Intel FakeCatcher deepfake detector
Vendor Intel Corporation Official site intel.com/content/www/us/en/newsroom/news... How this card is sourced
Updated · AI Detector Index editors
Intel FakeCatcher "can detect fake videos with a 96% accuracy rate", per Intel's newsroom post of November 14, 2022, which says the company "has productized FakeCatcher" as "the world's first real-time deepfake detector that returns results in milliseconds" (intel.com newsroom, read 10 November 2026). This card is a status record. FakeCatcher is a video detector for faces, and on 10 November 2026 Intel publishes no product page, price, sign-up, download or API for it: the three Intel pages that describe it were last modified in 2023, and the Responsible AI page modified on 5 November 2026 does not name it. The card joins the detector index so the claim has a home with its date, and how this card is sourced explains why a status record carries empty plan and access fields.
What Intel FakeCatcher detects
| Field | Value | Source |
|---|---|---|
| Content types detected | Video | intel.com/content/www/us/en/newsroom/news... |
| Generators covered | No generator list is published. The newsroom post describes deepfakes that use a real video of a person: videos of celebrities doing or saying things they never actually did, The paper evaluates on FaceForensics, FaceForensics++, Celeb-DF and the authors' Deep Fakes Dataset of in-the-wild fake portrait videos (arXiv 1901.02212, abstract) | arxiv.org/abs/1901.02212 |
| Input limits | No public submission route: the 2022 newsroom post says FakeCatcher runs on a server and interfaces through a web-based platform and that the real-time detection platform can run up to 72 different detection streams simultaneously on 3rd Gen Intel Xeon Scalable processors, returning results in milliseconds. The paper analyses faces across video segments (signals from various facial regions, under image distortions, with varying segment durations). No file size, duration, format or URL input is published, and no page accepts an upload. | intel.com/content/www/us/en/newsroom/news... |
| Free tier | None published: no sign-up, trial, price or access form exists on intel.com for FakeCatcher (read 10 November 2026); the newsroom post names potential users (social media platforms, global news organizations, nonprofit organizations) and the infographic lists use cases, with no route to the platform | intel.com/content/www/us/en/newsroom/news... |
| Data retention | No product privacy statement exists. Intel Privacy Notice (Last reviewed: March 27, 2026, Last modified: March 27, 2026), Retention: We retain personal information for the period necessary to: Provide requested Intel Services, as needed to comply with legal obligations [...], as agreed in an individual consent, to resolve disputes, and to otherwise fulfill the purposes, rights and obligations outlined in this Privacy Notice. Retention periods can vary significantly based on the type of information and how it is used. No period is stated for video submitted to FakeCatcher, which has no public submission route. | intel.com/content/www/us/en/privacy/intel... |
| Launch year | 2022 | intel.com/content/www/us/en/newsroom/news... |
| Ownership | Intel Corporation (Nasdaq: INTC), whose newsroom post of November 14, 2022 (Intel Introduces Real-Time Deepfake Detector, updated Nov. 15, 2022) says the company has productized FakeCatcher, a detector designed by Ilke Demir, senior staff research scientist in Intel Labs, in collaboration with Umur Ciftci from the State University of New York at Binghamton; the method is the paper FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals (Ciftci, Demir and Yin, arXiv 1901.02212, submitted 8 January 2019, v3 19 July 2020, IEEE Transactions on Pattern Analysis and Machine Intelligence, document 9141516). Status record: on 10 November 2026 Intel publishes no product page, price, sign-up, download or API for FakeCatcher. The three Intel pages that describe it are the 2022 newsroom post (last modified 3 November 2023), the Trusted Media research blog (last modified 11 May 2023) and a demonstration video page, Trusted Media: Real-time FakeCatcher for Deepfake Detection (last modified 15 March 2023); the Responsible AI page (last modified 5 November 2026) does not name it. The newsroom post describes the platform as running on a server and interfacing through a web-based platform with no public address. The DeepFake-O-Meter v2.0 paper (arXiv 2404.13146, Table I) lists FakeCatcher (Intel) among detection platforms with no website. The Intel Privacy Notice (Last modified: March 27, 2026) governs intel.com. | intel.com/content/www/us/en/newsroom/news... |
FakeCatcher reads blood flow. The newsroom post explains that "most deep learning-based detectors look at raw data to try to find signs of inauthenticity", while "FakeCatcher looks for authentic clues in real videos, by assessing what makes us human, subtle 'blood flow' in the pixels of a video. When our hearts pump blood, our veins change color. These blood flow signals are collected from all over the face and algorithms translate these signals into spatiotemporal maps" (intel.com newsroom, read 10 November 2026). The method is remote photoplethysmography, and the Trusted Media research blog says Intel "has already incorporated deepfake detection technology on the Intel Xeon Scalable Processor" using it (intel.com, Trusted Media, read 10 November 2026). The modality is video of faces; no generator list is published.
The method has a paper, written by its inventors. "FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals" by Umur Ciftci, Ilke Demir and Lijun Yin was submitted to arXiv on 8 January 2019 and revised on 19 July 2020, and the abstract reports "96%, 94.65%, 91.50%, and 91.07% accuracies, on Face Forensics, Face Forensics++, CelebDF, and on our new Deep Fakes Dataset respectively" (arXiv 1901.02212, read 10 November 2026). The index treats the paper as the vendor's account: Demir is "senior staff research scientist in Intel Labs" and Ciftci is at the State University of New York at Binghamton, per the newsroom post, and the datasets are the authors' choices.
The vendor is Intel Corporation. The newsroom post's "Editor's Note" says it "was updated on Nov. 15, 2022, with information to better describe the collaboration behind FakeCatcher", and the page's metadata gives a last modification date of 3 November 2023 (intel.com newsroom, read 10 November 2026). The sibling video cards with a public route are Deepware Scanner, a free web scanner for face manipulation, and Sensity, a forensic platform sold by demo; both state what a buyer can submit, which FakeCatcher does not.
Limits and plans
No submission route exists. The newsroom post says the platform "runs on a server and interfaces through a web-based platform" and that "the real-time detection platform can run up to 72 different detection streams simultaneously on 3rd Gen Intel Xeon Scalable processors" (intel.com newsroom, read 10 November 2026). No address for that platform is published, no page accepts an upload, and no file size, duration, format or URL input is stated. The infographic linked from the post repeats the 72 streams and names four uses, "Content Creation Tools", "Social Media", "Media & Broadcasters" and "AI for Social Good", with no route to any of them (FakeCatcher infographic, PDF, read 10 November 2026).
No price, trial or plan exists, so the plans table on this card is empty, the integrations field is empty and the API field is marked unavailable. The newsroom post lists "several potential use cases": "Social media platforms could leverage the technology", "Global news organizations could use the detector" and "nonprofit organizations could employ the platform to democratize detection of deepfakes for everyone" (intel.com newsroom, read 10 November 2026). The demonstration page "Trusted Media: Real-time FakeCatcher for Deepfake Detection" introduces "Fakecatcher, now optimized for real-time deepfake detection", carries a video and no form, and was last modified on 15 March 2023 (intel.com, demonstration page, read 10 November 2026).
Retention has no product statement. The Intel Privacy Notice, last modified March 27, 2026, says "We retain personal information for the period necessary to: Provide requested Intel Services, as needed to comply with legal obligations [...] and to otherwise fulfill the purposes, rights and obligations outlined in this Privacy Notice" and that "Retention periods can vary significantly based on the type of information and how it is used" (intel.com, Privacy Notice, read 10 November 2026). With no public upload, there is no submitted video for a period to govern, and the retention rank field stays empty.
What the vendor claims
Newsroom post (November 14, 2022): the company has productized FakeCatcher, a technology that can detect fake videos with a 96% (accuracy) rate. Intel's deepfake detection platform is the world's first real-time deepfake detector that returns results in milliseconds. Infographic: 96% FakeCatcher's accuracy for deepfake detection. Paper abstract (arXiv 1901.02212): We evaluate FakeCatcher on several datasets, resulting with 96%, 94.65%, 91.50%, and 91.07% accuracies, on Face Forensics, Face Forensics++, CelebDF, and on our new Deep Fakes Dataset respectively; pairwise separation 99.39% (accuracy). Trusted Media blog: a related source-detection model detects fake videos with 97.29 percent accuracy and detects the source model behind fake videos with 93.39 percent accuracy; a gaze model achieved 92.48 percent accuracy on FaceForensics++, and 99.27 percent accuracy on DeeperForensics.
The one product figure is 96%. The newsroom post's standfirst says the detector "analyzes 'blood flow' in video pixels to return results in milliseconds with 96% accuracy", and the infographic prints "96%" over "FakeCatcher's accuracy for deepfake detection" (FakeCatcher infographic, PDF, read 10 November 2026). Neither page names the dataset behind the 96%, the sample count or the test date. The paper's four figures, 96% on Face Forensics, 94.65% on Face Forensics++, 91.50% on CelebDF and 91.07% on the authors' Deep Fakes Dataset, suggest the product figure is the highest of the four, which the paper itself reports first (arXiv 1901.02212, read 10 November 2026). The card records the 96% as the vendor's claim and quotes the paper's spread beside it.
Related figures on Intel's pages belong to other models. The Trusted Media blog says a source-detection model by the same researchers "detects fake videos with 97.29 percent accuracy and detects the source model behind fake videos with 93.39 percent accuracy", and a gaze-tracking model "achieved a 92.48 percent accuracy on FaceForensics++, and 99.27 percent accuracy on DeeperForensics datasets" (intel.com, Trusted Media, read 10 November 2026). None of these is a FakeCatcher figure, and the card does not merge them. No false positive rate is published for the product or in the paper's abstract.
The caveats are general. Intel's site footer states "No product or component can be absolutely secure. Your costs and results may vary. Performance varies by use, configuration, and other factors" (intel.com newsroom, read 10 November 2026). The paper's abstract says the authors analysed signals "under image distortions, with varying segment durations, from different generators, against unseen datasets", which is the closest the vendor's material comes to a statement on edited or re-encoded video (arXiv 1901.02212, read 10 November 2026). A 96% with no named dataset and a method paper from 2019 are what a reader has; the card keeps them apart.
What independent studies found
No record in the index tested Intel FakeCatcher, read 10 November 2026 in the data file. The one video record, DFDC 2020, scores Kaggle teams rather than named products, and the study count on this card is zero. One outside paper confirms the access picture rather than the accuracy: the University at Buffalo's platform paper, DeepFake-O-Meter v2.0, lists "FakeCatcher (Intel)" in its table of detection platforms with no website address beside it, where the other commercial tools carry one (arXiv 2404.13146, Table I, read 10 November 2026). That platform has its own card in the index.
Until a primary source scores the product, the 96% is Intel's, and the card ranks on nothing. A buyer comparing video detectors has, on this card, a 2022 announcement, a 2019 paper by the inventors and no way to submit a clip.
Where Intel FakeCatcher appears in the index
Rankings
Fits these use cases, questions and comparisons
The card appears on AI detectors by content type with video as its only column, and in studies by modality with zero records against one non-text type, read 10 November 2026 in the data file. It is absent from the study-based rankings, which need an indexed record, from the free tier and retention rankings, which find no allowance and no period, and it carries its 96% into the claim gap field with no study to pair it. The voice and video question lists the cards that claim both media; this card claims video alone. Every detector card on the register shows the same fourteen fields, so a status record with no plan sits beside the detectors that sell one.
Questions about Intel FakeCatcher deepfake detector
Can you use Intel FakeCatcher?
Not through any published route: on 10 November 2026 intel.com carries no product page, price, sign-up, download or API for FakeCatcher, and the 2022 newsroom post describes a server platform with no public address.
How accurate does Intel say FakeCatcher is?
96% (accuracy), per the newsroom post of November 14, 2022 and its infographic, with no dataset named; the inventors' paper reports 96%, 94.65%, 91.50% and 91.07% on four datasets (arXiv 1901.02212), read 10 November 2026.
What does FakeCatcher detect?
Face manipulation in video, by reading "blood flow" signals in the pixels of a face and classifying their spatiotemporal maps, per Intel's newsroom post and Trusted Media blog read 10 November 2026. It does not analyse audio or text.
Who built FakeCatcher?
Ilke Demir of Intel Labs with Umur Ciftci of the State University of New York at Binghamton, per the newsroom post; the method paper with Lijun Yin was submitted to arXiv on 8 January 2019, read 10 November 2026.
Has Intel updated FakeCatcher since 2022?
Intel's three FakeCatcher pages were last modified on 3 November 2023, 11 May 2023 and 15 March 2023 by their page metadata, and the Responsible AI page modified on 5 November 2026 does not mention it, read 10 November 2026.
Sources and revisions
Sources
- intel.com/content/www/us/en/newsroom/news...Read 10 November 2026
- intel.com/content/www/us/en/research/blog...Read 10 November 2026
- intel.com/content/www/us/en/research/trus...Read 10 November 2026
- download.intel.com/newsroom/2022/new-tech...Read 10 November 2026
- arxiv.org/abs/1901.02212Read 10 November 2026
- intel.com/content/www/us/en/privacy/intel...Read 10 November 2026
- intel.com/content/www/us/en/artificial-in...Read 10 November 2026
- arxiv.org/abs/2404.13146Read 10 November 2026
