NBA Highlights Video Generator

Requires Python 3.15.1 or newer.

Videos

Posted on NBA Full Play Highlights.

Author

Ronen Huang

Time Frame

August 2025 to Present

FFmpeg Build

Download from https://www.gyan.dev/ffmpeg/builds/.

Ollama Build

Download from https://ollama.com/download and pull one time.

ollama pull lfm2.5-thinking:1.2b

NBA Team Abbreviations

• atl - Atlanta Hawks
• bkn - Brooklyn Nets
• bos - Boston Celtics
• cha - Charlotte Hornets
• chi - Chicago Bulls
• cle - Cleveland Cavaliers
• dal - Dallas Mavericks
• den - Denver Nuggets
• det - Detroit Pistons
• gsw - Golden State Warriors
• hou - Houston Rockets
• ind - Indiana Pacers
• lac - Los Angeles Clippers
• lal - Los Angeles Lakers
• mem - Memphis Grizzlies
• mia - Miami Heat
• mil - Milwaukee Bucks
• min - Minnesota Timberwolves
• nop - New Orleans Pelicans
• nyk - New York Knicks
• okc - Oklahoma City Thunder
• orl - Orlando Magic
• phi - Philadelphia 76ers
• phx - Phoenix Suns
• por - Portland Trail Blazers
• sac - Sacramento Kings
• sas - San Antonio Spurs
• tor - Toronto Raptors
• uta - Utah Jazz
• was - Washington Wizards

Manually Make Videos

The full play videos can be made from the reliable play by play rather than the unreliable box score.

from nba_video_generator.src import pipeline

pipeline(
    [
        ("DiVincenzo", "2026-04-20", "min"),
    ], 
    {
        "ffmpeg_path": r"C:\Users\ronen\Documents\Projects\nba_video_generator\src\nba_video_generator\ffmpeg-9.0.2-essentials_build\bin\ffmpeg.exe"
    }
)

A date range can be specified as well.

from nba_video_generator.src import pipeline

pipeline(
    [
        ("Booker", "2026-04-19", "2026-04-27", "phx"),
    ],
    {"ffmpeg_path": r"C:\Users\ronen\Documents\Projects\nba_video_generator\src\nba_video_generator\ffmpeg-9.0.2-essentials_build\bin\ffmpeg.exe"}
)

Process

1. Specify the player last name (as per NBA.com website), team abbreviation, and date(s) (yyyy-mm-dd).
2. Programs crawls through play by play by quarter, keeping a list of links and times.
3. Events within 5 seconds of each other are merged to a single event.
4. Plays are concatenated together to make the video.

Successfully generated player/team/date videos are recorded in processed_performances.json in the current working directory. Later calls to pipeline() skip dates already recorded; set NBA_AGENT_STATE before importing the package to use a different state-file path.

A video of the process is provided below.

https://www.youtube.com/watch?v=84GDSAL5CeE

Automatically Make Videos With Agent

Makes videos of the LLM determined best player performances of specified date (default today).

from nba_video_generator.nba_agent import run_agent

date = "2026-06-10"

ffmpeg_path = r"C:\Users\ronen\Documents\Projects\nba_video_generator\src\nba_video_generator\ffmpeg-9.0.2-essentials_build\bin\ffmpeg.exe"

run_agent(date, ffmpeg_path)

Process

1. Specify the date (yyyy-mm-dd).
2. LLM chooses players from pool of candidates that meet thresholds in src/nba_video_generator/nba_agent/config.py.
3. The videos are made with the pipeline method. The thumbnails with relevant statistics are made afterward.
4. (Optional) Upload to YouTube with Python. Example program in src/nba_video_generator/example_youtube_upload_scripts. Link to data api - https://developers.google.com/youtube/v3/guides/uploading_a_video.

A video of the process is provided below.

https://www.youtube.com/watch?v=IC8SrrvS4xE&feature=youtu.be

Automatically Make Game Top Plays Shorts - NEW

Makes shorts of the top 4 plays from each specified date (default today). Made scoring plays are excluded when they occur with 5:00 or less remaining in the fourth quarter and the team was already leading by at least 20 points. Fourth-quarter buzzer-beaters are also excluded unless they tie the game or change which team is leading. When choosing the top plays, scoring prioritizes buzzer-beater winners, buzzer-beater ties, other game-winning baskets, game-tying baskets, first-through-third-quarter buzzer-beaters (1–3 seconds), and-one dunks, and-one threes, and other dunks, in that order; other plays are ranked afterward by their existing style and situation scores.

from nba_video_generator.nba_shorts import make_shorts_for_date

date = "2016-06-19"

ffmpeg_path = r"C:\Users\ronen\Documents\Projects\nba_video_generator\src\nba_video_generator\ffmpeg-9.0.2-essentials_build\bin\ffmpeg.exe"

make_shorts_for_date(date, ffmpeg_path)

Web Interface

Run the video-generation and upload workflows from a local browser dashboard:

python -m pip install --upgrade nba-video-generator
nba-video-generator-web

After the new release is published, install it from PyPI as shown above. The dashboard opens at http://127.0.0.1:8765. It includes forms for the player pipeline (including date ranges and multiple players), the daily agent (load finished games for a date and choose which to analyze), and top-plays Shorts. Each video workflow lets you choose the output folder. When the player pipeline cannot match a steal, block, turnover, or miss to an exact NBA play event, it uses the action ID immediately before the first feed action found at that game clock. Jobs run in the background and stream their output to the activity panel. The YouTube upload workflow lets you select a video folder and Google client_secrets.json using local file pickers. Uploads default to private, preserve local files, and use a saved OAuth token under your home directory. Only one workflow can run at a time. To choose another port or keep the browser closed:

nba-video-generator-web --port 8766 --no-browser