Ṣíṣeto: Fíìmù ìfilọ́lẹ̀ àwòrán Ora 2

Beere lọwọ mi fun: orukọ ọjà mi, ìró ohùn kan ṣoṣo, àmì mi, àwòrán ìṣàfihàn ọjà náà tó tó 3 sí 6, àwòrán tó ju 40 lọ ti ohun tó ṣẹ̀dá (tàbí àwòrán rẹ̀), àwọ̀ àti fọ́ọ̀ǹtì àmì ìdánimọ̀ mi, kọ́kọ́rọ́ API ElevenLabs àti ID ohùn, pẹ̀lú orin tí kò ní owó ọba pẹ̀lú ìjáde tí ó hàn gbangba (fáìlì ohùn àti àmì ìjáde àkókò). Tí mo bá fo ohunkóhun, lo àwọn àyípadà wọ̀nyí: "Ora 2", "àwòrán àwòrán tí a ṣe fún ìtọ́wò", àwọn àwòrán ọ̀fẹ́ láti Unsplash àti Pixabay, Geist, ink #111214, ohùn Samara X (19STyYD15bswVz51nqLf) lórí eleven_v4, àti Pixabay "Cascade Breathe" (129.83 BPM, ju sílẹ̀ ní 75.39s).

<script>
Kọ ohùn ìlà mẹ́sàn-án ní ọ̀nà yìí kí o sì ṣe àtúnṣe sí gbogbo ìlà náà sí ọjà mi:
"Èyí ni [orúkọ]. Ẹ̀ka kan... tí a kọ́ fún [iye]. [Ìṣe] ní ọ̀nà èyíkéyìí. Fún un ní [ìbáṣepọ̀] - ó sì kàn... yé e. Bí o ṣe ń lò ó tó, bẹ́ẹ̀ náà ni yóò ṣe dára sí i. Ṣàtúnṣe [ìbáṣepọ̀] rẹ̀. Ó jẹ́ [ìbáṣepọ̀]. Ó jẹ́ [ìbáṣepọ̀]. Gbogbo [ìjáde] - gẹ́gẹ́ bí o ṣe yàwòrán rẹ̀. [Orúkọ]. Jáde nísinsìnyí."
Lo àwọn àmì ìmọ̀lára díẹ̀díẹ̀: [pẹ̀lẹ́pẹ̀lẹ́] lórí àwọn ìlà ìdákẹ́jẹ́ẹ́ àti [ní ìtara] lórí ìlà [Ìṣe]. Má ṣe fi àmì sí ohun tí ó ń ṣí, kí o sì fi ìdádúró pípé parí rẹ̀.
</script>

Pe ElevenLabs API taara nipa lilo bọtini mi, laisi MCP. Ṣe agbekalẹ iwe afọwọkọ mẹrin ni kikun gba ni akoko kan ki o jẹ ki n yan. Ti ila kan ko ba dun daradara, tun ila yẹn ṣe ni igba mẹrin ki o si pin ẹya ti o dara julọ sinu ero mi.
Kúrú àwọn ìdádúró tó gùn ju 0.28s lọ sí 0.2s, lẹ́yìn náà mú gbólóhùn kọ̀ọ̀kan wá sí 85% ti ọ̀nà sí ìpele àárín, kí o sì ṣàtúnṣe èrè nígbà ìdádúró nìkan.
Fún àkókò ọ̀rọ̀, pín kíkà náà ní àkókò ìdádúró 100ms tàbí jù bẹ́ẹ̀ lọ, kọ apá kọ̀ọ̀kan sí ara wọn pẹ̀lú Whisper, kí o sì so ọ̀rọ̀ àkọ́kọ́ rẹ̀ pọ̀ mọ́ ìbẹ̀rẹ̀ tí a wọ̀n. Ìjáde Whisper kan ṣoṣo lè gbé àwọn ọ̀rọ̀ sí ìdajì ìṣẹ́jú-àáyá.

Fíìmù ìfilọ́lẹ̀ onígun mẹ́rin 22, 1440x1440 ní 60fps, ní irú ìgbékalẹ̀ yàrá AI: káfíìfù funfun, ìkọ̀wé kan, tàdáwà dúdú, àwọn àwòrán gidi, àti àwọn àkọlé tí a tẹ̀ ọ̀rọ̀ sí ọ̀rọ̀ ní àkókò tí a ń sọ wọ́n.
Gbogbo iṣakoso (ibi ti o ni kiakia, ibi ti a ṣeto, ati apo akọle) lo gilasi olomi: fi omi ṣan oju iṣẹlẹ ti o wa lẹhin rẹ, tẹ oju iṣẹlẹ naa ni eti bi gilasi ti o nipọn, lẹhinna fi didan oke, eti didan ati ojiji gbigbe rirọ kun. Gilasi yoo parẹ lori funfun lasan, nitorinaa fi awọ pastel ti o lọra kun si ami iyasọtọ mi tabi fifọ aworan ti o bajẹ lẹhin rẹ.
Èyí tí ó tẹnu mọ́ àwọn irú ọ̀rọ̀ ní ìpele díẹ̀ nínú àwọn àwọ̀ ilé iṣẹ́ mi, pẹ̀lú ìmọ́lẹ̀ díẹ̀ lẹ́yìn rẹ̀.
Àwọn òfin ìṣípo: jẹ́ kí kámẹ́rà máa rìn kiri (titẹ lati 1.0 si 1.04 fun iṣẹlẹ kọọkan; nigbati akoonu ba gbe lọ si iṣẹlẹ ti o tẹle, bẹrẹ ni zoom nibiti iṣẹlẹ ti tẹlẹ ti pari). Ko si ohun ti o han lẹsẹkẹsẹ: lo awọn ipadanu ti o dinku ti o kere ju 0.3s. Yi awọn aworan pada lori awọn akọsilẹ 16th, pẹlu titẹ fun ọkọọkan. Awọn iyipada aye ilẹ nigbati gbolohun ba bẹrẹ, ki o si ṣe deedee awọn idasilẹ orin pẹlu laini [Ìṣe]. Maṣe fi awọn iduro kikun han lori iboju.
Ti a ti gbesele: awọn afihan apoti yiyan, awọn slams ti a ti lu-snapped, awọn filasi funfun, awọn awoṣe 3D, awọn awoṣe.

Àwọn ìṣẹ̀lẹ̀ méje, tí a fi àkókò sí ohùn náà:

1. "Èyí ni [orúkọ]": àkójọpọ̀ àwòrán mi máa ń jáde bí orúkọ ṣe ń jáde ní ńlá, lẹ́yìn náà ó máa ń yí ààyè padà sí ìlà ẹ̀ka.
2. "tí a ṣe fún [iye]": àwọn àwòrán mi tó dára jùlọ tí wọ́n tàn lórí ojú ìbòjú gbogbo yí padà sí àwọn àkọsílẹ̀ 16 nínú ìsàlẹ̀ náà, pẹ̀lú ìlà tí a fi funfun tẹ̀ lórí wọn.
3. Àmì ìfàsẹ́yìn: ọ̀pá ìfàsẹ́yìn gilasi kan máa ń kọ ìfàsẹ́yìn kúkúrú kan. Káàdì èsì náà máa ń yí ìrísí padà ní gbogbo àkọsílẹ̀ 16, nígbà tí ìfàsẹ́yìn dúdú kan máa ń yí padà sí orúkọ tuntun kọ̀ọ̀kan lórí ìlà kékeré kan.
4. "Fi moodboard hàn": Àwọn àwòrán 9 fò sínú grid 3x3, lẹ́yìn náà yíká àbájáde náà bí "Ó ṣe ń mú ọ" ṣe ń tẹ̀ lórí ibojú.
5. "Bí o ṣe ń lò ó tó, bẹ́ẹ̀ náà ni yóò ṣe máa rí": tẹ ìlà méjì, pẹ̀lú ọ̀rọ̀ tí a tẹnu mọ́ nínú ìpele.
6. Àwọn ètò náà: fi páálí dígí kan hàn pẹ̀lú àwọn ìfàsẹ́yìn mẹ́ta, ọ̀kọ̀ọ̀kan ń gbéra bí ó ṣe ń pe orúkọ rẹ̀. Yí àwọ̀ àwòrán náà padà, yí irúgbìn tuntun ní ọjọ́ kẹrìndínlógún, lẹ́yìn náà yí i padà sí pósítà. Yí ìfọ́ àwọn àwọ̀ rẹ̀ padà sí ẹ̀yìn gígí náà ní ìṣọ̀kan.
7. "Gbogbo [ìjáde], gẹ́gẹ́ bí o ṣe rí i": fọ́ 120 nínú àwọn àwòrán mi jáde láti àárín lẹ́yìn ìkọ̀wé gíláàsì kan, lẹ́yìn náà kó wọn jọ sínú àmì mi bí orúkọ náà ṣe ń jáde. Fi "Jáde nísinsìnyí" sí ìsàlẹ̀ kí o sì di í mú fún ìṣẹ́jú méjì.

Rọpo ọpa ìbéèrè, awọn esi ati awọn sliders pẹlu awọn titẹ sii gidi, awọn abajade ati awọn eto ọja mi.

Ṣètò orin náà kí ìṣàlẹ̀ náà balẹ̀ lórí ìlà [Ìṣe], kí o sì gbé àkójọ orin náà síbẹ̀.
Fi ẹ̀gbẹ́ orin mẹ́ta bolẹ̀ lábẹ́ ohùn náà (ìwọ̀n tó kéré sí 30%, àárín 85%, afẹ́fẹ́ 55%; 0.3s dúró, 50ms wo-ahead, 0.5s release). Lẹ́yìn náà, ṣe àtúnṣe àárín gbólóhùn kọ̀ọ̀kan títí ohùn náà yóò fi dúró ní ìwọ̀n 9 dB lórí orin náà láàárín 300 Hz àti 4 kHz.
Lo Mixkit SFX kan ti a gba lati ayelujara fun iṣẹlẹ kan, ti a gbe kalẹ nipasẹ oke ti a wọn: titẹ kan fun iyipada aworan kọọkan, ohun pataki fun gbogbo lẹta ti a tẹ, ibalẹ rirọ lori ọrọ ti a tẹnumọ, whooshes, ati awọn ipa lori isalẹ ati aami. Gé gbogbo wọn ni ayika oke rẹ (ọpọlọpọ awọn riser jẹ iṣẹju-aaya 4). Pa orin naa run ni ọna ti o wa ni dB labẹ kaadi ipari. Gbọn-un si -14 LUFS.

1. Kanfasi HTML kan. Gbogbo fireemu jẹ iṣẹ mimọ ti akoko inu search(t), ati gbogbo akọle ati gige tẹle tabili ọrọ naa.
2. Gíláàsì: Ya àwòrán tí ó wà lẹ́yìn àwòrán náà, tàn án sí i ní ìwọ̀n 14px fún ara náà, ya àwòrán tí ó ṣókùnkùn díẹ̀ tí a gbé sókè ní ìwọ̀n 1.06x nínú ẹ̀gbẹ́ 12px ní etí rẹ̀, lẹ́yìn náà fi wàrà, dídán, ẹ̀gbẹ́ àti òjìji kún un.
3. Ṣe àgbékalẹ̀ pẹ̀lú Playwright ní 60fps nípa lílo àwọn subframes motion-blur 8, lẹ́yìn náà fi ffmpeg ṣe àgbékalẹ̀.
4. Kí o tó fi ohunkóhun hàn mí, ṣe ìwé ìfọwọ́sowọ́pọ̀ àwọn àwòrán, ṣe àyẹ̀wò ìyàtọ̀ fírẹ́mù fún àwọn àwòrán férémù kan ṣoṣo (àwọn àmì 16th nìkan ló lè fò), kí o sì ṣàyẹ̀wò ìpíndọ́gba ohùn-sí-orin fún gbogbo gbólóhùn.

Ohun èlò ìṣí [pẹ̀lú ìgbóná] lè dún bí ẹni pé a sọ ọ́ ní ohùn pẹ̀lẹ́pẹ̀lẹ́, nígbà tí [ìdùnnú] lè dàbí ohun tí kò ṣeé ṣe. Àpótí àmì tí ó wà lẹ́yìn ọ̀rọ̀ kan dà bí àṣàyàn ọ̀rọ̀ Windows. Tí kámẹ́rà bá tún padà sí 1.0 nígbà tí a bá ń fi nǹkan sílẹ̀, ìró ìró náà yóò yọ́. Ìparẹ́ 0.05s dà bí ìró ìró ìró lẹ́sẹ̀kẹsẹ̀.

Beere lọwọ mi fun awọn ohun ti o kọ silẹ, kọ iwe afọwọkọ naa, fi ohùn mẹrin ranṣẹ si mi lati yan ninu rẹ, lẹhinna fi awọn aworan 8 han mi ṣaaju ki o to ṣe afihan kikun.

Ask me for: my product name, a one-line pitch, my logo, 3 to 6 screenshots of the product, 40+ images of what it creates (or photos of it), my brand colors and font, an ElevenLabs API key and voice ID, plus a royalty-free track with a clear drop (the audio file and the drop timestamp). If I skip anything, use these defaults: "Ora 2", "an image model made for taste", free images from Unsplash and Pixabay, Geist, ink #111214, the voice Samara X (19STyYD15bswVz51nqLf) on eleven_v4, and Pixabay "Cascade Breathe" (129.83 BPM, drop at 75.39s).

<script>
Write a 9-line voiceover in this format and adapt every line to my product:
"This is [name]. A [category]... built for [value]. [Verb] in ANY style. Give it a [input] - and it just... understands. The more you use it, the BETTER it gets. Adjust its [setting]. Its [setting]. Its [setting]. Every [output] - exactly as you pictured it. [Name]. Out now."
Use emotion tags sparingly: [softly] on the quiet lines and [excited] on the [Verb] line. Don't tag the opener, and end it with a plain full stop.
</script>

Call the ElevenLabs API directly using my key, without an MCP. Generate 4 full-script takes in one pass and let me choose. If a line doesn't sound right, regenerate just that line 4 times and splice the best version into my take.
Shorten pauses longer than 0.28s to 0.2s, then bring each phrase 85% of the way toward the median level, adjusting gain only during pauses.
For word timings, split the read at pauses of 100ms or longer, transcribe each segment separately with Whisper, and align its first word to the measured onset. A single Whisper pass can place words up to half a second late.

A 22-second square launch film, 1440x1440 at 60fps, in the style of an AI lab launch: white canvas, one typeface, black ink, real images, and captions typed word by word at the exact time they're spoken.
Every control (prompt bar, settings panel, caption pill) uses liquid glass: frost the scene behind it, bend that scene along the edge like thick glass, then add a top sheen, bright rim and soft lift shadow. Glass disappears on plain white, so add a slow pastel aura in my brand colors or a blurred wash of the photo behind it.
The one emphasized word types in a gradient of my brand colors, with a faint glow behind it.
Motion rules: keep the camera drifting (push from 1.0 to 1.04 per scene; when content moves to the next scene, start at the zoom where the previous scene ended). Nothing appears instantly: use eased fades of at least 0.3s. Change images on 16th notes, with a click for each. Land scene changes on phrase starts, and align the music drop with the [Verb] line. Don't show full stops on screen.
Banned: selection-box highlights, beat-snapped slams, white flashes, 3D, templates.

Seven scenes, timed to the voice:

1. "This is [name]": a collage of my images drifts out as the name types in large, then gives way to the category line.
2. "made for [value]": full-screen flashes of my best images switch on 16th notes into the drop, with the line typed in white over them.
3. The drop: a glass prompt bar types a short prompt. The result card changes style every 16th note, while a black style chip rolls to each new name above a thumbnail strip.
4. "Show it a moodboard": 9 images fly into a 3x3 grid, then circle the result as "It gets you" types on screen.
5. "The more you use it, the better it gets": type two lines, with the emphasized word in the gradient.
6. The settings: show a glass panel with 3 sliders, each moving as she names it. Shift the image's hue, roll new seeds on 16ths, then turn it into a poster. Shift a blurred wash of its colors behind the glass in sync.
7. "Every [output], exactly how you see it": burst 120 of my images outward from the center behind a glass caption pill, then bring them together into my logo as the name types on. Put "Out now" underneath and hold for 2s.

Replace the prompt bar, results and sliders with my product's real input, outputs and settings.

Set the song so the drop lands on the [Verb] line, and anchor the beat grid there.
Duck the music under the voice with a 3-band sidechain (lows 30%, mids 85%, air 55%; 0.3s hold, 50ms look-ahead, 0.5s release). Then adjust the mids of each phrase until the voice sits about 9 dB above the music between 300 Hz and 4 kHz.
Use one downloaded Mixkit SFX per event, positioned by its measured peak: a click for each image change, a key sound for every typed letter, a soft landing on the emphasized word, whooshes, and impacts on the drop and logo. Trim each around its peak (many risers are 4 seconds). Fade the music along a dB curve under the end card. Loudnorm to -14 LUFS.

1. One HTML canvas. Every frame is a pure function of time inside seek(t), and every caption and cut follows the word table.
2. Glass: capture the canvas behind the shape, blur it about 14px for the body, draw a lightly blurred copy magnified about 1.06x in a 12px band along the edge, then add the milk, sheen, rim and shadow.
3. Render with Playwright at 60fps using 8 motion-blur subframes, then encode with ffmpeg.
4. Before showing me anything, make a contact sheet of stills, scan frame differences for single-frame pops (only the 16th-note runs may jump), and check the voice-to-music ratio for every phrase.

A [warmly] opener can sound whispered, while [excited] can feel fake. A highlight box behind a word looks like a Windows text selection. If the camera resets to 1.0 during a handoff, the zoom snaps. A 0.05s fade looks like an instant pop-in.

Ask me for the inputs, write the script, send me 4 voice takes to choose from, then show me 8 stills before the full render.

Àṣẹ tí a mú láti ifiweranṣẹ @rossaxbt.