Unmask AI Content

Instantly detect AI-generated text and images with open-source, privacy-first analysis. Free to use.

Is it AI?

0/60,000

Sign in to DeepFlag

Continue with Google or use a one-time code sent to your email.

or

Your scan limit is exhausted

Try again later, or send a request below.

Request a higher limit

Tell us what you're using DeepFlag for and we'll review raising it.

Sample result — not a real scan
42%
Confidence
The signals were too close to call confidently either way, so this is the model's confidence in the more likely side (human or AI) — not a measure of "how mixed" the content is. Check the per-metric breakdown below for more detail.
Mixed / Uncertain
perplexity

34.21

How predictable the text is to a language model. Human writing tends to be less predictable (higher perplexity); AI text often reads smoother and more predictable (lower perplexity).
burstiness

0.47

Variation in sentence length and structure. Human writing tends to 'burst' between short and long sentences; AI text is often more uniform.
entropy

4.82

Randomness in word choice. Lower entropy can indicate more repetitive, predictable — and often more AI-like — phrasing.
classifier score

0.61

Score from a RoBERTa-based classifier trained to tell AI-generated text apart from human writing (0 = human-like, 1 = AI-like).
binoculars score

1.08

Cross-perplexity ratio between two language models (the Binoculars method) — a strong zero-shot signal for detecting AI-generated text.
language

en

The detected language of the submitted text.
language confidence

0.98

Confidence in the detected language. Our models are English-only, so low confidence or a non-English language means the other scores may be less reliable.

I've been using this feature for a few weeks now and honestly it's been a mixed bag for me personally.The implementation leverages a novel multi-stage pipeline that seamlessly integrates natural language processing capabilities, enabling users to derive actionable insights from unstructured textual data with unprecedented efficiency.

Likely human
Uncertain
Likely AI

How It Works

Multiple detection signals combined into a single confidence score. No black boxes — every metric is transparent.

Text Detection

Analyze text using perplexity scoring, burstiness analysis, entropy measurement, and a RoBERTa-based ML classifier — all combined into a weighted ensemble.

Image Detection

Detect AI-generated images through FFT frequency analysis, EXIF metadata inspection, and a dedicated image classifier model — no detail goes unnoticed.

Developer API

RESTful API with API key authentication, full OpenAPI documentation, and rich JSON responses including confidence scores, analysis metrics, and flagged segments.

Why AI Detection Matters

AI-generated content is accelerating across the web. As large language models produce text indistinguishable from human writing and image generators create photorealistic visuals at scale, the ability to verify content authenticity has become a critical tool for publishers, educators, and platforms.

50%+

of Google searches now trigger an AI Overview response

527%

growth in AI-referred web sessions between January and May 2025

900M

weekly active ChatGPT users generating and consuming AI content

3×

more likely — AI text scores lower on perplexity and burstiness than human writing

Frequently Asked Questions

Everything you need to know about AI content detection with DeepFlag.

What is DeepFlag.ai?
DeepFlag.ai is a free, open-source AI content detector that analyzes text and images to determine whether they were generated by an AI. It combines multiple detection signals into a single confidence score and shows a transparent per-metric breakdown — no black boxes.
How does AI text detection work?
Text detection uses a weighted ensemble of four signals: perplexity scoring (AI text tends to choose highly predictable word sequences), burstiness analysis (AI text has low sentence-length variation), entropy measurement (AI text shows less randomness in word selection), and a RoBERTa-based ML classifier trained on human and AI-written samples. Each signal contributes to a final confidence score from 0 to 100%.
How does AI image detection work?
Image detection runs three checks: FFT (Fast Fourier Transform) frequency analysis to identify unnatural patterns in pixel distribution that AI generators leave behind, EXIF metadata inspection to flag missing or inconsistent camera data typical of synthetic images, and a dedicated image classifier model trained on real versus AI-generated images. All three signals are combined into one confidence score.
Is my text or image stored or used for training?
No. All content you submit is processed in real time and immediately discarded, regardless of whether you're signed in. DeepFlag.ai does not store, log, or use your submissions for model training.
Is DeepFlag.ai free to use?
Yes. The web detector is completely free — sign in with Google or a one-time email code to view your results, no payment required. A Developer API is also available for teams that need to integrate AI detection into their own applications or content moderation pipelines.
Does DeepFlag.ai have an API?
Yes. The DeepFlag API is a RESTful service with API key authentication and full OpenAPI documentation. Responses include a confidence score, per-metric breakdowns, and flagged segments — making it easy to build automated content moderation workflows.