AI-Powered Basketball Highlight Generation Platform

SUMMARY

Get Clipped partnered with Folio3 AI to build the MVP of an automated basketball highlight platform, with the goal of proving that computer vision could replace manual clip editing end-to-end before the product scaled to other sports.

The brief was to close the gap between a finished game and a shareable clip. Manual editing took hours per game and limited how many games could be covered in a night. Get Clipped needed a system fast enough to get highlights onto social platforms while the result still mattered.

Folio3 AI developed a video processing and event detection pipeline capable of ingesting uploaded game footage, extracting and buffering frames for analysis, classifying basketball-specific events such as field goals, free throws, assists, rebounds, and turnovers, tracking individual players across possessions, and merging confirmed highlight plays into a single game reel reviewable from an admin dashboard.

ABOUT THE CUSTOMER

Get Clipped is a sports content company that turns basketball game footage into short, social-ready highlight clips for players, teams, and leagues. The clips go out to Twitter and Instagram, where a good moment from the game can reach an audience within minutes of it happening.

The business depends on speed. A highlight posted within an hour of the final buzzer gets shared and saved, while the same clip published the next afternoon gets scrolled past. Footage was never the problem, because full game recordings were ready almost as soon as the game ended.

The problems with manual clipping:

• Time: A single game took hours of editing before the first clip was ready, pushing most highlights past the window where they held any real value.
• Effort: Every game needed a trained editor watching it end to end, so coverage was capped by how many games one person could sit through in a night.
• Accuracy: Clip boundaries are decided hundreds of times across a game, and consistency drops as the hours go on, so plays were cut short or missed entirely.
• Duplicated work: Player stats could not be captured during editing, which meant a second pass through the same footage to pull the numbers.

What the automated pipeline resolved:

• Turnaround: Hours of manual editing replaced by a path from confirmed play to finished clip in under 30 seconds.
• Editor dependency: Clips are trimmed and merged into a full-game reel automatically, so no one has to watch the game back to produce them.
• Consistency: Clip start and end points are set by scoring logic rather than editor judgment, applied the same way on every play.
• Single-pass capture: Nine event types, including player identification and stats, are detected during the same processing run as the clips.

THE CHALLENGE

Highlight production is still a manual craft. Someone watches the game, marks the timestamps, trims the clips, and exports for each platform. That workflow does not scale past a few games a week, and it does not run overnight.

Automating it is harder than general object detection. Basketball plays are dense, fast, and visually similar. A missed three and a made three look nearly identical for most of the frames that contain them. The system has to read the outcome, not just motion.

Three challenges defined Phase 01:

• Basketball-specific event detection at frame level: The models needed to distinguish 2PT, 3PT, and free throw attempts from makes, capture shot location, and separate offensive from defensive rebounds across continuous footage.
• Player identification and possession continuity: Stat attribution only works if a player keeps the same identity through occlusion, camera motion, and player clustering under the basket. Assists and turnovers depend entirely on tracking holding through the play.
• Deciding what counts as a highlight: Detection produces far more events than any viewer wants to watch. The pipeline needed confidence scoring and relevance logic to decide which detected plays were worth trimming, and where to set the clip boundaries so the play reads as a complete moment.

OUR DELIVERY APPROACH

TOOLS & TECHNOLOGIES

THE SOLUTION

Folio3 AI developed an AI-powered basketball highlight generation platform that connects video understanding to automated content production. An admin uploads a game through the dashboard. The system ingests the file, extracts frames continuously, and buffers them so models have enough surrounding context to judge a play rather than a single moment.

Detection models then analyze the buffered frames for basketball events: points scored by type with shot location, field goals made and attempted, three-point makes and attempts, free throws, assists, rebounds split into offensive and defensive, turnovers, player identity, and approximate minutes played. Every detected event is stored and tagged with a highlight relevance score.

Once a play clears the relevance threshold, the system trims a segment around it, formats the clip, and queues it. After the game has been fully processed, the queued clips are merged into a single highlight video covering the whole game. The admin reviews the output and tracks performance through dashboard analytics.

THE FIVE-STAGE HIGHLIGHT GENERATION PIPELINE

1. Ingestion and frame buffering: Uploaded MP4 game footage is ingested, decoded, and continuously buffered into frame windows sized for contextual analysis rather than single-frame inference.
2. Event detection and classification: Computer vision models identify basketball events across the buffered frames, classify shot type and outcome, and record shot location on the court.
3. Player identification and stat attribution: Tracking maintains player identity across possessions so assists, rebounds, turnovers, and approximate minutes played attach to the correct individual.
4. Highlight scoring and clip generation: Detected events are scored for highlight relevance, and confirmed plays trigger automatic trimming of a short segment with appropriate lead-in and follow-through.
5. Merge, review, and analytics: Generated clips are merged into one full-game highlight video, surfaced in the admin dashboard for review alongside basic engagement and view metrics.

SOLUTION ARCHITECTURE

RESULTS ACROSS THE MVP PIPELINE