家计簿应用广告动画

MiniMax H3で、架空の家計簿アプリ「マネミル」のCMを生成。

MiniMax H3の良い点はたくさんありますが、このようなCMの動画生成部分を無料で作れる、というのは凄いことです😳

今回は、ストーリーボード生成に10円ちょっと（ElevenLabs経由でGPT Images 2.0を使用）。
動画生成は、Hailuoを使えば15秒の動画を140円くらいで作れます（768pの場合）。

また、MiniMax H3をローカル（パソコン）にインストールすれば、無料で作れます。

そして、もう一つ良いところが、ストーリーボードさえあれば、そこから簡単に動画化できること。
企画段階でストーリーボードを作っておけば、そのまま動画にできるので、アイデアを形にするハードルがかなり下がります😌

MiniMax H3で使用したプロンプトはこちら↓
------------------
subject_definitions:
<Subject 1> is the anime girl in <Picture 1>: Japanese TV-anime style, early twenties,
shoulder-length mint-green hair, large golden eyes, soft cream knit top.
<Subject 2> is the graphic style in <Picture 1>: soft flat 2D vector work on white and mint -
light ribbons, round pastel icons, translucent charts, particles, clean phone UI.
<Subject 3> is the app branding in <Picture 1>: a rounded teal app icon holding a smiling face
and bar chart, with the logotype "マネミル" in rounded gothic, mint green.

summary:
[reference generation] A 15-second Japanese commercial for a household budgeting app, drawn as clean
anime-style 2D animation over smooth motion graphics with animated Japanese type, carried by a
calm female voiceover. <Subject 1> goes from not knowing where her money goes to seeing it
clearly and breathing out. <Picture 1> is a storyboard sheet used only for casting and style.

retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 5], [Shot 8]): partially_preserved - her mint hair,
golden eyes, face and cream knit top stay identical; only pose and expression change.
<Subject 2> (appears in every shot): weak_reference - only the mint-and-white palette, flat
vector feel and particles carry over; the storyboard's grid and borders are not reproduced.
<Subject 3> (appears in [Shot 10]): fully_preserved - the icon shape, letterforms and mint
green are exact.

detailed_description:
[Shot 1] Clean Japanese TV-anime 2D animation over soft flat vector motion graphics with
animated Japanese type. <Subject 1> rests her chin on her hand on a pale field,
brow troubled, as blank receipts and pale icons drift around her. Charcoal type fades up word
by word on the left reading "いくら使ったか わからない…". <Subject 1>, a soft
close-mic young woman's voice (S2), murmurs to herself: <d>[Japanese] わからない…</d> Push In,
small amplitude, slow speed.
[Shot 2] At 00:01.500, the blank receipts sweep into a spiral filling the frame in soft
gray-teal, then dissolve into mint and gold particles. White type assembles at center reading "そのモヤモヤ、". A calm Japanese woman in her late twenties, soft low voice (S1), says
off-screen: <d>[Japanese] そのモヤモヤ、</d> Roll Clockwise, small amplitude, fast speed.
[Shot 3] At 00:03.000, a mint-green field wipes in. Light ribbons stream in from the top carrying
wordless white pictograms - a bank, a card, a cart, a train - and converge into a phone whose
screen is a plain glowing panel with no writing. White type draws itself across the upper left
reading "つなぐだけで". (S1):
<d>[Japanese] つなぐだけで、</d> Truck Right, small amplitude, slow speed.
[Shot 4] At 00:04.500, the phone rises upright into a teal field and its screen builds itself
carrying only this Japanese text: the heading "今月の支出" at the top, the total "¥236,800"
counting up beneath, then a donut chart in five soft colors, then five list rows sliding in one
after another and reading exactly "食費 ¥48,600", "交通 ¥16,800", "日用品 ¥12,300",
"趣味・雑貨 ¥10,200", "その他 ¥8,900", each row a color dot, its label, then its amount, set
large enough to read. White type appears at the upper right
reading "家計が見える". (S1): <d>[Japanese] 家計が、見える。</d> Push In, small amplitude,
slow speed.
[Shot 5] At 00:06.000, a warm white field wipes in. <Subject 1> lights up with a bright smile as
translucent unlabelled bar charts and a rising line graph slide in behind her. Type scales up
gently on the left reading "気づきが増える". (S1): <d>[Japanese] 気づきが、増える。</d>
Pull Out, small amplitude, slow speed.
[Shot 6] At 00:07.500, a pale field wipes in and <Subject 1> is gone. Four round pastel icons pop in
one at a time - a fork, a bus, a bottle, a bed - each with one label rising beneath and nothing
else: "食費", "交通", "日用品", "その他". Type then appears at the lower left reading "ムダが見つかる". (S1): <d>[Japanese] ムダが、見つかる。</d> Static Shot.
[Shot 7] At 00:09.000, a clean white field wipes in. A bar chart draws itself column by column across five
months whose axis labels read only "4月", "5月", "6月", "7月", "8月", a dotted
trend line traces the tops, and the final bar lifts in warm gold. Type appears at the upper
left reading "未来がイメージできる". (S1): <d>[Japanese] 未来が、見えてくる。</d> Push In,
small amplitude, slow speed.
[Shot 8] At 00:10.500, a soft mint field with a gentle glow wipes in. Close on <Subject 1>, eyes closed
and hands clasped, letting out one quiet breath as her shoulders drop. Particles drift upward
and nothing else moves. No type and no voice here. Static Shot.
[Shot 9] At 00:12.000, a bright mint field with radiating light wipes in. The phone rises at center and a
message card builds on its screen carrying only this Japanese text: the heading "今月のまとめ"
and "いつも おつかれさまです！" beneath it, a round wordless mascot beside it. White type
appears at the lower left reading "そっと支えてくれる". (S1): <d>[Japanese] そっと、支えてくれる。</d> Pedestal Up, small
amplitude, slow speed.
[Shot 10] At 00:13.500, a clean white field wipes in. The teal app icon of <Subject 3> assembles at center
from scattered pieces, its logotype "マネミル" draws itself beneath, and a small button fades
up reading "無料ではじめる". Mint particles spread. (S1),
warm and unhurried: <d>[Japanese] マネミル。</d> The frame holds still until exactly 15.000
seconds. Static Shot.
Throughout: every character on screen must be Japanese and must come only from the quoted
strings above. Never invent words, never write Chinese or Korean, never produce pseudo-CJK or garbled
glyphs, no romaji. Any surface with no quoted string - buttons, receipts, chart bodies,
mascots - stays blank. Keep one clean Japanese TV-anime style with crisp
linework and soft cel shading; never photorealistic or 3D. Keep motion smooth and elegant. Every shot lasts exactly 1.5 seconds and cuts on its
stated timestamp. Only (S1) and (S2) speak, exactly the lines above: no ad-libs, no singing,
no lip sync except in [Shot 1]. Every line starts within 0.2 seconds of its shot beginning and
ends before that shot ends - never carry a line over, never leave one unspoken. On-screen type is rounded Japanese gothic that builds as animated shapes, never a
subtitle strip. No storyboard grid, borders or split screen. Keep <Subject 1>'s mint hair and
golden eyes identical, cut sharp, palette warm white, mint, teal, pale gold, soft pink and
light blue.

overall_soundscape:
A dry paper rustle carries the receipts and rises as they spiral, then breaks into an airy
shimmer. Digital ticks land as each icon reaches the phone and as each list row arrives, a
chime rings when the donut chart completes, and glassy pops mark the pastel icons. One quiet
exhale sits alone in [Shot 8].

non_diegetic_music:
An original 15-second gentle uplifting Japanese track at a moderate tempo: felt piano,
marimba, plucked synth, light brushed beat, airy pads, low under the voice. It opens muted,
brightens at 00:01.500, brings the beat in at 00:03.000, drops to pad and one bell at
00:10.500 so the breath has silence around it, returns at 00:12.000, then lands on one warm
chord that decays at 15.000 seconds.