Epstein Photos Network Visualization

PostedAugust 5, 2026

UpdatedSeptember 6, 2026

ByEpsteinWiki

Epstein.photos is an interactive visual research tool created by Decoherence Media. It uses facial recognition, manual verification, image analysis, and network mapping to identify people who appear in photographs released through the United States Department of Justice Epstein Library and House Oversight disclosures.

The tool connects two people when they appear in the same photograph. Researchers can explore those relationships through an interactive graph, browse identified and unidentified faces, search by name or document number, and inspect clusters of visually similar images.

Epstein.photos does not claim that every person shown knew about or participated in Jeffrey Epstein’s crimes. A photographic appearance can establish presence in an image. It does not independently establish the nature of a relationship, criminal conduct, or knowledge of abuse.

Snapshot

Website: Epstein.photos

Creator: Decoherence Media

Launch date: May 6, 2026

Resource type: Facial recognition database and photographic network visualization

Named identities: 433 manually verified people when the methodology guide was published

Unidentified identities: 151 people included as unknown subjects

Source images extracted: 2,751,081

Exact duplicates flagged: 153,240

Images with detected faces: 23,421

Primary sources: Department of Justice Epstein Library and selected House Oversight productions

Core technology: AWS Rekognition, InsightFace, Gephi, NetworkX, D3.js, DINOv2, and UMAP

Source code: Publicly available on GitHub

Key safeguard: Images containing victims or nudity are not displayed

What Is Epstein.photos?

Epstein.photos transforms a large photographic collection into a visual map of people and photographic appearances.

The About page explains that each circle in the main graph represents a person. A line connects two people when they appear together in at least one photograph.

Users can hover over a person or connection for a summary. Clicking opens additional information and a preview image. Opening a person or connection in the search view displays photographs containing that person or combination of people.

The website was created by Decoherence Media, a nonprofit newsroom that uses open source and data driven methods.

The Four Main Research Views

Epstein.photos provides four complementary ways to examine the collection.

The Graph view maps photographic connections. People are represented as nodes, while shared photographs create edges between them.

The People view displays face clusters and identity labels. Researchers can filter named, unknown, unreviewed, and excluded subjects.

The Search view accepts one or more names or an EFTA document number. File extensions and punctuation variations can be ignored when matching a document identifier.

The Explore view arranges visually similar photographs near one another. This can reveal repeated locations, events, clothing, activities, and groups that may be difficult to find with a name search.

How The Network Graph Works

The main graph is a coappearance network. Two people receive a connection when they appear in the same image.

Node colors represent broad categories, often based on occupation or social role. Examples include fashion and modeling, academia and science, politics, finance, and people associated with the United States Virgin Islands.

The project documentation describes node size as a measure of photographic prominence. The current About page says size reflects the number of different people with whom a person appears. The original project guide describes size in relation to the number of images containing that person.

Researchers should therefore treat node size as a visual navigation aid rather than an exact measure of importance, influence, culpability, or relationship strength.

A thick line or dense cluster means people repeatedly appear together in images. It does not explain why they were together or what occurred outside the frame.

How The Image Pipeline Was Built

Decoherence Media published a detailed methodology guide and the processing pipeline on GitHub.

The project reports that it extracted 2,751,081 images from the source PDFs. It flagged 153,240 exact duplicates and detected faces in 23,421 images.

AWS Rekognition was used to index and compare faces. InsightFace provided an initial local face detection step. The team used AWS moderation tools and manual review to detect nudity and sensitive material.

The pipeline grouped similar faces into clusters. Researchers then attempted to identify each cluster through reference photographs, reverse image searches, public records, news reporting, social media, and contextual evidence from released emails.

The network was constructed with NetworkX and Gephi, then displayed with D3.js.

Facial Recognition And Manual Verification

The project used a 99 percent facial similarity threshold for comparisons with reference images. Decoherence Media says every named identification was manually verified.

Automated similarity scores are not proof of identity. Facial recognition can produce false matches, merge two people into one cluster, or separate photographs of one person into several clusters.

Age, image quality, lighting, camera angle, partial faces, glasses, compression, and changes in appearance can affect results.

Manual review reduces those risks but does not eliminate them. Any identification used in an EpsteinWiki article should be compared with the original image and at least one independent reference source.

The website provides forms for submitting a new identity or reporting a correction.

Named, Unknown, Excluded, And Unreviewed Faces

The project uses four identity statuses.

Named refers to people whose identities were manually verified. The project reported 433 named identities when its guide was published.

Unknown refers to 151 unidentified people considered potentially relevant because of their apparent role or proximity to identified people.

Excluded refers to 6,630 face clusters that did not meet the project’s inclusion rules. This category includes people appearing only inside news screenshots, memes, thumbnails, cartoons, or other material that did not appear to be an original photograph connected to the archive.

Unreviewed refers to 4,242 remaining face clusters that had not yet received a full inclusion or identification review.

These numbers describe the state of a developing database and may change as the project processes corrections and additional material.

The Original Photograph Standard

The released files contain many kinds of images. They include original photographs, scanned documents, social media screenshots, news articles, video thumbnails, identification cards, memes, and repeated copies.

Decoherence Media requires at least one apparently original photograph before including a person in the public network.

The project gives Alex Jones as an example. His face appears repeatedly in news article screenshots within the files. Those appearances do not establish that he was photographed with Epstein or belonged to Epstein’s network.

This distinction is essential. A person whose face appears in a screenshot about the case is not equivalent to a person shown in an original private photograph.

Researchers should always determine what type of image produced the match.

Searching By Person Or Document Number

The photo search allows users to enter one or more names. A combined search can isolate images in which two identified people appear together.

Users can also search by EFTA document number. This creates a direct path from a document citation to the photographs extracted from that file.

Search results place boxes around identified faces. The original source file can be opened from the results page.

Researchers should record the EFTA number, source page, identity label, and date of access. Names, classifications, and image counts can change when corrections are applied.

How The Photo Atlas Works

The Epstein Photo Atlas groups visually similar images into clusters.

The project created mathematical image representations with DINOv2. It then used UMAP to place similar representations near one another in a two dimensional space.

Nearby images may share a location, composition, activity, object, or subject. Clusters can reveal groups of party photographs, identification documents, vehicles, tables, water activities, interiors, or repeated scenes.

Proximity does not prove that two photographs were taken at the same event. The model measures visual similarity, not historical identity. Researchers must compare backgrounds, clothing, architecture, metadata, and source documents before joining photographs into one event.

Survivor Privacy And Content Safeguards

The site’s content policy places survivor privacy at the center of the public interface.

People identified as victims are anonymized with labels such as Victim 1. Their real names are not displayed.

Images containing victims are not served by the website, even when the image contains no nudity. Victim nodes may remain in the network so that its overall structure is visible, but the associated photograph is withheld.

Images containing nudity are also excluded. The project used automated moderation followed by manual review because automated systems allowed some false negatives.

People who appear only as minors are not publicly identified. Young women without evidence of a more significant association were generally excluded from the network.

These safeguards reduce harm, but users should still report any mistaken identification or sensitive image through the project’s correction form.

What A Photographic Connection Can Establish

A verified original photograph can establish that two visible people appeared in the same image.

When provenance is reliable, it may also help establish a location, event, approximate date, recurring social group, property, aircraft, item of clothing, or sequence of photographs.

Repeated coappearances can support a finding that two people were photographed together more than once.

The image must still be evaluated for cropping, editing, duplication, source context, and whether it depicts a photograph inside another document.

What A Photographic Connection Does Not Prove

An edge in the graph does not prove friendship, business dealings, travel, abuse, conspiracy, criminal participation, or knowledge of Epstein’s conduct.

A large public event can place strangers in the same frame. A group photograph can include people who never interacted. A private photograph can document social access without revealing what the subjects knew.

The number of photographs may also be affected by duplication, bursts of similar frames, uneven preservation, and the particular files chosen for government release.

Network position must never be described as a measure of guilt or complicity.

Relevant Epstein Data Evidence

Epstein Data provides additional tools for checking image results against the original evidence collection.

The Epstein Data image database contains more than 92,000 analyzed image records. Each record can include an EFTA number, source PDF, page number, extracted text, visible objects, setting, activity, and an automated description.

The Epstein Data research database also provides reverse image and face search across a larger indexed image collection. These tools can help locate duplicate photographs, related scenes, and the original EFTA source.

EFTA00068334 discusses the government’s possession of approximately 40,000 images described as not containing nudity, along with thousands of nude or partially nude images. This record helps explain the scale and sensitivity of the photographic evidence handled by investigators.

EFTA00013222 contains related court language concerning the seized image collection.

The FBI digital evidence investigation examines image media and evidence directories, including materials associated with Little Saint James and other locations.

Researchers should use these sources to verify document numbers and image provenance before publishing an identification.

How To Verify An Epstein.photos Result

A responsible verification process should include several steps.

Open the identified person or connection in the Search view.
Record the identity label and EFTA document number.
Open the original source file.
Confirm that the image is original rather than a news screenshot, thumbnail, or photograph of another image.
Compare the face with reliable reference photographs.
Inspect other images assigned to the same face cluster.
Check whether the image contains a date, location, caption, or recognizable event.
Search the EFTA number in Epstein Data.
Compare the result with emails, calendars, flight logs, financial records, and testimony.
Report uncertain or incorrect identifications to the project.
State only what the photograph proves.

How The Tool Supports EpsteinWiki Research

Epstein.photos makes visual relationships easier to examine across a massive and inconsistent evidence release.

The graph can identify recurring groups. The People view can surface unknown faces. The document search can connect an EFTA number with its extracted photographs. The Atlas can reveal repeated settings and visually related scenes.

The tool complements the Epstein Search Hub, EpsteinWiki research on mapping the Epstein network, and the Dafydd Jones Jeffrey Epstein photo archive.

EpsteinWiki should cite the original EFTA record whenever possible. Epstein.photos should be credited as the identification and discovery tool, while the source document should remain the evidentiary citation.

Key Takeaways

Epstein.photos maps people who appear together in photographs released by the Department of Justice and House Oversight.
The project reported 433 manually verified named identities and 151 unidentified included subjects.
Its pipeline extracted approximately 2.75 million images and detected faces in 23,421 images.
Users can browse a network graph, a face directory, a document search, and a visual photo atlas.
Searches can combine multiple names or use an EFTA document number.
The project uses AWS Rekognition with a 99 percent similarity threshold and manual verification.
Facial recognition can still produce errors and must be independently checked.
The project excludes people who appear only in news screenshots or other nonoriginal images.
Images containing victims or nudity are not displayed.
People shown only as minors are not publicly identified.
A connection means two people appear in the same photograph. It does not establish guilt, knowledge, or the nature of their relationship.
Original EFTA records should be cited through Epstein Data whenever a photograph is used as evidence.

Why Epstein.photos Matters

The Epstein files contain an enormous number of visual records mixed inside PDFs, screenshots, duplicates, news articles, and evidence files. Traditional text search cannot reliably identify a face or connect repeated appearances across that collection.

Epstein.photos provides a visual index for those relationships. It helps researchers move from a face to a name, from a name to a photograph, and from a photograph to an EFTA source.

Its greatest strength is discovery. Its greatest risk is overinterpretation. Used carefully, the tool can reveal patterns and locate evidence. The original photograph, source document, and corroborating records determine what can responsibly be claimed.

Sources

Tags:

Searchable Epstein Databases