LMS-2241 Production log

Pilot: a Grade 8 math concept video made 100% with AI

Deliverables

Item	File
Korean video (1080p, H.264, 30 fps, 4:27)	`LMS-2241_linear-function_ko_1080p.mp4`
Korean script (subtitles + TTS input, timestamps)	`LMS-2241_script_ko.md`
English video (1080p, 3:51)	`LMS-2241_linear-function_en_1080p.mp4`
English script	`LMS-2241_script_en.md`
Production log	`LMS-2241_production-log_en.md` (this file), `LMS-2241_production-log_ko.md` (Korean)

Unit: Linear functions and graphs — slope and y-intercept (the ticket's recommended unit)
Length: 4:27 = 4 s intro + 4:18 lesson + 5 s outro (requirement: under 5 minutes ✅)
Intro and outro: W Labs title card (grade, unit), closing card (next lesson preview, "made with AI" note), short code-synthesized chime
Subtitles: Korean subtitles burned into the video ✅
Teacher avatar: an original illustrated character drawn in code. No real person's likeness ✅
Textbook copying: none. The example (water tank, 1 L + 2 L per minute) and the practice problem (y = −2x + 4) are newly written ✅
Curriculum: 2022 revised curriculum, middle school math, "Change and Relationships" area — [9수02-15] understand linear functions and draw their graphs, [9수02-16] understand and apply the properties of linear function graphs (Grade 8). ⚠️ Matched using a secondary source (Namuwiki); still needs checking against the official NCIC text

English version

Same scenes, graphics and avatar; only the script, voice, subtitles and on-screen text are in English.

Voice: Microsoft Edge neural TTS en-US-AvaNeural, rate −6%
Notation: y = ax + b, same as the Korean version (not the US-style y = mx + b)
Length: 3:51 (English narration is faster; includes intro and outro)

Style variants

The same lesson was also rendered in four designs (each in Korean and English).

Variant	Design	Motion
Classic (default)	Light paper background, blue/orange, Noto Sans KR	Calm and clean, 0.4 s
Chalkboard	Green chalkboard in a wood frame, chalk colors, Gaegu handwriting	Soft, 0.5 s
Night Lab	Dark navy, cyan/magenta, IBM Plex Sans KR, teacher on the left	Sharp, 0.3 s
Pastel Pop	Cream, bold outlines, sticker-style shadows, Jua	Playful with a slight bounce, 0.3 s

Tools

Step	Tool	Notes
Script, direction, visuals, avatar, editing	Claude Code (Claude Opus 5.5)	Script, formulas, graphs and animation all AI-written
Korean voice	Microsoft Edge neural TTS, `ko-KR-SunHiNeural`, rate −4%	No API key needed
English voice	Microsoft Edge neural TTS, `en-US-AvaNeural`, rate −6%
Lip sync	ffmpeg extracts voice loudness → avatar mouth shapes
Video assembly and rendering	Remotion 4.0.535 (React)
Intro/outro chime	Synthesized in code with ffmpeg (sine-wave arpeggio)	No stock audio
Fonts	Noto Sans KR, Gaegu, Gowun Dodum, IBM Plex Sans KR, Jua	All Google Fonts, SIL OFL
Direction guides	Agent skills: Remotion skills, motion-art-direction, animation-principles, shot-composition

Prompts

User → Claude Code prompts (verbatim, in order):

Okay, let's test. Here create a video for this. https://uedu-dev.atlassian.net/browse/LMS-2241
Answers to Claude's questions: voice = "Edge neural TTS", avatar = "Illustrated character"
Let's improve a bit. Is the mouth fade transition really the right decision for this? It's a bit weird looking. → mouth changed to snap between 4 fixed shapes
Is it possible to create an english version? → English version
Please create other variants, shuffle the styles and design a bit. → 3 style variants
I think we also need to fix the neck? The hair is layered above the neck? Is that intentional. Also, I think the rendering is stuck. → neck/hair layering fixed, all videos re-rendered
Now, can we add an intro and outro section? → intro and outro added, all videos re-rendered

(Setup prompts — installing Remotion, the Remotion skills and the motion design skills — are excluded from production time.)

Direction rules (set by Claude using the motion-art-direction skill):

One calm, clean motion style throughout; scenes change with crossfades only
One focal element per screen (the graph or formula); the teacher avatar is secondary
Each visual appears when the sentence that mentions it starts (synced to narration)

TTS input was written separately for pronunciation, so math reads naturally (e.g. y = 2x + 1 → "y equals 2 x plus 1"). Full text in LMS-2241_script_en.md.

Time

Step	Time
First Korean video: ticket analysis, direction, script, TTS, scenes/avatar/subtitles, review, render	~30 min
Mouth improvement + re-render	~10 min
English version	~25 min
3 style variants	~30 min
Neck/hair fix + re-render of all 8 videos (render ~34 min, unattended)	~40 min
Intro/outro design, chime, review + re-render of all 8 videos (render ~35 min, unattended)	~55 min
Total	~3 h 10 min (first Korean video alone: ~30 min)

(Excludes one-time setup such as installing Remotion and packages.)

Cost

Item	Cost
TTS (Edge neural TTS)	₩0
Remotion (free license for small teams*)	₩0
Fonts, ffmpeg	₩0
Claude Code usage	Within existing subscription (no extra charge)
Total	₩0 extra

* Remotion is free for companies with 3 or fewer employees; larger companies need a company license. Check which license fits W Labs.

Math check (AI self-check — Korean reviewer still needed)

y = 2x + 1: x = 0, 1, 2, 3 → y = 1, 3, 5, 7 ✔
Slope: (3 − 1)/(1 − 0) = 2/1 = 2, (7 − 3)/(3 − 1) = 4/2 = 2 ✔
y-intercept: x = 0 → y = 1, the graph crosses the y-axis at (0, 1) ✔
Sign: y = x, y = 3x rise to the right; y = −x, y = −3x fall to the right; larger |a| is steeper ✔
Practice y = −2x + 4: slope −2, y-intercept 4, points (0, 4) → (1, 2) [−2·1 + 4 = 2] ✔
Definition: "y written as a first-degree expression in x, y = ax + b (a ≠ 0)" — matches the standard textbook definition. The water tank example only has x ≥ 0, so the graph scene says "if we let x be any number" before drawing the full line.

Open items

Korean reviewer confirms zero math errors — against LMS-2241_script_ko.md
2022 curriculum check — matches [9수02-15], [9수02-16] per a secondary source; confirm against the official NCIC text
For production use, consider a licensed TTS (e.g. Azure Speech) — Edge TTS is used unofficially