The Faces Behind the Advertisements: Using Image and Video Forensics to Trace Reused Promotional Media

Investigative media-forensics workstation examining blurred photographs, videos, metadata and file fingerprints in a Telegram advertising investigation.

Cyber Truth Times examines Telegram image and video forensics, metadata, hashes and perceptual similarity to investigate reused promotional media across Telegram datasets.

Introduction: Telegram Image and Video Forensics

Photographs and videos are among the most prominent elements of online promotional advertising. In messaging-platform channels, visual material may accompany profile descriptions, promotional claims, pricing information, contact references and other content intended to attract an audience.

But visual media also presents an investigative opportunity.

A photograph may appear in more than one advertisement. A video may be redistributed in a different context. A file may contain metadata that provides clues about its processing history. Two visually similar images may have different file hashes because they were resized, compressed, cropped or edited.

These possibilities make image and video forensic analysis a potentially valuable component of digital investigations.

Cyber Truth Times’ investigation into The Telegram Sex-Advertising Economy examined five Telegram datasets associated with commercial sex advertising across Malaysia, Cambodia, Myanmar, Dubai and a Russian-language advertising channel. The datasets contain photographs, videos and promotional media associated with advertisements, alongside text, digital references and other records.

This report examines how forensic methods can be applied to that visual material. It focuses on file preservation, metadata, cryptographic hashes, perceptual similarity, video-frame analysis, reverse-image searching and the limitations of media-based attribution.

The objective is not to identify the people depicted in the advertisements or to make assumptions about their circumstances. Instead, the investigation examines how visual media can be compared, documented and contextualised to determine whether similar or identical files appear in multiple records, whether the media’s provenance can be examined and what further corroboration may be required.

<text weight=semibold inline>Visual similarity is an investigative lead, not proof of identity, ownership, exploitation or trafficking.</text>

1. Why Visual Media Matters in Telegram Investigations

The five datasets examined by Cyber Truth Times contain promotional content in which images and videos form part of the advertising presentation.

Visual media can potentially provide information that is not immediately apparent from the accompanying text.

For example, the same photograph may appear in different advertisements, or a video may be uploaded in a different format while retaining substantially similar visual content.

Such occurrences may raise questions about the provenance and distribution of the media.

However, the existence of visual similarities does not automatically establish that the advertisements were created by the same person, that the people depicted are the same individuals, that the media was produced for a particular advertiser or that the content was distributed with the consent of the people depicted.

Media analysis should therefore focus initially on the files themselves and the records in which they appear.

The investigation may examine:

  • Whether files are exact duplicates.
  • Whether visually similar files appear across different posts.
  • Whether files contain metadata.
  • Whether media appears to have been resized, recompressed or converted.
  • Whether videos contain repeated or overlapping frames.
  • Whether an earlier public appearance can be documented.
  • Whether the media’s context differs between advertisements.
  • Whether the available records support a chronology of appearances.

These are forensic questions about digital files and their documented context, not questions that can be answered reliably through appearance alone.

2. The Five Datasets and Their Visual Evidence Context

The five datasets provide different contexts for examining promotional media.

Malaysia: Promotional photographs and videos

The Kuala Lumpur dataset contains promotional content, including media associated with advertisements and recorded forwarding-related occurrences.

The media can potentially be examined for file-level duplication, visual similarity and contextual differences between records.

The 235 forwarding occurrences discussed in the earlier report are distribution-related records. They do not independently establish that photographs or videos were forwarded 235 times, nor do they establish how many people saw the media.

Media analysis must remain separate from the forwarding analysis unless the source records establish a specific relationship.

Poipet: Multilingual promotional material

The Poipet dataset contains multilingual promotional advertisements and media.

Visual examination may help document whether files within the dataset are exact duplicates or appear visually similar.

However, visual similarity cannot independently establish that the media was captured in Poipet, that the individuals depicted were located there or that the advertisement accurately represents the circumstances described in its text.

Yangon: Media accompanying advertisements

The Yangon dataset contains photographs and videos associated with promotional posts, alongside contact references and other digital records.

The media may provide material for examining file properties, duplication and contextual differences between advertisements.

But the presence of a photograph or video in an advertisement does not establish the identity of the person depicted, the location of capture, the date of capture or the circumstances in which the media was produced.

Dubai: Promotional media and advertising context

The Dubai dataset contains promotional media associated with advertising claims and multiple Telegram destinations.

The media may be examined alongside the text of the advertisements to document how visual material appears within the dataset.

However, media associated with a hotel, transport claim, location description or promotional package does not independently verify that the advertised arrangement occurred.

Images cannot establish hotel involvement, location, transportation, commercial transactions or the identity of individuals without corroborating evidence.

Russian-language dataset: Profile-based media

The Russian-language dataset contains profile-style advertisements and associated media.

This material may be examined for duplicate files, perceptual similarity and metadata, while preserving the distinction between visual appearance and identity.

The media does not independently establish that the profile descriptions are accurate or that the images depict the individuals described in the advertisements.

The five datasets offer different contexts for media analysis, but the presence of visual material across them does not establish a unified media-distribution operation or common ownership.

3. File Metadata: What Photographs and Videos May Reveal

Digital image and video files may contain metadata.

Depending on the file format, device, software and processing history, metadata can include information such as:

  • File format.
  • Image dimensions.
  • Video resolution.
  • Frame rate.
  • Codec information.
  • Date and time fields.
  • Camera or device information.
  • Editing software information.
  • Orientation.
  • GPS-related fields, where retained.
  • File-processing or export information.

Metadata may provide investigative clues.

For example, a file may contain a recorded software field indicating that it was processed through an editing or export application. A video may contain technical information about its encoding or frame rate.

However, metadata should not be treated as automatically authentic.

Metadata can be removed, altered, rewritten, incorrectly interpreted or introduced through a processing workflow.

A timestamp embedded in a file does not necessarily establish when the depicted event occurred.

A GPS field does not independently establish where the photograph was taken.

A software tag does not necessarily establish who edited the file.

A file with missing metadata is not automatically suspicious. Many platforms and messaging applications strip or alter metadata during processing, while ordinary users may also remove metadata intentionally or inadvertently.

Metadata should be treated as a source of leads that require contextual verification.

4. Cryptographic Hashes: Identifying Exact File Matches

Cryptographic hashing is one of the foundational methods for documenting digital files.

A cryptographic hash function generates a fixed-length value from a file’s data. The resulting hash can be used to compare files and determine whether they are byte-for-byte identical.

For example, two files with identical SHA-256 hashes are overwhelmingly likely to contain exactly the same data, subject to the assumptions of the hash function and the integrity of the process used to calculate and record the hashes.

If two files have different cryptographic hashes, they are not byte-for-byte identical.

However, that does not necessarily mean they depict different content.

An image that has been resized, recompressed, cropped, rotated or edited will generally produce a different cryptographic hash.

Likewise, a video that has been re-encoded or placed into a different file container may produce a different file hash even when its visual content remains substantially similar.

Cryptographic hashes are therefore useful for exact-file comparison and evidence-integrity documentation, but they cannot alone identify visually similar media.

In an investigation, the distinction between a hash used to preserve evidence and a hash used to compare content is important.

Integrity hashes document a file’s digital fingerprint for preservation and integrity checks.

Matching hashes may identify exact duplicate files.

Neither establishes the authenticity of the depicted scene, the identity of a person or the truth of an accompanying advertisement.

5. Perceptual Hashing: Comparing Visually Similar Images

Perceptual hashing is designed to identify similarities between media that may not be byte-for-byte identical.

Unlike a cryptographic hash, a perceptual hash attempts to represent visual characteristics of an image in a form that can be compared with other images.

This can be useful when files have undergone routine changes such as resizing or compression.

A perceptual comparison may identify candidate images that appear similar despite differences in file encoding or certain visual transformations.

However, perceptual hashing is not an identity-recognition system.

It compares visual characteristics of media, not the identity or circumstances of the person depicted.

Perceptual hashing also has limitations.

Different images may produce similar perceptual hashes. Similar images may fail to match depending on the transformation, algorithm or threshold used. Cropping, overlays, changes in composition or other modifications can affect the comparison.

A similarity score should not be treated as a probability that two images depict the same person, that one image was copied from another or that the files originated from the same advertiser.

A perceptual match should be treated as a candidate relationship that requires visual review and contextual examination.

Exact matches versus perceptual matches

MethodWhat it can supportWhat it cannot establish
Cryptographic hashExact file-level matchVisual similarity after changes
Perceptual hashCandidate visual similarityIdentity, ownership or source
Visual reviewContextual comparisonDefinitive provenance on its own
Reverse-image searchPossible earlier appearancesComplete publication history
Metadata reviewRecorded technical or contextual fieldsAuthenticity of metadata or scene

The distinction between exact duplication and visual similarity should remain explicit in any published finding.

6. Video Forensics: Examining Moving Images

Videos present additional challenges because they contain sequences of frames, audio, encoding information and potentially multiple processing layers.

A video file may be re-encoded, resized, cropped, edited, screen-recorded or placed into another container.

These transformations can change file-level hashes without necessarily changing the underlying visual content in a meaningful way.

Video analysis may examine:

  • File-container information.
  • Codec and encoding details.
  • Frame rate.
  • Resolution.
  • Duration.
  • Audio and video stream properties.
  • Frame-level similarities.
  • Repeated or overlapping frames.
  • Differences between versions.
  • Embedded metadata.
  • Possible earlier appearances.

Video-frame analysis can help researchers compare visual content across files.

However, a frame extracted from a video is only a representation of a portion of the original moving image.

A single frame may omit important contextual information, including what happened before or after the moment shown.

A video may also contain multiple scenes or edits.

Therefore, frame-level similarities should not be interpreted as proof that two videos are identical, that one was copied from the other or that both depict the same event in the same context.

The examination should consider the full video when practical and document any limitations caused by incomplete files, processing, compression or missing audio.

7. Reverse Image Search and Earlier Appearances

Reverse-image searching may help investigators identify earlier indexed appearances of a photograph or similar visual material.

This can be useful when a photograph appears in an advertisement but the surrounding context is uncertain.

A search may locate:

  • An earlier public appearance.
  • A different caption.
  • A different promotional context.
  • A publication that identifies a photographer or source.
  • Another version of the image.
  • A cropped or edited version.
  • A public webpage containing similar media.

However, reverse-image search is not a comprehensive database of the internet.

Private channels, closed groups, unindexed websites and restricted platforms may not be accessible to search engines.

An image that produces no search results may be new, unindexed, private, altered or simply unavailable to the search engine.

A negative result does not prove that an image is original.

Likewise, the earliest indexed appearance is not necessarily the source.

Investigators should document the search date, the service used, the results observed and the limitations of the search.

Where multiple versions appear, the differences should be examined without assuming that the earliest result represents the authentic original.

8. Establishing a Media-Comparison Framework

A responsible media investigation should distinguish between the observations made about a file and the conclusions drawn from those observations.

For the five Telegram datasets, a media-comparison framework could document:

File identification

Record the original filename, dataset reference and file format.

File integrity

Calculate and document a cryptographic hash where appropriate.

Technical metadata

Record available file properties without treating them as independently verified facts.

Visual comparison

Document whether the files appear identical, substantially similar or materially different.

Perceptual comparison

Record the method used and the limitations of the similarity result.

Context

Document the post, advertisement or dataset in which the media appears.

Chronology

Record the dates associated with the dataset entries, while distinguishing post dates from media capture dates.

Corroboration

Seek independent evidence before attributing media to a person, location, event, advertiser or organisation.

Publication review

Consider whether publication of the image, filename, metadata or other identifying information could harm individuals or facilitate unwanted contact.

The framework should preserve the original media and keep analytical copies separate where practical.

The aim is to produce a transparent record of the analysis rather than a conclusion based solely on a similarity score.

9. What Reused Media Might Suggest and Why Context Matters

If the same photograph or visually similar media appears in multiple advertisements, the observation may raise questions about the advertising material’s provenance.

Possible explanations include:

  • The same promotional media was reused.
  • A third party copied the media.
  • Advertisements were redistributed.
  • Media was reposted without the original context.
  • Files were edited or recompressed.
  • The images were independently obtained from a common source.
  • The records contain a coincidental or misleading visual similarity.

The available media alone may not distinguish between these possibilities.

Reuse does not automatically establish that an advertiser misrepresented an individual, that the media was used without consent, that an individual was trafficked or that multiple channels share an owner.

Such conclusions require evidence specific to the allegation.

The investigation should document the reuse pattern, if confirmed, and identify the questions it raises rather than treating a candidate match as proof of wrongdoing.

10. Privacy, Consent and the Ethics of Visual Investigations

Visual investigations involving commercial-sex advertisements require particular care.

The individuals depicted may be vulnerable, may have limited control over the distribution of their photographs or videos, may be misrepresented by promotional text or may be unaware that their media appears in multiple datasets.

Publication of photographs can expose individuals to harassment, stalking, unwanted contact, reputational damage or other harms.

Investigators should consider whether the public-interest objective can be achieved through redacted images, blurred faces, cropped screenshots, descriptive text or hash-based documentation instead of publishing identifiable media.

The fact that an image is publicly accessible does not automatically justify republishing it.

The fact that an image appears in an advertisement does not establish consent to its use in that advertisement.

The fact that an image is reused does not establish exploitation.

Where the investigation concerns advertising infrastructure rather than the identities of the people depicted, publication should generally prioritise the digital evidence and advertising patterns over unnecessary identification of individuals.

The investigation should avoid facial recognition or identity attribution based solely on appearance and should not publish unverified identities.

11. What the Current Investigation Can and Cannot Establish

The five datasets provide a basis for examining the photographs and videos associated with promotional material.

They contain media that may be examined for exact-file duplication, visual similarity, technical properties and contextual differences between records.

However, this report does not claim completed forensic measurements for individual files unless those measurements have been directly processed, recorded and independently reviewed.

The available dataset summaries do not independently establish:

  • The identities of the people depicted.
  • The source of each photograph or video.
  • The date or location of media capture.
  • Whether the depicted individuals consented to the advertisements.
  • Whether images were manipulated.
  • Whether images were reused without consent.
  • Whether reuse was coordinated.
  • Whether a person was recruited or trafficked.
  • Whether an advertisement represents a real transaction.
  • Whether the five datasets share common ownership.
  • Whether media reuse establishes a criminal organisation.

The absence of a match in a forensic or reverse-image search does not establish originality.

The presence of a match does not establish identity, ownership, consent, exploitation or trafficking.

The investigation must distinguish confirmed file-level observations from similarity-based leads and unresolved questions.

12. Further Investigative Questions

Further media examination could address questions such as:

  • Which files are exact duplicates?
  • Which files appear visually similar despite different hashes?
  • Which media contains metadata?
  • Which metadata fields are preserved?
  • Which fields may have been altered or removed?
  • Are multiple versions of the same video present?
  • Are there earlier indexed appearances of the photographs?
  • Are the earliest indexed appearances independently corroborated?
  • Does the media context change between advertisements?
  • Are there independent sources documenting the media’s provenance?
  • What privacy safeguards are necessary before publication?

These questions provide a basis for additional forensic work without prematurely assigning identity or responsibility.

Conclusion: Images Can Provide Leads, but They Do Not Identify the Truth on Their Own

Photographs and videos can be valuable evidence in digital investigations.

Cryptographic hashes can help identify exact files and document integrity. Perceptual hashes can identify candidate visual similarities. Metadata can provide technical or contextual clues. Video-frame analysis can help compare moving-image content. Reverse-image searches can locate possible earlier appearances.

But none of these methods independently establishes the identity of the person depicted, the authenticity of the scene, consent, exploitation or trafficking.

The five Telegram datasets examined by Cyber Truth Times provide material for a media-forensics investigation into promotional photographs and videos associated with commercial sex advertising.

The objective is to examine the media’s digital properties, document confirmed similarities and differences, preserve the original files and identify questions requiring further corroboration.

Cyber Truth Times will distinguish exact file matches, candidate perceptual matches, contextual observations, analytical interpretations and unresolved questions.

This report forms Part 09 of The Telegram Sex-Advertising Economy, a 12-part investigative series examining five Telegram datasets and the digital advertising patterns visible in the records.

Bibliography

These sources provide technical and methodological context for the article. They are not evidence of the specific media files contained in the five Telegram datasets.

  1. NIST — Digital Evidence Preservation: Considerations for Evidence Handlers — Preservation, hashing and digital evidence handling.
  2. Ofcom — Overview of Perceptual Hashing Technology — Explains cryptographic and perceptual hashes and limitations of similarity matching.
  3. EEAS — OSINT Guidelines: Digital Image Verification — Metadata, reverse-image search and verification considerations.
  4. Media Forensics on Social Media Platforms: A Survey — Research overview of media forensics, provenance and authenticity verification.
  5. Hany Farid — Image Forensics — Academic review of image manipulation and forensic methods.
  6. Multimedia Stream Hashing: A Forensic Method for Content Verification — Research on stream hashing and content verification across multimedia transformations.
  7. UK Government — Digital Imaging and Multimedia Procedure v3.0 — Technical procedure and integrity methods for digital image evidence.

Investigative Disclaimer

This report examines methods for analysing photographs and videos contained in Telegram datasets supplied for journalistic examination. It does not claim that individual media files have been conclusively authenticated, that the people depicted have been identified or that reuse, consent, ownership, exploitation or trafficking has been established. File metadata, cryptographic hashes, perceptual similarity, reverse-image search results and visual comparisons must be interpreted in context and independently corroborated where necessary. Cyber Truth Times distinguishes technical observations, candidate matches, investigative leads, analytical interpretations and unresolved questions throughout this report.

For deeper investigations on such topics, see our Cybercrime Investigations & Exposes.

Leave a Reply

Your email address will not be published. Required fields are marked *