Getting Started
Welcome to itsreal.media! Follow these steps to start analyzing images for AI-generated content.
-
1
Sign Up
Create your account at app.itsreal.media using your email or social login
-
2
Choose Your Workflow
Select between individual image analysis or multi-image claim risk scoring based on your needs
-
3
Upload & Analyze
Upload your images for AI-powered analysis using the detector or create a claim with multiple images
-
4
Review Results
Get detailed analysis reports with model signals, provenance data, and risk scores for your content
Understanding the Report
The it'sREAL.media Detector analyzes images to determine whether they are authentic photographs or AI-generated synthetic media. The analysis combines multiple independent detection methods to produce a single Synthetic Media Score.
Three layers of analysis
Neural network signals, cryptographic provenance (C2PA), and metadata forensics work together. When they agree, confidence is highest.
| Step | How to read the report |
|---|---|
| 1. Synthetic Media Score | The top-level score from 0% authentic to 100% synthetic. The gradient bar gives you an instant visual read. |
| 2. Summary panel | Quick overview: verdict, Frame:Origin and Frame:Edit results, C2PA status, and metadata summary. |
| 3. Detail panels | Expand any section below the summary for full signal breakdowns, raw model outputs, and metadata dumps. |
Upload an Image
Upload any image from the detector page in the dashboard.
Image requirements
- Formats
- JPEG · PNG · WebP · AVIF
- Max size
- 10 MB
- Min dimension
- 256 px on the shortest side
Images below the minimum dimension are rejected.
The analysis pipeline runs all detection modules and typically returns results in a few seconds depending on image resolution.
Synthetic Media Score
The score represents the overall likelihood that the image is synthetic. It combines all available evidence: model predictions, C2PA provenance, and metadata analysis.
The gradient bar runs from green for authentic through orange for uncertain to red for synthetic. A white indicator marks the exact position.
C2PA and metadata override
When a cryptographically valid C2PA manifest is present, the score is overridden entirely: strongly synthetic if AI-generated content is declared, strongly authentic if verified real. When AI-generated metadata is detected without valid C2PA, the score is pushed toward synthetic. All model signals and metadata details are still displayed in the report.
Image Data
Technical properties extracted from the image file: format, pixel dimensions, megapixel count, aspect ratio, and compression level.
| Property | Description |
|---|---|
| Format | Image file format such as JPEG, PNG, WebP, or AVIF. |
| Dimensions | Width and height in pixels. |
| Megapixels | Total pixel count in megapixels. |
| Compression | Estimated compression level based on bits-per-pixel when file size is available. |
Frame:Origin 1.2
Determines whether an image was entirely generated by an AI system. It runs five independent detection signals, each analyzing a different aspect of the image.
| Signal | What it analyzes |
|---|---|
| Surface Analysis | Pixel-level noise patterns on flat surfaces. AI generators leave characteristic signatures that differ from real camera sensor noise. |
| Edges Analysis | Edge transitions and boundary artifacts. AI-generated edges show subtle regularities not found in optical captures. |
| Frequency | Frequency components via wavelet transforms. AI generators produce distinctive frequency-domain fingerprints. |
| RGB Entropy | Entropy distribution across color channels. Real photographs have natural entropy patterns that AI struggles to replicate. |
| Generalized Model | Broad-spectrum detector trained on raw pixel data without preprocessing, designed to catch novel generation techniques. |
An ensemble model combines all five signals into a single confidence score.
Frame:Edit 0.8
Detects whether a real photograph has been partially modified using AI tools such as inpainting, face swaps, object removal, or background replacement.
| Signal | What it analyzes |
|---|---|
| Entropy RGB | Entropy inconsistencies between modified and unmodified regions across color channels. |
| Surfaces | Texture discontinuities where AI-edited regions meet original content. |
| Generalized Model | Broad detector for partial manipulation patterns without domain-specific filtering. |
| Frequency | Wavelet analysis tuned for localized frequency artifacts left by editing tools. |
| NLE Maps | Non-linear enhancement maps that highlight editing artifacts invisible to the naked eye. |
The edit verdict shows Authentic or Image Modified. This verdict is produced by a dedicated ensemble model that aggregates the five edit signals above.
C2PA Provenance
C2PA (Coalition for Content Provenance and Authenticity) is a cryptographic standard that embeds signed certificates into image files, creating a tamper-evident chain of provenance.
| Status | Meaning |
|---|---|
| Verified Origin | Captured by a certified device and unmodified since signing. |
| Transparent Edit | Edited, but every edit is documented in the signed manifest. |
| Altered | A C2PA manifest exists but the signature is broken, which implies tampering. |
| Unverified | A C2PA manifest exists, but authenticity could not be established because of trust, compliance, or validator limitations. |
| No C2PA | No provenance manifest was found, so other methods must be used. |
C2PA details include the signer, claim generator, actions performed, and source ingredients when available.
Metadata Analysis
Scans embedded metadata fields such as IPTC, XMP, and EXIF for traces that reveal how the image was created.
| Result | What it means |
|---|---|
| Synthetic Media Clues | Metadata contains AI tool identifiers such as IPTC 2025.1 fields or legacy XMP traces like CreatorTool or DigitalSourceType. |
| Standard metadata | EXIF, XMP, or JPEG structure data may be listed for reference; it is not used as an authenticity signal. |
| None | No meaningful clues were found. Metadata may have been stripped during export or upload. |
Raw Metadata
Complete, unprocessed metadata dumps from the image file. Especially useful for forensic investigation and debugging.
| Source | Contents |
|---|---|
| EXIF | Camera and capture metadata: make, model, lens, exposure, GPS, timestamps. |
| XMP | Editing tools, creation dates, and AI properties. |
| IPTC | Editorial metadata such as photographer, caption, keywords, copyright, and location. |
| PNG Text | Text chunks often used by AI tools to record generation parameters. |
| ICC Profile | Color space and display profile information. |
| JPEG Info | JFIF version, subsampling, and quality estimates. |
Claim Risk Score
A claim bundles one or more images with a description and produces a single risk score that represents the likelihood that any of the submitted images contain synthetic media. Scores range from 0.0 (authentic) to 99.9 (synthetic).
| Verdict | Score range | Meaning |
|---|---|---|
| LOW RISK | 0 – 32.9 | All images appear authentic. No significant synthetic signals detected. |
| MEDIUM RISK | 33.0 – 65.9 | Mixed or uncertain signals. Manual review is recommended. |
| HIGH RISK | 66.0 – 99.9 | Strong indicators of AI-generated or modified content. |
Each image in the claim is scored independently through a three-layer priority system, then the claim-level score is determined by worst-case aggregation (the highest per-image score wins).
Worst-case aggregation
A single high-risk image is enough to elevate the entire claim. If a claim contains four authentic photos and one AI-generated image, the claim score reflects the synthetic image.
| Layer | Source | How it works |
|---|---|---|
| A. C2PA Provenance | Cryptographic manifest | Highest priority. If a verified C2PA manifest indicates AI-generated content (via IPTC digital source type or known AI generator in the signer/software agent), the score is set to 99.0. If it confirms a real camera origin, the score is set to 1.0. Altered, unverified, or missing C2PA falls through to the next layer. |
| B. Model predictions | Frame:Origin / Frame:Edit ensemble | The final ensemble prediction (or the Frame:Origin binary classifier as fallback) provides a label and confidence. The confidence is converted to a synthetic probability: real at 95% confidence maps to 5.0, while ai-generated at 90% confidence maps to 90.0. |
| C. Metadata adjustment | IPTC, XMP, EXIF clues | Applied on top of the model score. When AI metadata is detected (XMP AI tags, EXIF tool signatures), the score is pushed strongly toward synthetic (weight 0.80). When no AI metadata is found, the model score stands unchanged. Non-AI metadata (EXIF, JPEG structure) is shown for transparency but does not adjust the score. |
Score breakdown
Each claim report includes a detailed Score Breakdown panel showing the contribution of each layer (C2PA, model predictions, metadata) for every image. Expand the breakdown to understand exactly why a score was assigned.
Claims can be created from the New Claim page in the dashboard. Up to five images can be attached per claim. All images are analyzed in parallel and the risk score is finalized once every image has been processed.
API Overview
The it'sREAL.media API provides programmatic access to image analysis. Authenticate with an API key using the Authorization header. Full interactive docs are also available on Swagger UI.
https://api.itsreal.media
| Header | Format |
|---|---|
| API Key | Authorization: Token YOUR_API_KEY |
Generate an API key from the API Key page. Each user can have one active key at a time.
The analyze endpoint is the only public API endpoint. All other endpoints are used internally by the frontend application.
Monthly quota
Each account has a monthly image analysis quota (default 1,000). Every response includes X-QuotaLimit-Limit and X-QuotaLimit-Remaining headers. When the quota is exceeded, the API returns 402 Payment Required. Check your current usage on the Account page.
Analyze Endpoint
The primary endpoint for AI image detection. Upload an image and get instant analysis results.
Accepted files
- Formats
- JPEG · PNG · WebP · AVIF
- Max size
- 5 MB
- Min dimension
- 256 px on the shortest side
Images below the minimum dimension are rejected with a 422 Unprocessable Entity error.
Upload and analyze an image to determine if it's real or AI-generated. Supports JPEG, PNG, WebP, and AVIF formats up to 5 MB.
| Parameter | Type | Description |
|---|---|---|
| file required | binary | Image file to analyze (JPEG, PNG, WebP, AVIF, max 5 MB) |
| Status | Description |
|---|---|
| 400 | Invalid file type, file too large, or missing file |
| 401 | Unauthorized — invalid or missing token |
| 402 | Payment Required — monthly quota exceeded |
| 422 | Unprocessable Entity — image below minimum dimensions (256 px) |
| 504 | Gateway Timeout — analysis timed out |
Quick Example
Troubleshooting & FAQ
Common issues and solutions to help you get the most out of itsreal.media.
Check the supported formats and size limits in the Upload an Image section above. Common causes include exceeding the file size limit, unsupported formats, or images below the minimum dimension. Try refreshing the page and uploading again.
If you're experiencing issues not covered here, please contact our support team at support@itsreal.media or use the contact form on our website. We typically respond within 24 hours.