I Love The Smiths AI Video: the 500 Days of Summer elevator scene, two photos, fifteen seconds, explained

An I Love The Smiths AI video is a 15-second clip in a small, wood-panelled office elevator: one person stands on the left in big over-ear headphones, the other stands on the right, hears music leaking from the headphones, and says “I love the Smiths” [official]. The listener does not hear, slides one headphone cup off one ear and says “Sorry?”, the speaker repeats the line louder, hums a little, and walks out through the opening doors [official]. The listener is left alone in the elevator, frozen, as the doors slide shut [official]. It is made from two photos — nothing is filmed.

The scene is the elevator meet-cute from (500) Days of Summer (2009), directed by Marc Webb [reported]. On Day 4 of the story, Tom (Joseph Gordon-Levitt) rides the office elevator listening to The Smiths’ “There Is a Light That Never Goes Out” on big headphones [reported]. Summer (Zooey Deschanel), the new assistant, hears it leaking out and tells him she loves the band; he does not hear her, so she says it again, sings a line of the song, and steps out [reported]. The film’s studio, Searchlight Pictures, posted the scene on TikTok on 1 February 2024, captioned “I LOVE THE SMITHS” [reported]. That clip had 5.7 million views when read on 9 November 2026 [reported] — a 5.5 million read on 6 November is the same clip three days earlier, not a conflict [reported].

The scene had been circulating for years before any AI version existed: Zooey Deschanel reenacted it on TikTok with Jonathan Scott in October 2021 [reported], a lyric edit passed 2.2 million views in July 2025 [reported], and a film edit posted on 29 September 2026 passed 1.4 million views [reported]. A real-life remake on a park bench on 21 October 2026 passed about 770,000 views [reported]. The AI wave is the new part: the earliest AI clip found went up on 5 November 2026, recasting the scene with a famous couple, and a guide to making the scene from two photos was published the same day [reported]. On 6 November an AI clip of a Formula 1 driver and his partner passed 480,000 views [reported]; on 7 November two AI kittens in the elevator passed 466,800 views with 20,000 shares — about one share for every 23 views [reported] — and tutorials for Kling and CapCut templates appeared, the biggest passing 1.4 million views [reported].

The trend also travels under the names “Tom and Summer elevator scene”, “500 Days of Summer elevator AI trend”, “elevator duet transition”, “elevator meet-cute meme”, “I love X / Me Too meme” and “I Love The Smiths split screen trend” [official]. Those are not seven trends — they are seven framings of one elevator scene, and this page covers all of them. The “elevator duet transition” name comes from Higgsfield’s template clip of the full scene [official]; the “I love X / Me Too” name is the same scene reused for shared taste — one person says “I love [a band, a show, a food]”, the other says “me too” [official]. The split-screen framing is two takes of the scene stacked side by side [reported].

The template wave moved fast. AI video sites added the elevator scene as a two-photo template on 5 and 6 November 2026 [official], and a trend radar first flagged the AI version on 6 November [official]. Videosi’s page went up on 8 November [official] and a second guide was published the same day [reported]. By 9 November there were at least seven dedicated template or guide pages — videosi, MakeViral, Vid2Prompt, SameVid, Oimi, Starrd and AI Effect Studio — plus further guides and template listings [reported]. This page ships on day six of the AI wave.

Every one-tap template that works hits the same beats in one wood-panelled elevator, framed from the waist up so both faces stay large and sharp [official]. The timings below come from the published template prompt that films the scene from your photos [official]; the beats, not the tenths of a second, are what to check when a generation breaks.

0–3 seconds — the set-up. A medium two-shot from the front. The elevator hums upward. The listener stares straight ahead and nods very slightly to the music; faint, tinny guitar music leaks from the headphones. The one who speaks glances over, hears it, and a small smile grows [official].

3–5 seconds — the line. A medium close-up of the one who speaks, turned three-quarters to the listener. Warm and a little shy, they say: “I love the Smiths” [official].

5–7 seconds — the miss. A medium close-up of the listener, who did not hear. They slide one headphone cup off one ear, puzzled, and say: “Sorry?” [official].

7–11 seconds — the repeat. The one who speaks, louder, with a playful, confident smile: “I said I love the Smiths.” Then, eyes half closed, they softly hum a short wordless tune and sway a little [official].

11–13 seconds — the exit. A soft bell rings. The doors slide open on a bright office hallway. The one who speaks gives a small sideways smile and steps out without looking back [official].

13–15 seconds — the freeze. The listener alone in the elevator, from the chest up, frozen, eyes wide, mouth slightly open, as the doors slide shut in front of them. They whisper: “Wow” [official].

There is a second, longer version to know about. The full-scene copies — Higgsfield’s “Elevator Duet Transition” template clip and the guides that copy it — run about 24 seconds, close to the film scene’s own ~25 seconds [official]. In that version the first shot runs 0–4.7 seconds and carries all three lines, the second shot runs 4.7–11 seconds with a sung line and an “I love them”, and the third shot, 11–15 seconds, is the wide shot of the doors opening and the walk-out [official]. The 15-second templates are an abridgement of that scene, not a different scene — check which length you are buying before you pay.

The joke is recognition. Everyone has had the moment of hearing their own taste leak out of a stranger’s headphones; the scene just stages it as a meet-cute, and the audience already knows the lines before the second one lands. The scene had been circulating on TikTok for years — the studio clip alone has 5.7 million views [reported] — so viewers get the joke in the first second and the AI version needs no explanation.

The template has two parameters — who listens, who speaks — and that pair is a variant machine. Any two people fit the same elevator, and so does any pet in the headphones seat [official]. The same mechanism as the fish-cutting and flamingo-float templates — one fixed scene, swappable subjects — with a higher bar: both faces fill the frame for the whole clip, so likeness matters more here than in a wide shot [official]. Fan casting does the rest: most of the AI clips put a pairing people already care about into the elevator, from pop stars to TV characters [reported].

The pet branch needs no film knowledge at all. Two AI kittens in the elevator — one in headphones, one in glasses — passed 466,800 views with 20,000 shares on 7 November [reported], about one share for every 23 views, the highest share rate in the trend’s top ten [reported]. And the “I love X / Me Too” reading turns the scene into a format: swap the band for anything you love and the line still works [official].

The honest clock: the first AI clip went up on 5 November, the template wave started on 5–6 November, and this page ships on day six. That is early — but seven dedicated template or guide pages already hold a version, and the biggest tutorial already passed 1.4 million views [reported]. The durable part is not this elevator, it is the mechanism: one famous two-line exchange, two faces, one swappable pair.

There are five routes, and they are not the same product. The one-tap templates film a new 15-second scene from your two photos; the full-scene versions copy the trend clip shot for shot; the in-app CapCut and Kling templates live inside those apps. All of them use the same two roles: photo 1 is the listener with the headphones, who stays in the elevator at the end; photo 2 is the one who says the line and walks out [official]. Swap the photos if you want the other role [official].

Route 1 — videosi, the cheapest verified dollar price (13 credits). Upload two photos, pick 9:16 for TikTok, Reels and Shorts (1:1 or 16:9 on request), and press Generate — the 15-second MP4 comes out with the film sound already on it and no watermark, ready in 4–6 minutes [official]. It is made with ByteDance’s Seedance 2.0 or 2.5, your choice of tier, from your photos plus a motion reference of the elevator cuts [official]. One video is 13 credits, and credit packs start at $6.90 for 10 credits and $17.99 for 30, with credits that never expire and a failed render refunded [official]. Note the pack maths: the 10-credit pack does not cover a 13-credit video, so the effective entry price is the $17.99 pack, which puts one video at about $7.80 — or about $8.97 at the smaller pack’s per-credit rate [official]. No subscription is needed, and you only need an account to pay and download [official]. The catch, stated on the page itself: the film sound is on the video, but mouths move loosely — they do not lip-sync the dialogue word for word [official].

Route 2 — MakeViral, the 189-credit template with the prompts published (four versions). Add the listener first, then the one who speaks — the order decides who is who — and press Generate; the button shows the price first [official]. Seedance 2.5 films the scene straight from your photos, with no in-between picture, for 15 seconds with original sound and the trend’s song under it, as a 9:16 MP4 with no watermark, ready in a few minutes [official]. Four versions share the same two photos: Two people (189 credits), Couple (189 credits), You + your pet (189 credits), and Full scene (228 credits), which copies the trend clip shot for shot in wide 16:9 with the scene’s own audio [official]. A failed video gives its credits back [official]. The dollar cost of a MakeViral credit could not be verified [unknown], so compare the 189 credits against videosi’s 13 and Starrd’s 150 in credits, not in dollars. This is also the route with the most published detail: the full video prompts are on the page, about 2,900 characters each, with a six-shot breakdown and a troubleshooting table [official].

Route 3 — Starrd, the $7.50 guide route (150 credits). Starrd’s template restages the scene with two photos: photo 1 stands on the left in large silver over-ear headphones, photo 2 stands on the right holding a stack of folders, smiles and says the line, and walks out through the opening doors [reported]. Fifteen seconds, vertical 9:16, 480p, with sound, for 150 credits — about $7.50 — and no prompt to write [reported]. Find it in the Comedy & Viral section [reported]. The guide also covers the do-it-yourself alternative: feed the clip, your two photos and a still to a character-swap model [reported].

Route 4 — the rest of the wave: Vid2Prompt, SameVid, Oimi, AI Effect Studio, and the guides. Vid2Prompt makes a 15-second 16:9 video with sound in about 4 minutes, from one or two photos, on Seedance 2.0 Mini at 720p with no watermark — and, unlike the routes above, with spoken lines and a made-up guitar tune instead of the Smiths song [official]. Its two-people and pets templates both work from a single photo, with the second person or pet generated [official]. No price for its route could be verified [unknown]. SameVid’s template turns one or two portraits into a 14-second elevator moment at 720p with original sound for 380 credits [reported]; its dollar price could not be verified [unknown]. Oimi’s template recasts one or two photos into the scene with lip-sync and full audio, no prompt, no watermark, at 720p [reported]; its price could not be verified [unknown]. AI Effect Studio makes a 20-second landscape 16:9 MP4 at 480p or 720p for 180 credits, with a typical wait of 3–10 minutes; photos are deleted after 7 days and videos after 30 [reported]. Its credit packs are listed at $9.99–$49.99, but the per-video dollar cost could not be verified [unknown]. Zooop published a guide on 8 November 2026 for a 15-second 16:9 version at 480p or 720p [reported], and a further guide puts its version at about 24 seconds in 16:9 [reported]; neither guide’s price could be verified [unknown]. Memelry lists the scene at $14.99 once, with original sound included [reported].

Route 5 — the CapCut and Kling templates in-app (the closest thing to free). On 7 November 2026, tutorials appeared for both a Kling template (@tacy1094) and a CapCut template (@zora.pansy_win) [reported]. The CapCut tutorial is the most-viewed tutorial of the trend at 1.4 million views [reported]. Those templates live inside the Kling and CapCut apps and change often, so check inside the apps [official]. What they cost inside those apps, if anything, could not be verified [unknown] — treat them as the free route only until a price is confirmed, and expect the in-app version to be a photo template rather than the full two-photo AI scene.

The prompt route, by hand. MakeViral has published its own template prompts — about 2,900 characters for the two-people version — and the condensed prompt below is our wording of their structure, not a copy of their text [official]. Give it to any image-to-video model that takes two reference photos, with the original scene as the reference.

Film the famous elevator meet-cute meme like a real film scene: a small, old office-building elevator with warm dark wood-paneled walls, brass trim, a brass handrail and brushed metal doors, under soft warm overhead light on a quiet morning. Photorealistic, natural skin and natural colors, gentle film grain, a warm cinematic grade, not CGI, not a cartoon. Vertical frame. The two stand side by side facing the doors, about an arm's length apart, framed from the waist up, so both faces stay large, sharp and well lit.

The listener is the person in [Image 1], on the LEFT in every frame: copy the exact face from [Image 1] (the face shape, eye color and eye shape, eyebrows, nose, mouth, skin tone, age, hair color, hair length and hairstyle, and any beard, moustache, glasses or freckles), with the clothes from [Image 1]. They wear large over-ear headphones with no logo.

The one who speaks is the person in [Image 2], on the RIGHT in every frame: copy the exact face from [Image 2] with the same details, with the clothes from [Image 2], holding a thin stack of folders.

If an image shows an animal, that one is that animal with its own face, fur and markings, not a person. Never blend their faces, never swap their clothes. Each face stays the same from the first frame to the last.

The shots, 15 seconds:
1. 0-3 s: a medium two-shot from the front. The elevator hums upward. The listener stares straight ahead and nods very slightly to the music; faint, tinny guitar music leaks from the headphones. The one who speaks glances over, hears it, and a small smile grows.
2. 3-5 s: a medium close-up of the one who speaks, turned three-quarters to the listener. Warm and a little shy, they say: "I love the Smiths."
3. 5-7 s: a medium close-up of the listener, who did not hear. They slide one headphone cup off one ear, puzzled, and say: "Sorry?"
4. 7-11 s: the one who speaks, louder, with a playful, confident smile: "I said I love the Smiths." Then, eyes half closed, they softly hum a short wordless tune and sway a little.
5. 11-13 s: a soft bell rings. The doors slide open on a bright office hallway. The one who speaks gives a small sideways smile and steps out without looking back.
6. 13-15 s: the listener alone in the elevator, from the chest up, frozen, eyes wide, mouth slightly open, as the doors slide shut in front of them. They whisper: "Wow."

Audio: the elevator hum, faint tinny guitar music from the headphones (an original melody), the two voices in natural English with exact lip sync, the soft hum, the bell and the doors. No other dialogue. Only these two, nobody else in the elevator or in the hallway. No text, no subtitles, no logos, no brand names, no watermark. No other music, no famous melody, no lyrics.

For the pet version, the same prompt applies with one change: the listener is the pet in [Image 1], sitting upright on a small wooden bench so its head is at the person’s chest height, with small over-ear headphones made for its head [official]. At the first line it tilts its head and one headphone cup slips off one ear, and it makes one small, puzzled sound of its own kind [official]. The pet never talks [official]. At the end it is left alone on the bench, frozen, eyes wide, ears up, staring at the doors as they slide shut [official]. Caption ideas from the template sites: “I said I love the Smiths”, “when someone finally gets your music taste”, or the me-too twist — “I love [your thing].” “Me too.” [official].

Sound: there is no single official sound for the trend, and the routes disagree. MakeViral plays the Smiths’ own recording under the lines [official]; videosi lays the film scene’s audio over the video [official]; Vid2Prompt uses a made-up guitar tune and no Smiths song [official]. The biggest tutorial, the CapCut one, uses “ghostin” by Ariana Grande [reported]. Pick the sound when you post, from the platform’s own library, and keep the platform’s AI-generated label on.

This is an early bet, not a proven search term. The AI wave is six days old and the template wave four days old. Google autocomplete returns no suggestions for the full “i love the smiths ai video” phrasing (checked 2026-11-11) [reported]. The trend is real; the English demand for the AI-video phrasing is not yet established. Do not read the hashtag numbers as demand for this page: the 317.5 million is years of film-scene traffic, not this week’s AI wave [reported].

The 317.5 million needs discounting twice. #ILoveTheSmiths held 317.5 million TikTok views when read on 9 November 2026 [reported] — but that tag covers every film-scene edit since 2024, including the studio clip at 5.7 million views and fan edits at 2.2 and 2.1 million [reported]. The AI-specific tag tells the smaller, honest story: #ilovethesmithstrend held 655 videos and 2.8 million views on 8 November 2026 [reported]. The other big tag, #500DaysOfSummer at 6.9 billion views [reported], is the whole film, not this scene. And the two reads of the main tag — 313 million on 6 November, 317.5 million on 9 November [reported] — are dated reads three days apart, not a conflict.

The durations disagree across routes. One-tap templates run 14–15 seconds [official]; AI Effect Studio runs 20 seconds [reported]; the full-scene copies run about 24 seconds, close to the film scene’s own ~25 seconds [official]; and the AI clips watched on TikTok run 21–24 seconds in landscape [reported]. The 15-second versions are an abridgement: they keep the lines, the exit and the stunned look, and cut the sung line and the wide shot [official]. If you want the whole scene, buy the full-scene version, not a 15-second template.

Most prices are credits, not dollars — and five dollar figures could not be verified. Only videosi and Starrd give a verifiable dollar figure: about $7.80 and about $7.50 a video [official] [reported]. MakeViral (189 credits), SameVid (380 credits) and AI Effect Studio (180 credits) state credits only [official] [reported], and no dollar figure could be found for Vid2Prompt, Oimi or Zooop at all [unknown]. At credit counts alone the spread is wide — 13 credits on videosi against 380 on SameVid — but credits are not comparable across sites until each site’s pack price is known. Compare the finished-video dollar price before you pay, not the credit number.

The song is not yours, and the routes do not agree on it. MakeViral plays the Smiths’ own recording under your lines [official]; videosi lays the film scene’s audio over the video [official]; Vid2Prompt uses a made-up guitar tune and no Smiths song [official]. None of them gives you rights to the audio — the template buys you a video, not a licence. Adding a sound from TikTok’s or Instagram’s own library when you post is the safer path.

The sites disagree on whose photo you may use — and the stricter rule is the safer one. Videosi’s page says a photo of a partner, a friend, yourself or a public figure works on its tool [official]. Vid2Prompt’s guide says the opposite: do not upload photos of celebrities or anyone who has not agreed to be in the video [official]. This page follows the stricter rule: only use your own photo, or a friend’s with their yes; never a celebrity’s, a stranger’s, or a child’s photo. Turn on the platform’s AI-generated label before posting, and never present the clip as a real elevator encounter.

The failure modes are face mixing, mumbled lines, and loose lip sync. Two faces share the frame for the whole clip, so they can mix or swap sides — the fix the template sites give is one face per photo, with photo 1 always the listener on the left, and run it again [official]. A line can come out cut short or mumbled; every run is a new take of the scene [official]. On videosi the mouths move loosely and do not lip-sync the dialogue word for word [official]. On the pet version, the model can squash your pet or make it too small — use a photo where the pet sits and faces the camera, with the whole head in view [official]. In every case a failed video gives its credits back on the routes that state it [official].

Five prices could not be verified. No dollar figure could be found for MakeViral, SameVid, AI Effect Studio, Vid2Prompt, Oimi or Zooop [unknown], and no price at all could be found for the in-app CapCut and Kling templates [unknown]. They are marked, not smoothed over.