Back to videos
O

OpenAI

Introducing ChatGPT Images 2.0 A state-of-the-art image model that c...

Launch
Platform
x
Length
1:00
Views
12.9M
Median cut
0.77s
Face on screen
0%
Dark frames
0%

62 steals on the timeline

0:00 / 1:00

x-bookmarks

Word-Object Montage on White

medium effort

Steal this
Put the words inside the world instead of on top of it: every line of copy is a practical object in a real-looking scene, and the only graphic layer is the white canvas you reposition each image on.

Sixty seconds of roughly forty-two discrete micro-shots, each one a self-contained image in which a single word is built into the physical world of the frame (printed on a clock face, spelled in cereal, on a road sign, stamped on a toy badge). The shots are composited onto a plain white canvas at wildly varying sizes, aspect ratios and screen positions, so the frame itself becomes a layout tool rather than a fixed window. A recurring green chameleon character appears in nearly every scene as the visual thread that stops the montage feeling like a stock reel, and the whole thing resolves into a centred product lockup.

Opening move

A clean, sunlit wide of a suburban house with a chameleon-shaped topiary sits letterboxed on white, held long enough to read as a normal establishing shot. It sets the naturalistic, camera-real tone and plants the character silhouette, so the first hard cut to a macro clock face at ~0.8s lands as a jolt rather than a continuation.

Pacing

Relentless and even: 42 detected shot changes across 60s, median 0.77s per cut, which is close to one image per beat of a mid-tempo track. Hold the first shot noticeably longer (2s or so) and the final lockup longer again (3s+), and keep everything between at 0.6-1.0s. Vary shot length by no more than a third inside the body so the rhythm reads as machine-precise; use the mosaic frames and the empty-white two-up as your only rhythmic punctuation.

Graphics and type

There is no overlay typography at all in the body — every word is rendered as a physical object inside the shot (engraved, printed, spelled in food, painted on signage). All the graphic design happens in the canvas: pure white background, images placed as hard-edged rectangles with no drop shadow, no stroke, no rounded corners, at freely changing scales, aspect ratios and positions. Each source image keeps its own grade — bright daylight, tungsten night, pastel nursery, high-contrast studio — and the white surround is what unifies them. End card uses a thin geometric grotesque in near-black, set flush at the left and right thirds, generous letterspacing, image centred between.

Shot list
  • Sunlit exterior wide, house frontage, letterboxed 16:9 inside white frame
    Establishing beat and character plant; slowest shot in the piece so the eye can settle before the cutting starts
  • Extreme macro, clock dial filling frame edge to edge, high-key white with saturated green hands
    Scale whiplash after the wide; first example of a word living inside an object
  • Top-down flat lay, cereal bowl centred on grey worktop, spoon on the diagonal
    Changes camera axis to overhead so consecutive cuts never repeat the same geometry
  • Medium exterior, storefront sign board with subject perched on top, daylight, right third occupied by background signage
    Puts the word at true optical centre while the environment carries colour
  • Night medium, brown road sign lit by a hot practical flare, black surround
    The one dark-value beat; contrast valve in an otherwise bright reel
  • Multi-image mosaic: five images of different aspect ratios tiled with white gutters
    Compresses several beats into one frame; acts as a rhythmic accelerando
  • Asymmetric two-up: large image top-left, small image mid-right, lots of empty white
    Uses negative space as a beat of silence; deliberate off-balance layout
  • Environmental medium, child shot from behind at an aquarium, hand reaching in
    Human presence without ever showing a face, keeping attention on the object and the character
  • Stylised dark portrait-mode setup, hard top light, subject in costume, near-black surround
    Genre-shift shot; proves the format can absorb totally different lighting looks
  • Cut-out subject on flat sky, pushed hard into the right third of a white canvas
    Extreme composition offset; makes the following centred shot feel like an arrival
  • Macro, glitter badge on a nursery rug, shallow depth, pastel palette
    Texture and tiny-type payoff shot near the end
  • Centred small image with black wordmark set left and right of it on white
    Product lockup; the layout logic of the whole video resolved into a title card
How to shoot it
  1. Write your line first, then assign one word per shot; you need roughly one word per 0.8s of runtime.
  2. For each word, invent a physical object that could plausibly carry it in a real scene — dial, sign, badge, food, packaging — and note the lighting world that object lives in.
  3. Pick one recurring character or prop and place it in almost every shot; this is the only continuity device, so do not skip it.
  4. Generate or shoot each beat as a separate clean image or 1-2s clip. Vary axis deliberately: wide, macro, overhead, behind-shoulder, cut-out.
  5. Grade each beat to its own world rather than to a single house LUT — the white canvas will do the unifying.
  6. Build a white solid as the bottom layer of the timeline and place every asset on top as a rectangle, never full-frame.
  7. Change scale and position on every cut. Include at least one mosaic of 4-6 tiles and one deliberately empty asymmetric layout as rhythm breaks.
  8. Cut on a fixed grid at 0.77s. Do not use transitions; hard cuts only.
  9. Hold the opening establishing shot about 2s and let the last shot run 3s+ into the lockup.
  10. Finish with the product name set in a thin grotesque, flush left and right of a small centred image, on the same white.
Kit
  • Generative image/video tool for producing the individual word-in-scene beats (or a props kit plus label maker if shooting practically)
  • Camera capable of macro if shooting for real: 60-100mm macro plus a 24-70 for the wides
  • One hard key with a bounce card — most beats are single-source, high-key daylight
  • Small LED for the single night/practical-flare beat
  • Timeline editor with precise transform controls (scale/position keyframes) and a white solid as base layer
  • Beat-grid or metronome reference to lock the 0.77s cut cadence
  • Consistent character asset or prop so continuity survives 40+ shots
Not knowable from stills
  • All audio: whether there is voiceover reading the words, music, or diegetic sound design cueing the cuts
  • Whether the cut cadence is locked to a music beat
  • Whether shots contain internal camera or subject motion, or are stills with slow pushes
  • Whether assets were generated or practically shot and retouched
  • Exact typeface of the end card and whether it animates in

Written by anthropic/claude-opus-5 from the frames and measurements above. Technique only, and it cannot hear the video.

All frames 62Opening 1Shot changes 42Presenter 0