References in. Matches out.
Register reported images as references. Corvinth matches them against your platform's indexed images and new uploads.
IMAGE SAFETY INFRASTRUCTURE
One reported image.Find its matches across your platform.
Image safety infrastructure for platforms with user-generated content. Corvinth detects matching images across your platform's existing media and new uploads.
Corvinth returns the match result.
Your platform decides what happens next.
PLATFORM-SCOPED MATCHING · MANAGED OR CUSTOMER COMPUTE · NO DURABLE IMAGE-BYTE STORAGE
Platform-scoped detection
Same platform scopePlatform-selected
image
Selected by your team
Reference→
Platform-scoped
matching
existing content · new uploads
Image
matching
matching
Searching your
platform content
platform content
Match found
match_foundtrueclassificationEXACT
Corvinth detects. Your platform decides.
Platform-scoped matchingFind matching content across your platform.
Managed or Customer ComputeChoose where image compute happens.
Corvinth detects. Your platform decides.Match results integrate into your existing review and enforcement workflow.
Built for
Social Platforms
Dating Apps
Messaging Apps
Creator Platforms
AI Communities
Adult Content Platforms
Why Corvinth
Save months of engineering
Detection isn't just a model call. You need references registered, existing images indexed, new uploads checked, and every match routed back into your workflow. That's the system Corvinth provides — without becoming your next infrastructure project.
Fits into your existing media pipeline.
Send presigned image URLs for Corvinth-managed processing, or send derived representations from your own infrastructure.
Your policy. Our detection.
Corvinth tells you what matched and provides a structured record for every case. What you do with that is yours.
You define what to track. We find it.
Your team selects the images to track. Corvinth detects matching images across your platform's existing media and new uploads.
Trusted architecture
Choose where image compute happens.
Use Corvinth-managed compute, or keep image processing inside your own infrastructure. Either way, Corvinth does not durably store image bytes.
Mode A — Managed Compute
Managed computeCorvinth handles the image compute.
- Your storage
- ephemeral fetch access
- Corvinth compute
- Corvinth detection infrastructure
- Match result
Presigned URL sent to Corvinth. Corvinth fetches the image for processing.
Mode B — Customer Compute
Customer computeImage pixels stay with you.
- Your image
- your compute
- fingerprint
- Corvinth detection infrastructure
- Match result
Image processing stays inside your infrastructure. Corvinth receives the required fingerprint together with the platform-scoped object reference needed to correlate the match result back to content in your system — never the image itself.
No durable image-byte storageYou choose where images are processedSame detection infrastructure
Two compute modelsChoose the trust boundary that fits your platform.
No durable image-byte storageCorvinth does not build a repository of customer images.
Your policy stays yoursCorvinth returns detection results and relevant context; your platform decides what happens next.
How Corvinth works
Report it once. Catch it when it comes back.
When your platform chooses reported content to track, Corvinth uses it as a detection reference. Perceptual matching checks that platform's stored and future fingerprints.
01 / PLATFORM-SELECTED REFERENCE
Reported content
Your platform tells Corvinth what should be tracked.
02 / CONTINUOUS MATCHING
Continuous perceptual matching — both directions.
Continuous perceptual matching checks newly reported content against previously stored fingerprints from that platform, then keeps matching against new uploads from that platform as they arrive.
Previously storedThat platform's previously stored fingerprints
Referencereported fingerprint
Future uploadsThat platform's new upload fingerprints
03 / MATCH + STRUCTURED CASE RECORDCorvinth detects. Your platform decides.
Corvinth returns the match; the case record preserves the context.
Your platform decides what happens next.
Built for real cases
Detection is only half the job. The match needs evidence behind it.
Corvinth keeps platform-selected visual references, matches, and timing linked so your team isn't piecing together detection history from isolated hashes.
01Platform-selected visual reference
02Case
03Matched platform object
04Detection event + timing
01 / Structured case record
Every match ties back to its originating case and involved platform content.
02 / Traceable object identity
Evidence points to the exact platform object being investigated, rather than only an isolated hash or vector.
03 / Audit history
References, matches, case history, and timing stay connected so your team can trace the detection history of the case.
04 / Tenant-isolated matching
A platform's references, matching, and evidence remain scoped to that platform. Another tenant's case, object, or evidence metadata must not be returned or exposed.
Match↓Case + object + timing + evidence↓Your platform's policy
Corvinth records the detection evidence. Your platform decides what happens next.
Integration
Add the detection layer. Keep the rest of your stack.
Register references, connect your media pipeline, and route match results into your existing workflow.
01 Your platform
Start withReported reference /
content to check
content to check
Customer Compute
Derived representationDirect to Detection / matchingManaged path
Ephemeral fetch accessPresigned URL sent to Corvinth02 Corvinth
Match result + detection metadata→Your existing workflow
01 / Scope
Establish the platform boundary.
Corvinth operates inside that platform's isolated scope for references, matching, and evidence.02 / Connect detection
Connect the compute path that fits your stack.
Managed ComputeCustomer ComputeBoth paths connect into the same Corvinth detection layer.03 / Route results
Send matches and evidence into the workflow you already use.
Corvinth result → internal tooling → review / case handling → platform actionCorvinth detects and returns context. Your platform decides what happens next.
Your stack stays yours
- Media storage
- Reference selection and policy
- Moderation workflows
- Enforcement decisions
Corvinth provides
- Detection references
- Image matching
- Match results and detection metadata
Corvinth returns match results. You decide what happens next.
Live API demo
See Corvinth work.
One reported image. Another upload.
See what Corvinth detects.
A private demo. Your invitation.
- 01Report an image
- 02Try a re-upload
- 03See the result
Technical foundation
PDQ is a foundation, not the product.
PDQ is an established image-fingerprinting method used within Corvinth. Corvinth builds the operational detection system around it.
About PDQWhy now
Removing one image is not the same as finding its copies.
Under the TAKE IT DOWN Act, covered platforms must remove qualifying intimate imagery and make reasonable efforts to identify and remove known identical copies within 48 hours of a valid request. Corvinth provides image-matching infrastructure to support that discovery process.
The operating reality
- report
- support / admin
- engineer
- manual search
- removal
Engineering doesn't start the clock. For a covered platform, the 48-hour period begins when it receives a valid removal request through its TIDA notice-and-removal process.
FAQ
Questions platforms actually ask.
The useful answer is the precise one. These are the boundaries teams usually want clear before they connect a detection system to their workflow.
Corvinth returns the match result and supporting context. Your platform decides how that result is reviewed or acted on under its own policy.
Your platform decides whether to review, remove, allow, or otherwise act according to its own policy. Matching thresholds and review settings are detection configuration—not automatic enforcement by Corvinth.
Corvinth is not partnered with or integrated into StopNCII and does not represent StopNCII. Corvinth operates its own image-matching infrastructure, incorporating perceptual hashing and DINOv2-based semantic image matching.
That depends on your processing mode.
Managed Compute: Corvinth fetches and processes an image from a presigned URL.
Customer Compute: Your platform can send precomputed matching hashes, with its own content identifier when it needs match correlation, instead of an image URL.
Small teams are where the tradeoff can matter most: manual discovery competes directly with product and engineering work. Corvinth provides the detection layer without requiring the platform to build and operate that infrastructure itself.
Integration means establishing the platform scope, connecting the compute path that fits your infrastructure—Managed Compute or Customer Compute—and routing Corvinth's results into the workflow your team already uses.
The integration surface is bounded, but the timeline depends on your storage, upload, review, and case-handling workflow.
Detection infrastructure should make the boundary clearer—not add another layer of ambiguity.
For platform users
You report to the platform. Corvinth helps it search.
If a platform turns the item you reported into a detection reference, Corvinth can search that platform's own content for matches—helping find copies beyond the item you originally reported.
You do not report to Corvinth. Corvinth does not decide whether your report is valid or what gets removed. Those decisions stay with the platform.
Corvinth is not a reporting destination.
a note from the team
There's a human accountable for this.
Founder & sole engineer, Corvinth — building in public
I built Corvinth because small platforms often don't have a detection pipeline — they have an inbox. Someone files a report, it reaches an admin, gets forwarded to an engineer, and valuable time disappears into manual searching.
Corvinth is built to replace the manual-search part of that chain with detection infrastructure, so when a platform is working against a removal deadline, engineering is not burning most of that window manually hunting for copies.
Small teams shouldn't need a dedicated trust-and-safety engineering organization just to have this infrastructure. I built Corvinth to fit a startup's infrastructure budget and work alongside the systems the team already uses.
I'm not anonymous — email me directly with questions, including hard ones about what Corvinth can and cannot do.
founder@corvinth.comFounding Partner Program
3 platforms only.
$99/month
For your first 3 months from go-live.
$299/month thereafter — your exclusive founding-partner rate.
Standard pricing: $499/month
Included
100K images/month — Continuous PDQ matching and re-upload detection.
Pulse DINOv2 — Early Access — Historical library scanning and semantic similarity. Review-only, with agreed scan limits.
Direct founder onboarding — Personal integration support and priority feedback.