Inspiration
Every video contains moments worth sharing, but not every moment is easy to discover, verify, or monetize.
Our inspiration for ProofMint AI came from a fundamental problem in digital content commerce: creators produce valuable moments, yet licensing systems primarily treat videos as complete assets rather than collections of individually valuable segments.
A seven-second highlight from a sports event, a powerful emotional scene, or a memorable product interaction might be exactly what an advertiser, publisher, or content creator needs. However, finding that moment, establishing its provenance, negotiating usage rights, and completing payment often involves multiple disconnected processes.
We wanted to answer a simple question:
What if discovering and licensing a specific video moment could be as seamless as finding a product and checking out online?
This inspired us to combine computer vision, AI-assisted discovery, digital provenance techniques, and PayPal-powered payments into one platform.
With ProofMint AI, videos become searchable collections of micro-assets. AKAZE feature matching, RANSAC geometric verification, and experimental digital watermarking provide additional provenance evidence, while PayPal enables moment-level transactions with traceable, signed licensing receipts.
Our broader vision is a future of agentic commerce, where intelligent agents can help buyers discover relevant media, evaluate available provenance evidence, and facilitate licensing with explicit human approval.
We believe the next evolution of the creator economy will not simply be about monetizing more content—it will be about unlocking the value of individual moments while building trust into every transaction.
ProofMint AI was born from that vision: Discover the Moment. Verify the Evidence. License with PayPal.
What it does
ProofMint AI is an AI-assisted video discovery, provenance verification, and micro-licensing platform that transforms videos into searchable, individually licensable digital assets.
Instead of requiring buyers to license an entire video, ProofMint AI enables them to discover, verify, and purchase rights to specific moments through PayPal.
- Intelligent Video Moment Discovery
Creators upload videos, which are analyzed using OpenCV-based scene detection and motion analysis. The platform identifies candidate moments, extracts representative frames, and assigns timestamps, descriptions, and searchable metadata.
- Searchable Moment Marketplace
Buyers can search for relevant video moments using descriptive queries such as "calm opening" or "highlight candidate." Each moment includes its duration, contextual information, source reference, and licensing price.
- Visual Provenance Verification
ProofMint AI uses AKAZE feature detection and RANSAC geometric verification to compare candidate images against reference frames.
This provides evidence of visual correspondence, helping users assess whether an image matches a particular source moment.
- Keyed Digital Watermarking
The platform implements experimental DCT-based watermarking on representative video still frames. Watermarks can be embedded, downloaded, and verified to provide an additional layer of provenance evidence.
- Narrative Relationship Mapping
ProofMint AI organizes detected moments into a temporal graph, preserving their sequence and relationships within the original video.
- PayPal-Powered Micro-Licensing
Buyers can license individual video moments through PayPal Sandbox checkout. Each purchase is tied to the selected moment, its licensing terms, and payment transaction identifiers.
- Cryptographically Signed License Receipts
After successful payment, ProofMint AI generates an HMAC-SHA256-signed receipt containing:
Licensed moment and original video identifiers.
Exact segment timestamps.
Asset SHA-256 fingerprint.
License terms and transaction amount.
PayPal order and capture identifiers.
A tamper-evident signature.
Working Prototype
Our deployed application successfully demonstrates video ingestion, moment discovery, AKAZE matching, watermark verification, temporal graph generation, and PayPal Sandbox licensing.
In our demonstration, ProofMint AI identified four moments from a 30-second video, achieved 18 RANSAC inliers from 21 candidate matches, recovered 32/32 watermark bits, and generated a signed receipt for a $4.99 PayPal Sandbox transaction.
ProofMint AI brings video discovery, visual provenance evidence, and digital commerce together in one seamless workflow.
Note: Provenance checks are experimental technical evidence, not independent certification of copyright ownership. Watermarking currently operates on still frames rather than entire videos.
Discover the Moment. Verify the Evidence. License with PayPal.
How we built it
We built ProofMint AI as an end-to-end, AI-assisted video micro-licensing platform, combining computer vision, digital provenance verification, payment integration, and cryptographically signed receipts.
Our goal was to create a working application where users could upload videos, discover valuable moments, verify visual provenance, and license individual segments through PayPal.
- Building the Video Intelligence Pipeline
We started with Python, OpenCV, and FastAPI to develop the video processing backend.
Our pipeline analyzes uploaded videos, identifies candidate scene boundaries using visual and motion signals, and extracts representative frames.
Each detected moment is assigned a unique identifier, timestamps, motion-based descriptions, and licensing metadata.
We also developed a MomentWeave integration adapter to support structured moment-analysis exports, allowing the architecture to accommodate richer video intelligence workflows.
- Implementing Visual Provenance Verification
We integrated AKAZE (Accelerated KAZE) feature detection to identify distinctive visual features within video frames.
Using feature matching and RANSAC, we evaluate geometric consistency between a candidate image and a reference frame. This helps determine whether resized or cropped images visually correspond to the original content.
We also implemented perceptual hashing (dHash) as an additional lightweight image similarity check.
- Developing Keyed Digital Watermarking
We implemented an experimental Discrete Cosine Transform (DCT) watermarking mechanism.
The system embeds keyed watermark information into representative video still frames and provides a verification endpoint to recover and evaluate the embedded bits.
In our demonstration, the verifier successfully recovered 32 out of 32 watermark bits.
- Creating the Narrative Moment Graph
We designed a graph representation that connects individual video moments and preserves their temporal relationships.
The current implementation uses an OpenCV-based temporal fallback, producing nodes for detected moments and edges representing their sequence.
This provides a foundation for future semantic narrative understanding.
- Building the Marketplace Frontend
We developed the user interface using Next.js, React, and TypeScript.
The application allows creators and buyers to:
Upload and analyze videos.
Search and browse detected moments.
Preview source footage and representative frames.
Compare candidate images using AKAZE.
Generate and verify keyed watermarks.
Explore narrative relationships.
Purchase moment-level licenses through PayPal.
- Integrating PayPal Sandbox Payments
We integrated PayPal REST APIs to support micro-licensing transactions.
When a buyer selects a moment, the backend creates a PayPal Sandbox order and directs the buyer to the approval flow.
After the approved payment is captured, the application records the transaction identifiers and generates a moment-specific licensing receipt.
We successfully demonstrated a $4.99 Sandbox transaction linked to an individual video segment.
- Implementing Signed License Receipts
To connect provenance evidence with commerce, we implemented HMAC-SHA256 receipt signing.
Each receipt records the purchased moment, original video reference, asset fingerprint, exact timestamps, licensing terms, and PayPal payment identifiers.
The signature provides tamper-evident integrity protection when verified using the corresponding secret.
- Database, Testing, and Cloud Deployment
We used PostgreSQL with SQLAlchemy to maintain video assets, detected moments, purchases, and licensing receipts.
We containerized the FastAPI backend using Docker and deployed the application on Render, with separate frontend, backend, PostgreSQL, and persistent media storage components.
During integration, we resolved CORS configuration issues and PostgreSQL foreign-key insertion ordering to ensure reliable browser-based video ingestion.
We also implemented automated backend tests, achieving 26 passing tests, and validated the main application workflow on the deployed environment.
Technology Stack
Frontend: Next.js, React, TypeScript
Backend: Python, FastAPI, SQLAlchemy
Computer Vision: OpenCV, AKAZE, RANSAC, perceptual hashing, DCT watermarking
Database: PostgreSQL
Payments: PayPal REST APIs and Sandbox
Security and Provenance: SHA-256, HMAC-SHA256, keyed watermark verification
Infrastructure: Docker, Render, persistent media storage
Final Outcome
We successfully built and deployed a working prototype demonstrating the complete workflow:
Upload Video → Discover Moments → Search → Verify Visual Provenance → Generate Watermarks → License with PayPal → Issue Signed Receipt
Our implementation combines video intelligence, experimental provenance techniques, and payment infrastructure to demonstrate a more granular model for digital media commerce.
ProofMint AI turns individual video moments into discoverable, traceable, and licensable micro-assets.
Challenges we ran into
g Visual Provenance Across Image Transformations
Establishing whether a candidate image corresponds to an original video frame was technically challenging, particularly when images were resized, cropped, or compressed.
Simple pixel-by-pixel comparisons are too sensitive to these transformations.
How we addressed it: We integrated AKAZE feature detection with RANSAC geometric verification to evaluate visual correspondence using local feature matches.
Our deployed demonstration achieved 18 geometrically consistent matches from 21 candidate matches, producing an 85.71% inlier ratio.
We also included perceptual hashing as a lightweight complementary similarity check.
- Implementing Reliable Digital Watermark Verification
Developing a watermarking mechanism required balancing two competing objectives: embedding information without substantially degrading the image and recovering that information reliably.
Compression and image transformations can affect watermark detection.
How we addressed it: We implemented experimental keyed DCT watermarking for representative video still frames.
Our initial end-to-end verification recovered 32 out of 32 watermark bits, demonstrating successful embedding and detection on the tested image.
Robustness against aggressive compression, resizing, and other transformations remains an area for further evaluation.
- Connecting Provenance Evidence with PayPal Transactions
We wanted more than a conventional checkout experience. Every purchase needed to reference the exact video moment being licensed, its source fingerprint, timestamps, and licensing terms.
This required coordinating PayPal order creation, buyer approval, payment capture, database records, and receipt generation.
How we addressed it: We built a PayPal Sandbox micro-licensing workflow and implemented HMAC-SHA256-signed receipts that link payment references to moment-level metadata.
We successfully demonstrated a $4.99 PayPal Sandbox licensing transaction with a generated signed receipt.
- Resolving Cloud Deployment and Browser Security Issues
Moving from a local Windows development environment to a publicly hosted application introduced several unexpected challenges.
Our Next.js frontend initially encountered deployment dependency issues, while the FastAPI backend required adjustments to Docker paths, production environment variables, and CORS configuration.
A particularly difficult issue occurred when video uploads failed even though the API health endpoint was working.
How we addressed it: We traced browser preflight requests and discovered that the backend was rejecting the custom X-Demo-Upload-Token header. Updating the FastAPI CORS configuration resolved the preflight failure.
This reinforced the importance of testing real browser requests rather than relying solely on backend health checks.
- Debugging PostgreSQL Foreign-Key Constraints
After resolving CORS, the video upload still returned HTTP 500.
The Render logs revealed that SQLAlchemy was attempting to insert detected moments before their parent video record existed in PostgreSQL, resulting in a foreign-key violation.
How we addressed it: We introduced an explicit db.flush() after adding the parent video record. This ensured the video was inserted before its associated moments while preserving the transaction's atomicity.
After redeployment, video ingestion and moment generation worked successfully on Render.
- Balancing Technical Ambition with a Working Hackathon MVP
Our broader vision included richer narrative intelligence, autonomous commerce agents, robust watermarking, and advanced provenance verification.
However, implementing everything simultaneously risked producing an impressive concept without a reliable end-to-end demonstration.
How we addressed it: We prioritized a working vertical slice connecting video ingestion, moment discovery, AKAZE verification, keyed watermarking, PayPal Sandbox checkout, and signed receipts.
We clearly separated implemented functionality from experimental capabilities and future enhancements.
What We Learned
The biggest lesson was that integrating multiple technologies into one reliable workflow is often more challenging than implementing the individual algorithms.
We learned the importance of transaction ordering, practical computer vision validation, secure API configuration, cloud observability, and clearly communicating the limitations of experimental provenance techniques.
Most importantly, we demonstrated that AI-assisted video discovery, visual provenance evidence, and PayPal-powered micro-licensing can work together in a single deployed application.
What began as an idea evolved into a functional prototype with measurable computer vision results, automated tests, and a completed Sandbox licensing transaction.
These challenges made ProofMint AI more than a hackathon experiment—they helped establish the technical foundation for a future marketplace for licensable video moments.
Accomplishments that we're proud of
Building ProofMint AI has been an incredibly rewarding experience. What started as an ambitious idea—making individual video moments discoverable, verifiable, and licensable—evolved into a fully deployed, end-to-end working prototype that combines computer vision, digital provenance, and PayPal-powered commerce.
Here are the accomplishments we're most proud of:
- Turning an Ambitious Idea into a Live Application
Our biggest achievement was moving beyond a proof of concept and building a publicly accessible application.
We successfully deployed our Next.js frontend, FastAPI backend, PostgreSQL database, and persistent media storage on Render.
Live Application: https://proofmint-web.onrender.com
- Building an End-to-End Video Micro-Licensing Workflow
We connected the entire journey into one platform:
Upload Video → Discover Moments → Verify Provenance → License with PayPal → Generate Signed Receipt
Rather than demonstrating isolated technologies, we created an integrated workflow that brings video intelligence and digital commerce together.
- Achieving Measurable Computer Vision Results
We successfully analyzed a 30-second test video and generated four individually identifiable moments.
Our AKAZE and RANSAC verification achieved:
18 geometrically consistent matches out of 21 candidate matches.
85.71% geometric inlier ratio.
Successful candidate visual matching against a source video frame.
This demonstrated how established computer vision algorithms can support practical provenance verification.
- Successfully Implementing Keyed Watermark Verification
We developed an experimental DCT-based watermarking system capable of embedding and detecting keyed information in representative video frames.
In our deployed test, ProofMint AI successfully recovered 32 out of 32 watermark bits, achieving complete bit agreement on the tested watermarked image.
We are particularly proud of integrating watermark verification directly into the same interface used for moment discovery and licensing.
- Completing a Real PayPal Sandbox Licensing Flow
One of our most important milestones was completing a $4.99 PayPal Sandbox transaction for an individual video moment.
The application generated a moment-specific licensing receipt containing the source video reference, segment timestamps, asset fingerprint, licensing terms, PayPal order and capture identifiers, and an HMAC-SHA256 signature.
This demonstrated that moment-level content licensing could be connected to an actual payment workflow rather than remaining a theoretical marketplace concept.
- Creating a Traceable Provenance-to-Payment Record
We brought together multiple technical evidence layers:
SHA-256 source and asset fingerprints.
Exact video segment boundaries.
AKAZE-based geometric image matching.
Experimental keyed watermark detection.
Temporal relationships between moments.
PayPal transaction references.
HMAC-SHA256-signed licensing receipts.
Together, these form the foundation of a traceable, provenance-linked digital commerce workflow.
- Building and Testing a Reliable Backend
We developed automated backend tests and achieved 26 passing tests.
We also resolved real deployment challenges involving Docker configuration, CORS headers, PostgreSQL foreign-key constraints, and cloud environment settings.
Successfully debugging these issues and seeing the complete application work in production-mode deployment was one of our most satisfying achievements.
- Demonstrating a New Opportunity for the Creator Economy
Beyond the technology, we're proud of exploring a new commercial model: licensing individual moments rather than requiring the purchase of entire videos.
Our prototype demonstrates how creators, publishers, advertisers, and future AI-assisted commerce systems could discover and transact around precisely identified video segments.
Our Biggest Achievement
We didn't just build a video analysis application or integrate a payment button. We connected content discovery, visual provenance evidence, and PayPal-powered licensing into one working system.
ProofMint AI demonstrates how technologies that traditionally operate independently can come together to create new opportunities for trusted digital media commerce.
We're proud to have transformed a complex technical vision into a functioning, publicly accessible prototype with measurable results and a successful end-to-end licensing demonstration.
ProofMint AI — Discover the Moment. Verify the Evidence. License with PayPal.
What we learned
Building ProofMint AI taught us that creating an intelligent digital marketplace requires much more than combining AI models with payment APIs. The real challenge lies in making computer vision, provenance verification, secure transactions, and cloud infrastructure work together reliably.
Here are our biggest learnings from the PayPal AI Hackathon.
- AI Becomes More Valuable When Connected to Real-World Actions
One of our most important realizations was that AI-powered discovery becomes significantly more useful when it leads to a meaningful outcome.
Identifying interesting moments in a video is valuable, but enabling users to verify and license those moments creates an entirely new commercial opportunity.
We learned how AI-assisted discovery and PayPal-powered commerce can transform video intelligence into actionable transactions.
- Provenance Requires Multiple Layers of Evidence
We discovered that no single technique can establish the authenticity or ownership of digital media.
Through our implementation, we learned how different approaches complement one another:
SHA-256 identifies exact digital assets.
AKAZE and RANSAC provide geometric evidence of visual correspondence.
Perceptual hashing enables lightweight image similarity checks.
Keyed DCT watermarking provides additional evidence through embedded information.
HMAC-SHA256 helps protect the integrity of licensing receipts.
Combining these methods creates a more informative provenance workflow, although technical evidence alone cannot establish legal copyright ownership.
- Computer Vision Algorithms Have Practical Trade-Offs
Implementing AKAZE and RANSAC helped us better understand the challenges of matching images that have been resized, cropped, or otherwise transformed.
We learned that geometric consistency matters more than simply counting feature matches.
Our successful demonstration, with 18 RANSAC inliers from 21 candidate matches, reinforced the importance of evaluating match quality rather than relying on raw similarity scores.
We also learned that watermark recovery depends on image transformations and embedding parameters, making robustness testing an important area for future improvement.
- Payment Integration Is About Trust, Not Just Checkout
Integrating PayPal taught us that a payment workflow involves more than creating an order and redirecting a buyer.
We needed to associate each transaction with a precise video segment, licensing terms, payment identifiers, and a verifiable receipt structure.
Our successful $4.99 PayPal Sandbox transaction demonstrated how payment infrastructure can connect digital content discovery to traceable licensing records.
We also gained a deeper appreciation for explicit buyer approval, secure credential handling, and transaction verification.
- Cloud Deployment Reveals Problems That Local Testing Cannot
Our application initially worked locally, but deployment introduced challenges involving Docker paths, environment variables, CORS policies, and PostgreSQL transaction ordering.
One particularly valuable lesson came from debugging video uploads that returned HTTP 500.
We traced the failure to a PostgreSQL foreign-key violation and resolved it by explicitly flushing the parent video record before inserting its associated moments.
This taught us the importance of observability, database consistency, integration testing, and real browser-based validation.
- Reliable Systems Matter More Than Feature Count
We initially envisioned advanced narrative intelligence, sophisticated watermarking, and autonomous agentic commerce.
However, we learned that delivering a reliable end-to-end experience was more valuable than implementing numerous disconnected features.
By prioritizing video ingestion, moment discovery, provenance verification, PayPal checkout, and signed receipts, we created a working vertical slice that demonstrates the core business concept.
We also learned to distinguish clearly between production-ready capabilities, experimental components, and future enhancements.
- Agentic Commerce Needs Verifiable Context and Human Control
Exploring AI-assisted media licensing helped us recognize that intelligent commerce systems should not simply recommend products or initiate transactions.
They should provide meaningful context about what is being purchased, explain available evidence, respect licensing conditions, and require explicit buyer authorization.
This insight shaped our longer-term vision of AI agents that can assist with media discovery and licensing while keeping users in control of purchasing decisions.
- The Creator Economy Has Opportunities Beyond Whole-Asset Licensing
Perhaps our most exciting learning was that the value of digital media does not always reside in the entire asset.
A few seconds of footage can be valuable independently, provided users can identify the moment, understand its context, evaluate available provenance evidence, and obtain suitable licensing rights.
This opened our thinking toward future applications in advertising, entertainment, sports highlights, digital publishing, and creator marketplaces.
Our Biggest Takeaway
The future of AI-powered commerce is not simply about making content discoverable. It is about making discoveries actionable, traceable, and trustworthy.
ProofMint AI taught us how computer vision, experimental digital provenance, payment infrastructure, and cloud engineering can work together to support an entirely new approach to media licensing.
We leave this hackathon with a functioning prototype, stronger engineering experience, and a clearer vision for building trustworthy AI-assisted digital marketplaces.
What's next for ProofmintAI
Our journey with ProofMint AI doesn't end with the PayPal AI Hackathon. We've demonstrated that individual video moments can be discovered, linked to provenance evidence, and licensed through PayPal. Our next goal is to evolve this working prototype into a scalable, intelligent marketplace for digital media licensing.
- Advance AI-Powered Moment Intelligence
We plan to move beyond OpenCV-based scene and motion analysis by integrating multimodal AI models that understand visual context, dialogue, emotions, actions, and narrative transitions.
Our long-term goal is to automatically identify commercially valuable moments based on their meaning and potential use cases—not just visual changes.
We also envision richer narrative graphs that capture relationships between characters, events, emotions, and storylines.
- Build Autonomous, Human-Governed Commerce Agents
A major next step is developing intelligent agents capable of discovering relevant video moments, comparing licensing options, evaluating available provenance evidence, and recommending suitable content.
We want buyers to make requests such as:
"Find a five-second inspirational moment suitable for a technology advertisement, verify its provenance, and help me license it for under $10."
Our agents would coordinate discovery, evidence evaluation, and checkout preparation, while keeping final payment authorization under explicit human control.
- Strengthen Digital Provenance and Watermarking
We plan to extend our experimental still-frame watermarking implementation to full video sequences.
Future research will focus on watermark robustness against compression, cropping, resizing, frame extraction, and transcoding.
We also intend to improve AKAZE-based matching with stronger visual embeddings, confidence calibration, and systematic benchmarking against both matching and unrelated content.
Our goal is to provide transparent, measurable provenance evidence without overstating what technical verification can prove about legal ownership.
- Expand PayPal-Powered Licensing
We envision supporting more flexible licensing models, including:
Usage-based micro-licensing.
Time-limited and campaign-specific licenses.
Multiple licensing tiers for commercial reuse.
Creator-defined pricing and usage restrictions.
Automated payment reconciliation and transaction-linked receipts.
We also plan to introduce verified creator accounts, secure authentication, licensing approval workflows, and stronger receipt verification.
- Create a Creator-Centric Marketplace
ProofMint AI will evolve toward a marketplace where creators can upload media, manage rights declarations, organize individual moments, and make selected segments available for licensing.
Buyers will be able to discover content through semantic search, preview candidate moments, review available provenance evidence, and purchase appropriate licenses.
We envision applications across advertising, filmmaking, entertainment, sports, journalism, education, and digital publishing.
- Improve Scalability, Reliability, and Security
As we move beyond the hackathon prototype, we plan to introduce asynchronous video processing, cloud object storage, background job queues, stronger access controls, comprehensive observability, and production-grade transaction handling.
We also aim to expand automated testing, benchmark performance on larger video collections, and independently verify both PayPal payment status and receipt integrity.
- Explore Standards-Based Content Provenance
We want to investigate interoperability with industry provenance standards such as C2PA, enabling ProofMint AI to complement existing content authenticity and attribution ecosystems.
This would help connect moment-level technical evidence with broader, interoperable provenance workflows.
Our Long-Term Vision
We envision ProofMint AI becoming an intelligent commerce infrastructure layer for digital media, where valuable moments can be discovered, evaluated, and licensed individually.
Imagine a future where creators can monetize meaningful seconds of their content, businesses can find exactly the footage they need, and AI agents can facilitate licensing workflows with transparent evidence and secure, human-authorized payments.
The PayPal AI Hackathon helped us validate the technical foundation of that vision.
Our next mission is to transform ProofMint AI from a working prototype into a trusted, scalable platform for AI-assisted media commerce.
Built With
- akaze
- artificial-intelligence
- computer-vision
- digital-watermarking
- docker
- fastapi
- hmac
- nextjs
- opencv
- paypal
- postgresql
- python
- ransac
- react
- render
- sha256
- sqlalchemy
- typescript
