Frequently Asked Questions
Everything you need to know about PixelTruth AI image analysis.
Technology
How PixelTruth's detection engines work
PixelTruth utilizes a Multi-Engine Consensus approach for maximum identification accuracy. While our Standard (Free) tier relies solely on C2PA Protocol — detecting only images with embedded metadata — our PRO, ULTRA and ENTERPRISE plans unlock the full analytical power of all three independent layers:
C2PA (Coalition for Content Provenance and Authenticity) is an industry standard that embeds a tamper-evident metadata signature — a digital "birth certificate" — directly into an image file at the moment of creation.
When a camera, AI generator or editing tool adds a C2PA signature, PixelTruth can read and verify it to confirm where and how the image was created. Major supporters include Adobe, Microsoft, Google, Sony and many AI platforms.
Our Multi-Engine system currently covers the most widely-used AI image generators, including:
- DALL-E 3 (OpenAI / ChatGPT)
- Midjourney v5 & v6
- Stable Diffusion (including SDXL and community fine-tunes)
- Adobe Firefly
- Flux & Flux (Grok)
- Google Gemini Imagen (incl. Nano Banana)
- Bing Image Creator
The detection database is continuously updated as new models emerge. C2PA-signed content from any compliant tool is also detectable on the Free plan, regardless of the generator used.
Results & Accuracy
Understanding scores, verdicts and limitations
These two terms are often used interchangeably, but they describe fundamentally different things — and the distinction matters for how reliably they can be detected.
Modern AI generators produce images of extraordinary photorealism. A portrait generated by Flux or Midjourney v6 can be indistinguishable from a real photograph to any human viewer — yet our Multi-Engine system is specifically trained to identify the microscopic neural fingerprints left behind by each generator, even after compression, resizing or social-media re-encoding.
- AI-assisted deepfakes — face-swapping or body-replacement performed using AI tools (e.g. FaceSwap, DeepFaceLab). These leave detectable neural artefacts around manipulation boundaries and are identifiable by our Deepfake Anomaly Analysis engine.
- Manually edited deepfakes — images retouched or composited using graphic software such as Photoshop or GIMP, without any AI involvement. If the editing was performed carefully and no AI model was used, the image will contain no AI-generation signature whatsoever. Detection in these cases depends entirely on the presence of pixel-level inconsistencies, cloning artefacts or metadata anomalies — which may be absent in a professionally executed edit.
No system can guarantee 100% certainty — and any tool that claims otherwise should be treated with skepticism. PixelTruth uses the most advanced analysis available on mobile, providing a probability score based on current digital markers rather than a binary yes/no verdict.
A high confidence score (e.g. 88–95%) indicates a very high likelihood of AI generation or significant manipulation. Scores in the mid-range (40–70%) should be treated as inconclusive and may warrant further investigation.
Our Multi-Engine approach covers the tools responsible for over 92.8% of AI-generated images currently circulating online.
A "Suspicious" verdict indicates the detection of structural distortions or neural inconsistencies that suggest potential deepfake manipulation or advanced digital editing — but where a definitive conclusion cannot be reached.
Each scan produces several distinct metrics:
- AI Score — Overall probability that the image was generated by an AI model.
- Fake % — Likelihood of deliberate manipulation or composite editing.
- Anomaly % — Presence of structural or statistical irregularities unusual for authentic photography.
- Per-model scores (PRO/ULTRA/ENTERPRISE) — Breakdown of which specific generator the image most likely originated from, e.g. Midjourney v6: 89%, DALL-E 3: 7%.
The overall verdict (AI Generated / Authentic / Suspicious) is derived from a weighted consensus of all active engine scores.
Yes — partially, and with important caveats. The Deepfake Anomaly Analysis layer (PRO/ULTRA/ENTERPRISE) scans for pixel-level artefacts introduced by digital editing tools including Photoshop, Lightroom, GIMP and Affinity Photo. It also reads EXIF data and camera signatures to verify that the file's claimed origin matches its actual metadata history.
This is one of the most important advantages of our Multi-Engine approach. C2PA metadata is typically stripped entirely when an image is screenshotted, sent through a messaging app, or re-uploaded to a social platform. The Free plan would therefore return no result in most of these cases.
The AI Generation Detection engine (PRO+) works at the pixel and neural-pattern level — it does not rely on metadata at all. Even after heavy JPEG compression, resizing or platform re-encoding, characteristic patterns from AI generators are often still detectable, giving PRO/ULTRA/ENTERPRISE users a significant advantage when analysing images found in the wild.
Privacy & Data
Your images, your device, your control
No. Your image never leaves your device. The app performs an initial analysis pass on-device, and only the extracted scan data (numerical scores and feature vectors — not the image itself) is transmitted to our cloud engines for final processing. These results are never stored, never linked to your identity, and never used for AI model training.
Scan results (score, verdict, timestamp) may optionally be saved to your local device history. You can clear your scan history at any time from the Settings screen.
For full details, see our Privacy Policy.
PixelTruth supports all common image formats available through the Android gallery picker:
- JPEG / JPG — All quality levels supported
- PNG — Including transparent backgrounds
- WebP — Including animated WebP (first frame analysed)
- HEIC / HEIF — Standard iPhone export format
- BMP, GIF — Static images only
For best results, use the highest resolution version of the image available. Minimum recommended resolution is 512 × 512 px. Images smaller than this may return a "Suspicious" or inconclusive verdict.
The scan report remains intact — all the data associated with that scan (verdict, confidence score, metric breakdown, timestamp) is still available in your history. Only the image thumbnail will no longer be displayed.
This is by design. PixelTruth never stores your photos — not on our servers, not in the cloud. The thumbnail you see in your scan history is read directly from your device's local storage at the moment you view it. Once the original file is deleted from your phone, there is simply nothing left to display as a preview.
Yes — an active internet connection is required to perform a scan. PixelTruth's analysis engines run in the cloud, which enables far more powerful and up-to-date detection models than would be possible on a mobile device alone.
An internet connection is required for:
- Image analysis (all plans)
- Account login and subscription management
- Syncing scan history (all plans)
- Accessing the Scan Vault (Enterprise only)
- Downloading PDF Certified Reports (Enterprise only)
Your image never leaves your device — only the extracted scan data is sent over an encrypted connection to our cloud engines. This data is never stored or used for training.
Plans & Features
Free, Pro, Ultra and Enterprise compared
The key differences are scan volume, engine access and result depth:
No. Scan allocations reset on your billing date each month and unused scans do not carry over. If you find you're regularly hitting your limit, consider upgrading to the next plan tier.
PixelTruth supports two payment methods. The cancellation process depends on how you subscribed:
- Open the Google Play Store app
- Tap your profile icon → Payments & subscriptions
- Tap Subscriptions → find PixelTruth → Cancel subscription
- Log in to your account at pixeltruth.app
- Go to Account → Subscription
- Click Cancel plan and confirm
Alternatively, contact us at contact@pixeltruth.app and we will process the cancellation for you.
Cancellation takes effect at the end of the current billing period. You retain full access to your paid plan features until then. We do not process refunds for partial billing periods — see our Terms of Service for the full refund policy.
The Scan Vault is a private repository where you can save individual scan results — including the image thumbnail, full metric breakdown and verdict — for future reference. It is available on the Enterprise plan only.
Vault contents are stored locally on your device and are included in standard Android backups if you have Google backup enabled. Items can be deleted individually or cleared in bulk from Settings.