happens
- It shows you the exact folders it will read and asks before starting.
- It only reads files; it never changes, moves, deletes or uploads anything.
- It tells you how many encrypted photos have a copy (and how many of those were verified
byte-for-byte), and saves the list in
local/copies.csv. - On Start, photos with a copy are restored by copying that file; all others are repaired as usual. Finding no copies changes nothing else.
Real-world example: on one victim's 250 GB drive with 21,000 encrypted files, about 1 in 20 had an untouched copy, almost all of them WhatsApp images that had also been moved to another folder by a file transfer app. Camera photos almost never had copies.
Defaults work for most people. To change them, copy config.example.toml to config.toml
(or .env.example to .env) and edit. Every setting can also be an environment variable
PHOTOPHOENIX_<NAME>. The important ones:
| Setting | Meaning |
|---|---|
source_dir / output_dir |
default folders shown in the app |
donor_dirs |
folders of healthy JPEGs to learn cameras from |
copy_search_dirs |
folders searched for untouched copies — nothing else is ever searched |
recover |
both (default), photos or videos |
grey_area |
keep (default), crop or both — see The grey area |
workers, min_free_ram_gb, low_priority |
how hard PhotoPhoenix works your computer |
python -m photophoenix # the desktop app
python -m photophoenix build-library DIR [DIR] # learn cameras from healthy photos and videos
python -m photophoenix find-copies # find untouched copies (source_dir, copy_search_dirs)
python -m photophoenix recover FILE.jpg.coot -o OUT.jpg
python -m photophoenix recover FILE.mp4.coot -o OUT.mp4
(Set PYTHONPATH=src, or pip install -e . once and use the photophoenix command.)
PhotoPhoenix runs offline and reads only the folders you choose. The camera library
(local/library.json), reference-video list (local/video_refs.json) and copy list
(local/copies.csv) contain paths to your photos; they
stay on your computer and are excluded from git.
- ARCHITECTURE.md — components, data flow, stack, how to extend
- docs/methods.md — the recovery techniques and why they work
- docs/djvu-format.md — what STOP/Djvu does to a file, measured
- docs/validation.md — how each technique was tested, with results
- docs/safety.md — running gently on old or failing hardware
- CHANGELOG.md · CONTRIBUTING.md · COMMERCIAL.md
- The encrypted top part of each photo is gone; it is shown as a grey band or cropped off (see The grey area).
- Colours are estimated when the camera didn't use restart markers; usually close, sometimes off.
- Photos: only baseline JPEG (no PNG or progressive JPEG yet).
- Videos whose index was encrypted: only H.264 video with AAC sound (most phones since ~2012), and only with an original video from the same phone model. Old 3GP videos (H.263/MPEG-4) and AVI aren't supported yet. A few frames may glitch; the first moment of every video is lost.
- No documents (PDF, Word, …) or music yet.
- Photos re-saved by editing apps with custom ("optimised") Huffman tables can't be matched.
Free for non-commercial use under the PolyForm Noncommercial License 1.0.0: personal use, research, education, charities and public institutions. Commercial use requires a separate agreement — see COMMERCIAL.md. Contributions are welcome under the terms in CONTRIBUTING.md.