From one match video
to reusable tactical assets
FAA automates detection, tracking, pitch calibration and tactical analysis. It takes broadcast feeds, screen recordings or local footage — no multi-camera rig and no manual calibration required.
On-premise: footage never leaves your machine · Bilingual EN/CN tactical reports · Youth and women's editions available
Every interface, figure and clip on this site is real system output: Yangzhou vs Suzhou, Xuzhou vs Suqian, Jiangsu Football Super League 2026.
AI vision is cutting the cost of tactical data by an order of magnitude
These six items are the real daily cost for coaching staffs, analysts and content teams.
Manual clipping is expensive
Tagging the key events of a 90-minute match takes hours, and plays still get missed.
Metric definitions drift
Different analysts and tools compute possession and pressure differently, so nothing is comparable across games.
Tactical knowledge stays in heads
A coach's intuition is never written down, and new staff start from zero.
Each match is a one-off
Analysing games in isolation means no season trends and no player growth curves.
Live games outrun you
In-match decisions need insight in seconds — post-match workflows cannot help.
Multi-camera calibration breaks
Long focal lengths and goal-end angles defeat classical calibration; tracks visibly misalign.
Video in, insight out
Upload and run — nothing to configure. Kit clustering assigns the teams, and per-frame calibration aligns the pitch.
Automatic team and camera calibration
Kit-colour clustering, voting and Re-ID split the teams. Per-frame calibration with white-line anchors works on long-lens and multi-camera footage.
Occlusion handling and resumable runs
The football-specific jarvis2 model plus ByteTrack keeps IDs through occlusion; quality gates on tracks and ball chains drop false positives. A full match can pause and resume automatically.
Accelerated on your own machine
Apple MPS GPU acceleration — an ordinary Mac starts it with one click, and neither the footage nor the analysis leaves the machine.
12 capabilities you can use from day one
Live broadcast analysis
Point it at a broadcast or screen capture and it analyses near real time — annotations and data panels while the game is on.
Full-match deep analysis
Upload the whole match and run detection, tracking, tactics and events end to end, with automatic resume after an interruption.
Player recognition and tracking
jarvis2 + ByteTrack keep IDs across frames and occlusions; quality gates on tracks and ball chains remove false positives.
Automatic team and camera calibration
Kit-colour clustering, voting and Re-ID assign teams; per-frame calibration with white-line anchors suits long lens and multi-camera.
Tactics and formation analysis
Formation recognition, passing networks, PPDA, defensive line height, xG, transitions and half-by-half comparison.
Shots · shots on target · goals
Two-stage multimodal verification for shots and shots on target, scoreline OCR to confirm goals, plus timeline and clips.
Dynamic tactical sandbox for the full match
Both teams and the ball replayed on a real pitch map, with variable speed, formations, heatmaps, 3D view and multi-match compare.
Tactic maps and tactic library
Named tactic maps generated from pressing, progression and transitions, filed into a searchable library.
Player profiles and achievements
Goals, shots on target, assists, cards and saves collected automatically and aggregated across appearances.
Lineup OCR + commentary ASR
Starting XI, substitutes and shirt numbers read automatically; commentary transcribed into structured goals, shots and assists.
League analysis and team management
Season- and round-level team data aggregated into league reports, with data packs managed in one place.
AI Q&A and report generation
Conversational insight from a local model, and one-click bilingual EN/CN tactical PDF reports.
The interface as it really is — every conclusion jumps back to a frame
Screenshots come from the Chinese build of the product; reports and exports are available in English.






It doesn't stop at what happened
It accumulates tactical knowledge, profiles both teams and answers "what if we changed shape".

Tactic library: coaching experience as a searchable asset
24 named tactics filed under attacking / defending / transition / set pieces, each carrying relational data.
- ◆ Counterswhich opponent patterns this neutralises
- ▲ Countered bywhich opponent tactics break it
- ■ Preconditionsformations and game states it applies to, filterable
- Tactics usable against tonight's opponent are recommended automatically, with reasoning and risks

Duel simulation: what happens if you change shape
Two engines score every candidate formation against the current opponent.
- Engine A6 × 6 candidate formations, 300 randomised perturbations, scored across 15 dimensions
- Engine B916 duel segments cut from the full-match sandbox, possession transitions recomputed per segment
- Sample result: 4-2-2-3 scores 67% against this opponent, with counter / wide-area / deep-zone breakdowns
The interface itself labels this "simulated estimate · not a real probability · tactical reference only". xG strength and opponent style can be adjusted manually.

Tactical advice: a recommendation, not a literature review
Both team profiles compared automatically: possession 67.8% vs 32.2%, PPDA 6.21 vs 17.91, formation 4-1-3-2 vs 3-4-2-1.
- Strength / weakness cards for each team, each with verifiable evidence
- Defensive line height, transition efficiency and shot quality compared side by side
- Any conclusion jumps straight to the matching video segment

League album: single matches compound into season assets
Example: Jiangsu Football Super League 2026, all 13 teams filed into one album.
- Played / won / drawn / lost / GF / GA / GD / points accumulate match by match; the table exports as CSV
- Team × dimension capability matrix — attacking xG, defence, ball progression, fitness and distance — compared across the league
- Every additional match analysed makes the album richer
What one match actually produces
The figures below are the system's real output for Yangzhou vs Suzhou, not a mock-up.
| Metric | Yangzhou | Suzhou |
|---|---|---|
| Possession | 47.5% | 52.5% |
| Shots | 11.0 | 6 |
| Shots on target | 1.1 | 0.6 |
| Farthest shot | 37.8 m | 43.4 m |
| Pressing intensity | 40.7% | 33.9% |
| Off-ball runs | 65.6 | 109 |
The parameters view compares them on one screen: possession 68.7% : 31.3%, shots 28 : 13.
| Item | Result |
|---|---|
| Named players output | 39 |
| Whitelist hits | 39 / 45 |
| Coordinate system | 105 × 68 m |
| Unmatched tracks | excluded |


Three ways to run it
There is no standard price list: we quote by squad size, match volume and deployment model — demo first, contract after.
On-premise
One-click launch on a Mac. Raw footage and tactical data never leave the club — built for an in-house analysis room.
Cloud collaboration
The tactic library and match assets sync across devices, so the coaching staff annotates and reviews on one shared timeline.
Live match
Point it at a stream and analysis starts; a half-time report and opponent profile land before the second half.
Custom editions: youth academy · women's football · esports-style UI. Multi-tenant plans for academies and institutions by arrangement.
Downloads
No sign-up, no login — click and it downloads.
FAA Football AI Analyzer — system introduction
Why FAA → one match, one asset set → differentiators → delivery and outputs, with real interface captures and real match figures
Download PDF (EN) →FAA feature walkthrough
Paginated English introduction covering all 19 sections, with a real 3D sandbox capture in the middle (Chinese interface, all exports in English)
Watch online / download →Chinese materials (19-page PDF · 5 min 17 s original cut) are on the Chinese page. Original high-bitrate files (1904×900 · 49 MB) available on request — email us for a direct link.
Run it on the next match you already have
Send us one full match video. We return the sandbox, the event timeline and the tactical report, so you can see whether it fits straight into your workflow.
When you write, please add: club or organisation level · source of the footage · roughly how many matches per month