A launch film with motion graphics

opus 5.5 is f*cking goated at motion graphics - made this entire launch film for $0 - no external tools just a prompt everyone else posting "ai motion" content gatekeeps behind a paywall or an mcp install not this one steal the prompt below and run it in your own claude ↓ <inputs> Before you start, ask me for these: a single-word brand name for the wordmark (verbs work best), 9-12 high-resolution photos, a royalty-free track near 120 BPM that has a drop and a calm breakdown section (Mixkit, cleared for commercial use), and one free stock clip of a bare wall with shifting plant-shadow light (Pexels). </inputs> <direction> Think Apple keynote reel, flat 2D, one unbroken shot. Nothing ever cuts or fades. Each scene is literally built from the shape of the one before it. Type rises up out of a mask edge, icons snap in from nothing on a spring, bars stroke across, pages get shoved aside, and a solid black shape floods everything then shrinks down into whatever comes next. Palette: warm off-white background, black interface elements, iOS 26-style liquid glass laid over the photography. Wordmark font is Archivo at width axis 125 / weight 800; interface font is Geist. A visible cursor performs every interaction (real clicks, drags, long-presses), and the whole camera does a screen-studio zoom into each moment, scaling the cursor along with it. Do not use: crossfades, blur-to-reveal, brightness "developing" transitions, 3D flips, particle effects, glow effects, any hold over 1 second, or anything that reads as a stock template. </direction> <structure> Track runs 120 BPM across 54 beats, and every single beat triggers a visible change. Opening: the wordmark contracts inward like an accordion until it's swallowed by its own period, which balloons into a black pill; a label lifts up inside that pill. A click closes six iris blades over the label, then they snap open onto the first photo. That circle flattens to a square, shrinks, and a grid unfolds out from behind it like a map unfolding (center piece, then a plus shape, then the corners) before reflowing into a bento layout. One more click zooms into a single tile, timed exactly to the drop. Glass segment: the word "glass" types in letter by letter, liquefies into a droplet, then stretches into a full glass toolbar. Tapping the adjust icon turns it into a slider; dragging it relights the photo from daytime to golden hour using two matched exposures, and while held, the slider knob becomes a glass lens. That lens lifts into a floating glass orb; the next photo appears inside it as a circular crop; the orb then expands outward into a full lock screen: glass-style clock digits, a date line, and a home indicator bar that stretches into a glass music player. Device segment: the lock screen pulls backward to reveal a phone body, its bezel growing outward from the screen's own edge. The Dynamic Island stretches like liquid, splits off, travels across the frame and grows into a full Mac window that unrolls like a window blind. Long-press the desktop wallpaper, drag it into a Safari tab (the page shoves in behind it), release, and it becomes a landing page's hero image. On scroll, that hero image folds itself into a framed print inside a product card, the mat board and molding growing directly out of the photo's border. Choose a frame color (it paints across live) and a size, then the nav button collapses down into an "Order print" button. Order sequence: a single black shape keeps reshaping itself through four states: Ordered (checkmark) → Printing (percentage) → Out for delivery (a van tracing a route) → Delivered (checkmark). Closing: that final "delivered" circle expands to fill the entire frame edge-to-edge, holds on black for one beat, then contracts down into the framed print now hanging in the real wall-shadow footage. The same iris opens onto the print, closes again, and that black frame floods the screen once more before contracting all the way back into the pill, then the dot, then the letters spring back out into place, landing exactly on the beat where the track loops. Final frame matches the opening frame exactly. </structure> <build> 1. Single HTML file, 1440×1440 square canvas. Everything is computed live from an async seek(t) function based on elapsed time. No CSS transition properties, no setTimeout/setInterval, no carried-over state between frames. 2. Any animated value that needs to hit several targets in sequence is built as a sum of individual spring functions (closed-form, not simulated step-by-step), so the whole thing stays a pure function of t. 3. For the liquid glass look: each glass panel renders its own copy of the scene sitting behind it, then distorts it via an SVG feImage using a rounded-rectangle signed-distance field, run through three feDisplacementMap passes at slightly offset scales (this is what gives the chromatic-edge look), plus a rim-light pass on top. For glass lettering specifically, generate a per-glyph distance field on a canvas and use that for the displacement map, the mask, and the highlight. 4. For the melt/goo effect: apply blur, then threshold the alpha channel, then composite the un-blurred source back on top so glass edges stay crisp while the body stays fluid. 5. Iris wipe: build it as six blades arranged around a hexagon. Each blade is defined by its two vertices, the extension of both edges, and the short arc connecting them. 6. Wordmark contraction: every letter travels toward the central dot by an identical ratio, and its rendered width shrinks in sync (tighten the font's width axis, scale everything else proportionally) so letters never separate from one another. 7. Video handling: re-encode every clip as all-intraframe (ffmpeg -g 1), load via blob URL, and always await the 'seeked' event before drawing a frame. 8. Audio: source a real downloaded sound effect for every single event from Mixkit, never generate audio synthetically, and place each one at its measured waveform peak. Align the song so it starts right on a downbeat, the zoom-in lands exactly on the drop, the black-flood section sits in the breakdown, and the wordmark's return syncs to the beat. Normalize loudness to -14 LUFS. 9. Rendering: use Playwright, capture 4 sub-frames per output frame and blend them with ffmpeg's tmix filter at 60fps. Spot-check one frame per beat manually, then run a scan comparing frame-to-frame differences to catch single-frame pops (anything spiking to 3x its neighbors). </build> <gotchas> Chromium's backdrop-filter: url() handles displacement maps incorrectly; clone the background scene manually instead of relying on it. Any flood transition needs to overshoot past the frame's corners and take roughly 0.3 seconds, otherwise half the screen updates in a single frame, which reads as a pop. A child element set to visibility: visible will still render through a parent that's hidden; use visibility: inherit instead. Any text that swaps mid-morph needs its own dedicated mask. Don't serve the video with python's http.server, since it can't handle range-seeking; use a blob URL instead. </gotchas> <start> First ask me for the four inputs above. Once you have them, show me the beat-by-beat map plus four still frames (opening, glass, device, wall) before you build the full film. </start>