easton-reel

Decoded from the "this is a message to all finance bros" (Easton) reel format. Produces the signature look: one word at a time, huge red Times-Bold, red glow, punch-scaled to the beat of speech, over a dim crushed-black warm grade.

Genre anatomy (what makes it read as "Easton / finance-bro")

1. War-room shot — dim room, key-light face only, authority props behind

(US map, Steve Jobs poster, monitors). Subject does the alpha finger-pinch.

1. Red-serif kinetic caption — ONE word per beat, Times/Trajan bold caps,

red fill + red outer glow, punched (scale 78→100%) on the stressed syllable, upper-third. Hook + name-drop in this style.

1. B-roll cutaways — Rolex wrist, eye macro, prop gag (cup-noodles

"I'm gambling"). Money-aesthetic vs broke-reality contrast = the joke.

1. Clean white-cyc outro — full body, yellow CapCut sans subs, CTA "tune in".
2. Format = hook → name-brand → story/joke → CTA. Trending audio low under VO.

The script does #2 and #3 (grade) automatically from a raw clip. #1/#4/#5 are shooting/scripting choices — see "Shoot checklist" below.

Pipeline (one command)

python easton_reel.py INPUT.mp4 [-o OUT.mp4] [options]

Steps: faster-whisper word timestamps → build .ass (one-word red-serif punch) → ffmpeg cover-crop 9:16 → grade → burn subs → encode.

Options

• --text "..." manual caption mode — skips transcription, distributes the

words one-per-beat evenly across the clip. Use for SILENT b-roll / timelapse (no audio to transcribe). | inserts a pause beat. Audio track auto-dropped when the source has none.

• --model tiny|base|small|medium whisper size. Default small (good CPU

accuracy). Use tiny/base for speed, medium for noisy audio.

• --size N caption font px (default 110).
• --style red Easton red-serif glow (default). --style yellow =

CapCut yellow-sans (use for b-roll / second pass).

• --upper force caption into top third (auto-on for red style).
• --marginv N vertical offset from top when --upper (default 430).
• --no-grade skip the cinematic colour grade (keep source look).
• --keep keep the generated .ass (hand-tweak then re-burn).

Examples

# default Easton red punch caption + grade
python easton_reel.py "raw_talkinghead.mp4" --upper

# faster draft
python easton_reel.py raw.mp4 --model base

# b-roll line in yellow CapCut style, no grade
python easton_reel.py broll.mp4 --style yellow --no-grade

House rules / gotchas (hard-won on this machine)

• NVENC is broken here — GPU driver exposes nvenc API 12.2 but ffmpeg 8

needs 13.0 → h264_nvenc fails. Script auto-falls back to libx264 -crf 20. Don't "fix" the nvenc path; the fallback is the working path until the GPU driver is updated. (Matches the SkynetLabs video deploy gotcha: NVENC-broken → libx264.)

• Events `Format:` line MUST list `Name` — the V4+ Events column order is

Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text. Omit Name and every Dialogue row (which carries the empty Name field E,,0,0,0,,) mis-parses — libass leaks a phantom leading comma into the text (,YOU). Symptom: clean .ass text but a red comma renders before every word. Fix = add Name to the Format line. Don't strip the comma in code — fix the header.

• Force `-pix_fmt yuv420p` — the filter chain (eq/curves/subtitles)

promotes output to yuv444p (profile High 4:4:4), which Windows Media Player / Photos refuse with 0x80004005 "unsupported encoding settings". Both encode paths pin -pix_fmt yuv420p → profile High, universally playable. Don't ship without it.

• Strip punctuation — whisper emits ,BROS.; build_ass strips leading/

trailing .,!?;:"()[]—- so words render clean (BROS). Apostrophes kept (I'M, EASTON'S).

• ASS colour is &HAABBGGRR (alpha,blue,green,red) — red fill = &H000000FF&.

Don't swap for RRGGBB.

• Windows path escaping — the subtitles= filter needs the drive colon

escaped (C\:/...); handled in code. Don't pass a raw Windows path.

• Fonts used: Times New Roman (red) / Arial (yellow) — both ship with

Windows. Swap to Cinzel/Trajan/Playfair in STYLES for a more "luxury" serif.

• Output is always 1080x1920 cover-crop (no letterbox). Feed vertical or

near-vertical source; very wide source loses the sides.

Tuning the look (edit STYLES / build_ass in easton_reel.py)

• Bigger glow → raise blur (red style, default 9) and border (4).
• Slower/snappier punch → change \t(0,90,...) ms and the \fscx78 start scale.
• Hold each word longer/shorter → words hold from their start to the _next_

word's start (always one on screen). For gap-punches instead, end at b.

• Two-track look (red hook + yellow b-roll): run twice — pass 1 --style red

on the hook clip, pass 2 --style yellow --no-grade on the b-roll clip, then concat.

Shoot checklist (to feed this skill the right raw clip)

• Phone vertical 1080×1920, 30fps. Lock exposure on face.
• Key light on face only; kill background light → crushed-black look the grade

leans into.

• Authority backdrop (map / poster / monitors). Sit centred, mid-frame.
• Script: punchy hook in first ~1.5s, drop your NAME early (brand), payoff/joke,

CTA "follow + tune in". Keep it <40s.

• Shoot b-roll separately (watch, eyes, prop gag) for cutaways.

Verified

Tested 2026-06-09: (1) 6s speech trim — faster-whisper word timestamps OK. (2) Full run on a silent DJI laptop timelapse (2688×1512 landscape, no audio) via --text manual mode → cover-cropped 9:16, graded, red-serif glow punch words snapped to beats, audio auto-dropped, nvenc→libx264 fallback. Output clean after fixing the missing-Name-column comma bug. Functional, shipped a real reel (easton_laptop.mp4).