Tell what's real from what a machine made.
Ashes of the Phoenix, Inc. is a software development company based in Delaware, USA. We are building VeriLens AI, a privacy-first mobile app prototype that examines an image across independent forensic systems and returns an explainable result about synthetic generation or digital alteration.
A photograph stopped being evidence of anything.
Courts, insurers, newsrooms, HR departments, dating platforms and marketplaces all built processes on an assumption that quietly expired: that a photograph of a thing is evidence the thing happened. The tools to break that assumption are now free and take seconds. The tools to check it are not.
Claims and fraud
An insurance claim, a damage report, a marketplace listing. A generated image costs nothing to produce and is expensive to disprove after the payout.
Identity and reputation
Synthetic imagery of real people is trivial to make and, without a technical record, near-impossible for the subject to refute credibly.
Record and provenance
Newsrooms and archives need a defensible answer about an image's origin, produced consistently and repeatably rather than by an expert's intuition.
Watch an image get taken apart.
Pick a sample below. Each forensic dimension runs independently and reports its own reading; the fusion layer weighs them into one verdict and says how strongly it holds.
Illustrative simulation with pre-set outcomes. It demonstrates the analysis flow and the shape of a result — it does not run the engine in your browser.
Private image forensics, designed for the device in your hand.
The VeriLens AI mobile prototype is in development for Android and iOS. Its target workflow keeps analysis and history on the device, works without sending the selected image to a remote analysis service, and explains what each forensic system observed. It is a prototype, not yet a public release.
Local-first analysis
Choose an image and run the forensic pipeline on the device. The architecture is designed so the original image does not need to be uploaded for analysis.
Three engine components
DAE examines alteration evidence, OX1 estimates whether an image is synthetic, and OX2 estimates a generator family only when confidence supports naming one.
Evidence, uncertainty and abstention
The result keeps each reading visible, represents uncertainty explicitly, and can return inconclusive or withhold attribution instead of forcing a confident answer.
Readable results and local history
The prototype turns technical measurements into careful language and provides a history view so previous analyses can be revisited on the same device.
Four dimensions, deliberately independent.
Any single detector can be defeated by a technique aimed at it. Four detectors reading different physical properties of the same file are much harder to satisfy at once — and when they disagree, that disagreement is itself information the fusion layer uses.
Metadata schema
EXIF structure, encoder fingerprints and the internal consistency of what the file claims about itself. Cheap to strip, which is why it is never trusted alone.
Optical sensor noise
Real cameras leave a characteristic sensor-noise pattern across the frame. Generated imagery has to imitate a physical process it never went through.
Colour & illuminant
Whether light in the scene behaves as light does — consistent direction, colour temperature and falloff across every surface in the frame.
Learned detector
A trained model reading image content directly, benchmarked against generators it has never seen so the score means generalisation rather than memorisation.
Dempster-Shafer evidential fusion
The dimensions do not vote. Each contributes a belief mass, including mass assigned to "unknown" — which is what lets the system return Inconclusive instead of guessing. A result that does not know is more useful, and far more defensible, than a confident result that is wrong.
From a verdict to an infrastructure layer.
The engine and mobile prototype exist. What follows is validation, packaging and distribution.
Android and iOS prototype
A local-first mobile workflow integrates the DAE, OX1 and OX2 engine components, explainable results, abstention and on-device analysis history.
Native packaging and platform validation
Complete protected model packaging, calibrate the full release configuration and verify consistent behavior across supported Android and iOS devices.
Professional workstation & developer API
Bring the same engine to case-oriented desktop workflows and authenticated platform integrations once funding allows development to resume beyond the mobile prototype.
Provenance & C2PA
Reading signed provenance manifests where they exist and falling back to forensics where they do not — the two approaches are complements, not competitors.
Four ways the same engine is built to earn.
One forensic pipeline, sold into four different buying motions. Each reaches a different budget holder, and each one raises the value of the others: every image analysed anywhere improves the calibration everything else is sold on. These are the planned routes to revenue — VeriLens is early-stage and they are not yet established.
Professional subscription
Self-serve monthly and annual plans for journalists, investigators and analysts. Low friction to adopt, no sales cycle, and the entry point that seeds every other motion.
Platform API
Per-call screening for marketplaces, dating platforms and social products that need to check uploads at the point of submission. Revenue scales with the customer's own volume rather than with seat count.
Certified reports
A downloadable, verdict-bearing report suitable for attaching to a claim, a case file or an editorial record. Priced per document, bought by teams that need the paper trail rather than the dashboard.
Enterprise & licensing
Annual agreements for insurers, newsrooms and public bodies, including deployment inside the customer's own perimeter where the data cannot leave it. The highest-value motion, reached from the evidence the first three produce.
The pipeline runs on ordinary compute — no accelerator fleet to rent and no per-token bill to a model vendor. Gross margin is a function of commodity CPU seconds, and it improves with scale rather than tracking someone else's price list.
Volume drives the API, headcount drives seats, and evidentiary need drives reports. A customer can grow along any of the three without a replatform or a renegotiation.
Every advance in generative models widens the gap this closes. The market does not need to be created or educated — it is being expanded, continuously, by the industry we exist to answer.
Payments are processed by Stripe. Subscriptions run on Stripe Checkout and the Stripe Billing Portal, both hosted by Stripe — card number, expiry and CVC are entered on Stripe's own pages and never reach a VeriLens server. Subscription state changes only on a Stripe-signed webhook, so billing status cannot be forged from the client.
We're opening the round.
Ashes of the Phoenix, Inc. is a Delaware C corporation and the company behind the VeriLens AI project. The company is raising its first round through a SAFE — a Simple Agreement for Future Equity — offered under Rule 506(c) of Regulation D. It is open only to accredited investors resident in the United States, whose accredited status we are required to verify before any investment can be accepted.
If you invest in deep tech, trust infrastructure or security, we'd like to talk.
- The engine and mobile prototype are working. DAE, OX1 and OX2 produce four forensic dimensions and an evidential verdict, with uncertainty surfaced instead of hidden.
- The problem compounds. Every improvement in generative models widens the gap this closes. The demand curve is set by someone else's R&D budget.
- Defensibility is in the discipline. Held-out evaluation, hash-verified models and evidential wording that never overstates — the properties that matter when a verdict is challenged.
- Multiple routes to revenue. Professional subscription, per-call API for platforms, and reports for institutional workflows.