Reality Defender deepfake detector

Vendor Reality Defender, Inc. Official site realitydefender.com How this card is sourced

AIDI-D-0022

Updated ·

Reality Defender accepts video files of up to 250 MB and 30 minutes through its SDK and returns a score that the vendor reads as Authentic from 0 to 44, Suspicious from 45 to 55 and Manipulated from 56 to 100, according to the Reality Defender developer docs (SDK quickstart and media detail reference) read 9 November 2026. The vendor, Reality Defender, Inc., was founded in 2021 and sells to enterprises and governments, per the company page read the same day. Reality Defender is the first video-first card in the detector index; how this card is sourced explains which fields a deepfake vendor fills and which it leaves empty.

What Reality Defender detects

The Reality Defender register entry: field, value, source and read date
FieldValueSource
Content types detectedVideo, image, voice, textrealitydefender.com/faq
Generators coveredNot named by generator. The vendor lists manipulation types: face swaps, generated video, altered faces, synthetic speech on the audio track, and AI-agent callersrealitydefender.com/deepfake-video-detection
Input limitsSDK local files: images .jpg, .jpeg, .png, .gif, .webp up to 50MB; audio .mp3, .wav, .m4a, .aac, .ogg, .flac, .alac up to 20MB; video .mp4, .mov up to 250 MB and a maximum duration of 30 minutes; text .txt up to 900KB. Social media URLs from Facebook, Instagram, Twitter/X, YouTube, TikTok and Threads. The Free plan supports image uploads only.docs.realitydefender.com/sdks/quickstart
Free tierFree API plan: $0/month, 50 scans per month, up to 3 seats, image analysis only, API keys and an analytics dashboard; audio and video need a paid planrealitydefender.com/api
APIYes, docs.realitydefender.com/api-reference/qu...docs.realitydefender.com/api-reference/qu...
IntegrationsSDKs for Python, TypeScript, Go, Rust and Java, REST API with webhooks, RealMeeting for Microsoft Teams (live audio and video), RealCall in the IVR and live contact-centre calls, RealScan browser review, Social media URL submission (Facebook, Instagram, Twitter/X, YouTube, TikTok, Threads)realitydefender.com/api
Data retentionWebsite privacy policy (last updated August 10, 2026): We will not retain your personal information for longer than is reasonably necessary to carry out the applicable, identified purposes; the factors considered are listed and no fixed period is stated. Photos, images, videos, audio files and voice recordings are among the information collected. The FAQ states a range of US, UK and EU certifications.realitydefender.com/t-c/privacy-policy
Launch year2021realitydefender.com/company
OwnershipReality Defender, Inc. (privacy policy, last updated August 10, 2026). Founded in 2021, the company grew out of a nonprofit research initiative focused on AI-generated deception (company page). Products: RealScan, RealAPI, RealCall, RealMeeting.realitydefender.com/t-c/privacy-policy

Reality Defender detects AI-generated or manipulated media across four modalities. The FAQ describes "a deepfake detection platform for enterprises, governments, and platforms to detect AI-generated content and manipulations across audio, video, image, and text files" (FAQ, read 9 November 2026). The video page names what the detector looks for: face swaps, generated video, altered faces and synthetic speech on the audio track; the contact-centre products add AI-agent callers (video detection page, read 9 November 2026). No generator is named, so the generators field records manipulation types instead.

Four products share one engine: RealAPI for files submitted from an application, RealScan for review in a browser, RealCall for synthetic speech in the IVR and live calls, and RealMeeting for live audio and video in Microsoft Teams (realitydefender.com, read 9 November 2026). SDKs exist for Python, TypeScript, Go, Rust and Java, and the API accepts social media URLs from Facebook, Instagram, Twitter/X, YouTube, TikTok and Threads as well as uploads (API quickstart, read 9 November 2026). The FAQ adds that the models are "a combination of in-house models or proprietary models" and that the platform holds a range of US, UK and EU certifications.

Reality Defender is one of three cards covering video: Hive states image, video, voice and music, and Pangram states text, image and video. The image-only card closest in purpose is Was It AI.

Limits and plans

Reality Defender plans with price, limits and the date each was checked
PlanPriceLimitsChecked on
Free$0/month50 scans per month, up to 3 seats, image analysis, access to API keys, analytics dashboard. For trialing the API.
Business$399, Annual Billing (as shown)1,000 scans per month, 1 seat, image, audio and video analysis, results plus supporting detail, analytics dashboard, unlimited API keys and SaaS web platform, self-serve onboarding and chat support.
EnterpriseCustom pricing: custom scan volume, unlimited seats, image, audio and video analysis, unlimited API keys, in-app results and analytics dashboard, dedicated support. RealCall and RealMeeting are sold through sales.

The SDK docs give the file limits: images in .jpg, .jpeg, .png, .gif and .webp up to 50MB; audio in .mp3, .wav, .m4a, .aac, .ogg, .flac and .alac up to 20MB; video in .mp4 and .mov up to 250 MB with a maximum duration of 30 minutes; text in .txt up to 900KB (docs.realitydefender.com/sdks/quickstart, read 9 November 2026). The same page states: "The Free plan supports image uploads only. Video, audio, and text uploads require a paid plan." Larger files take longer and the API must be polled or a webhook set.

Prices are from the API page, read 9 November 2026. The page shows a dollar sign without a currency code.

The Business row is printed as "Annual Billing $399" with no monthly figure beside it, so the card reproduces the label as shown. On retention, the website privacy policy of August 10, 2026 says the company "will not retain your personal information for longer than is reasonably necessary to carry out the applicable, identified purposes", lists the factors it weighs, states no fixed period, and names photos, images, videos, audio files and voice recordings among the information collected (privacy policy, read 9 November 2026).

What the vendor claims

We use an inference system that grades each content with a 1-99% probability rating -- all without the need for the ground truth. (No numeric accuracy or false positive figure published; the home page cites a Gartner Emerging Market Quadrant for Deepfake Detection Market Shaper designation of 25 June 2026 and a bank case study of 2.4M+ calls analyzed in 2025 with about 1,010 synthetic voice calls detected.)
realitydefender.com/faq, read

Reality Defender publishes no accuracy rate and no false positive rate. The FAQ states the scoring method: "we use an inference system that grades each content with a 1-99% probability rating -- all without the need for the ground truth" (realitydefender.com/faq, read 9 November 2026). The home page cites a Gartner Emerging Market Quadrant for Deepfake Detection that named the company a Market Shaper on 25 June 2026, and a bank case study of 2.4M+ customer calls analysed in 2025 with about 1,010 synthetic voice calls detected; neither is a measured error rate, and the card files them as vendor statements.

The vendor's own limits are stated plainly on the video page: "A media finding addresses signs of manipulation. Identity verification remains with the controls responsible for establishing who the person is", and "a media assessment does not establish whether a recorded claim is true" (realitydefender.com/deepfake-video-detection, read 9 November 2026). The FAQ answers "Can anyone use Reality Defender?" with "No", describing the product as meant for "the largest entities and governments".

A deepfake score is a probability that the pixels or the waveform were manipulated, and the vendor says in its own words that the score does not establish who is on screen or whether what they said is true.

What independent studies found

No study record in the index has tested Reality Defender, and none of the nine records in the studies index evaluated a video, audio or image detector; every indexed sample is text. The detectors with a record are ranked on AI detectors ranked by independent testing, all of them text tools.

Deepfake benchmarks such as the DFDC and FaceForensics++ are on the candidate list and will be read in the primary source before any figure from them appears here. Until then the card carries the vendor's limits, prices and statements and no measured rate.

Where Reality Defender appears in the index

No ranking lists Reality Defender. The card has no study, no published false positive rate, a free allowance counted in scans rather than words, and no language list, so none of the four ranked fields carries a value (detectors.json, read 9 November 2026).

Reality Defender appears on AI detectors by content type, where video, image, voice and text are columns, and that view is the audience page that fits an enterprise deepfake vendor. The index of every AI detector sets the card beside the text detectors on the same fourteen fields, which is how a reader sees that the free plan here is 50 image scans a month while the text cards count words.

Questions about Reality Defender deepfake detector

Does Reality Defender have a free plan?

Yes, for the API: $0 a month for 50 scans, up to 3 seats, image analysis only, per the API page read 9 November 2026. Audio and video need the Business or Enterprise plan.

How much does Reality Defender cost?

The API page shows Business at "$399" under "Annual Billing" for 1,000 scans a month with image, audio and video analysis, and Enterprise at custom pricing, read 9 November 2026. RealCall and RealMeeting are sold through sales.

What file types and sizes does Reality Defender accept?

Images up to 50MB, audio up to 20MB, video up to 250 MB and 30 minutes in .mp4 or .mov, and text up to 900KB, per the SDK docs read 9 November 2026. Social media URLs from six platforms can be submitted instead of a file.

How does Reality Defender score a file?

With a score from 1 to 99: Authentic 0 to 44, Suspicious 45 to 55 and Manipulated 56 to 100, per the API media detail docs read 9 November 2026. The vendor says the score addresses signs of manipulation, not identity or truth.

Sources and revisions

Revision log

  1. Card created

Corrections log

Related reading