# Plurall AI: Frequently asked questions

Source: https://plurall.ai/faq/
Last updated: 2026-11-09

Short answers about what Plurall AI is, how it works and how to evaluate it. Figures link to the page they come from.

## What is Plurall AI?

Plurall AI, spelled with two l's, is a New York company founded in 2024 that builds deepfake detection, fraud defense and proof-of-humanhood verification for trust, safety, security and compliance teams. Its models read forensic signals across image, video, audio and identity and return a verdict with the evidence behind it, so a reviewer can act on it rather than guess.

More: [Mission](https://plurall.ai/mission/)

## What types of media does Plurall AI detect?

Plurall AI detects manipulated images, video and live-camera feeds today, with audio detection marked coming soon. Detection combines image, video, audio, metadata, spectral, temporal and web-intelligence signals so that a verdict is backed by more than one line of evidence.

More: [Engineering](https://plurall.ai/engineering/)

## What models does Plurall AI use?

Four in-house models feed one evidence layer: GaussMass for image detection, Influx for video detection with temporal analysis, Sage for live-camera liveness and injection detection, and Timbre for calibrated synthetic-speech detection (coming soon). The image model is named for Carl Friedrich Gauss.

More: [The verification toolbox](https://plurall.ai/engineering/)

## How fast is Plurall AI?

The GaussMass image model runs in 14.07 ms on a mobile NPU, 37.22 ms on GPU and 74.13 ms on CPU, with a 5.27x NPU speed-up over CPU. It is 8 to 15 million parameters and 90.24 MB, compared with 80 to 300 million parameters for ViT-Base and ViT-Large detectors, which is why it can run on-device as well as in the cloud. Timbre scores audio in under 30 ms.

More: [Architecture figures](https://plurall.ai/#hiw-heading)

## How accurate is Plurall AI?

Plurall AI reports accuracy by operational slice on disjoint splits, never as one pooled number, because pooled numbers hide weak spots on specific generators and capture conditions. On the current V1.2 slices, synthetic recall moved by 0.64 percentage points after benign edits (93.89% edited-synthetic recall) and a 512-image real-portrait guard held at 512 authentic. On the V1.3 research track, a controlled portrait benchmark reached 99.22% real recall and a 24-image real-only stress fixture produced no false positives. Every reported rate carries a 95% Wilson interval, and borderline items are routed to review rather than guessed.

More: [Validation](https://plurall.ai/#hiw-heading)

## Does Plurall AI run on-device or only in the cloud?

Both. The models are small enough to run on a mobile NPU, which supports on-device and edge deployments where media cannot leave the device. Cloud and API deployments are available for batch review and workflow integration.

## What happens when a new deepfake generator appears?

Plurall AI retrains end to end and ships a re-scored release within 24 to 48 hours. A run converges in 50 to 100 epochs, four to eight hours on one GPU, so retraining is routine rather than a project.

More: [Retraining](https://plurall.ai/#hiw-heading)

## Who uses Plurall AI?

Plurall AI is built for trust and safety, fraud, security and compliance teams at digital platforms, marketplaces, financial services and other organizations where content authenticity matters. Small and mid-sized businesses can access it through managed service providers such as eMazzanti Technologies.

More: [Partner with us](https://plurall.ai/partner/)

## How does Plurall AI compare to Reality Defender, Sensity AI and Hive?

Reality Defender is positioned as a broad multimodal enterprise platform, Sensity AI as a forensic and threat-intelligence tool, and Hive as a content-moderation API for platforms at scale. Plurall AI differs in two ways: its models are small enough to run on-device with low latency, and every result is delivered as explainable evidence rather than a single score.

More: [Best deepfake detection companies in 2026](https://plurall.ai/compare/best-deepfake-detection-companies/)

## Does Plurall AI offer an API?

Yes. Plurall AI provides a bearer-authenticated REST API with media scan workflows, JSON responses, webhooks for asynchronous results and cursor-based pagination, so teams can integrate detection into their own review tools, case management and onboarding flows.

More: [Developer docs](https://plurall.tech/docs)

## What is proof of humanhood?

Proof of humanhood is Plurall AI's verification surface for confirming that a real, present person is on the other side of the camera. It provides verification workflows, liveness status and audit trails, so a team can show not only that a check was passed but what evidence supported it.

More: [Proof of humanhood](https://plurall.ai/proof-of-humanhood/)

## How much does Plurall AI cost?

Plurall AI offers a free trial and pilot programs; enterprise pricing depends on volume, modalities and deployment model. Start a free trial or ask for a pilot through the partner page.

More: [Try it for free](https://plurall.tech/signup/start)

## Is there a free trial?

Yes. You can try Plurall AI for free, or start with a scoped pilot for your deployment.

More: [Try it for free](https://plurall.tech/signup/start)

## Is there a browser extension?

Yes. The Plurall AI Deepfake Defender extension for Chrome scans images and video in the browser and flags signs of manipulation.

More: [Chrome Web Store](https://chromewebstore.google.com/detail/djdpgmejbeoongebfkobaipgpemphofg)

## Where is Plurall AI based and who founded it?

Plurall AI is based in New York and was founded in 2024 by Sandra Cai. The company was selected for TechCrunch Disrupt 2025 Startup Battlefield.

More: [TechCrunch Startup Battlefield](https://techcrunch.com/startup-battlefield/company/plurall-ai/)

## What laws apply to deepfakes?

Regulation is moving quickly in the US, EU and UK. Plurall AI maintains a summary of synthetic media law across the three jurisdictions in its Blog and Press section.

More: [Blog and Press](https://plurall.ai/notes/)
