Under the fan · prompt kit + generator
The AI ceiling fan transition is the TikTok outfit change filmed from the ceiling: you lie on a rug, a ceiling fan spins between the camera and you, and every blade that sweeps across the frame reveals a new outfit. It spread in late September 2026, often to Duran Duran's "Girls on Film". This page gives you a prompt that works in any AI video tool, the second each outfit should change, and theme prompts, and its generator makes the 15-second vertical clip from one face photo, with a theme pack or your own four outfits, from 23 credits.
Make a ceiling fan transition15 s · 9:16 vertical · 480p · looks change at about 5, 9 and 13 s
There is no prompt to write: the fan, the overhead camera and the blade timing are built in. Wan 3.0 is the default; Seedance 2.0 Mini at 480p is the cheapest test. Seedance 2.5 first checks your photos with the provider and charges nothing if they are refused.
The preview is a credited example made with the same reference shot, not your video. Yours keeps the overhead fan shot and the timing, with your face and the four looks you pick.
A one-shot outfit-change video seen from straight above. The person lies still on the floor, a ceiling fan turns close to the lens, and each blade crossing the frame acts as a wipe: when it has passed, the clothes are different while the face, pose and room are not.
Why it works: a blade covers the whole body for a few frames, so the cut is hidden. Every AI prompt for this trend is really about two things, keeping that camera locked and making the change happen only behind a blade.
Filmed or AI: creators first did it with a phone fixed to the ceiling and one take per outfit. The AI version starts from a single photo, so you need no fan, no tripod and no change of clothes, but the model has to keep your face across every look.
Other names: ceiling fan outfit change, under the fan, fan transition, fan outfit transition and fan blade outfit change. Hashtags usually say #ceilingfantransition or #ceilingfantrend.
| Question | Answer |
|---|---|
| The shot | One locked overhead camera, a person lying on a rug, a ceiling fan turning between them; a new outfit after each blade pass |
| When it spread | Late September 2026 on TikTok (The Tab wrote it up on 28 September 2026); the filmed version was around earlier in 2026 |
| Song | No single official sound; "Girls on Film" by Duran Duran is a common pick |
| Typical clip | 8–15 seconds, 9:16 vertical, three or four outfits |
| This generator | 15 s, 9:16, one person; four looks from a theme pack or your four outfit photos; the outfits change at about 5.4, 9 and 12.9 s |
| Models here | Wan 3.0 (default), Seedance 2.0 Mini, Seedance 2.0 Fast, Seedance 2.0, Seedance 2.5, MiniMax H3 |
| Cost here | 23 to 464 credits per video; Wan 3.0 at 480p is 67 credits |
| Watermark | Accounts without a purchase get a moving watermark; any plan or pack removes it, also from videos already made |
| Failed video | Credits come back automatically |
Three steps, in order, so you know what people mean when they say "the ceiling fan trend".
Early 2026
The filmed version: a phone points down from the ceiling, the creator records one take per outfit in the same pose and cuts each take where a blade covers them. An Instagram tutorial by @colinpomeroyyy from around January 2026 credits @jeffmachad0 for the idea.
Late September 2026
The AI version takes off on TikTok: one photo in, an overhead fan clip with several outfits out. The Tab explained it on 28 September 2026 as the "AI ceiling fan trend".
October 2026
Dozens of AI video sites add a ceiling fan template, and themed versions (autumn knits, Halloween costumes, wedding and night-out looks) take over the hashtag. Duran Duran's "Girls on Film" stays a frequent song, but there is no one official sound.
For Kling, Dreamina, Runway, Wan, Hailuo or any image-to-video tool. Attach one clear photo of the person as the first image, replace the four looks in brackets and keep the rest; every line is there to stop a specific failure (see the table below).
Vertical 9:16, 10 seconds, one continuous shot. The camera is fixed to the ceiling and looks straight down at the person from the reference image, who lies on their back on a light rug, arms relaxed, whole body in frame. A ceiling fan with exactly four matte blades turns slowly clockwise just below the lens, so each blade sweeps across the entire frame. The camera, room, light and pose never change. Keep the person's face, hairstyle, skin tone and body identical to the reference image in every frame. Only the clothes change, and only while a blade fully hides the body: 0–2.5 s [look 1]; 2.5–5 s [look 2]; 5–7.5 s [look 3]; 7.5–10 s [look 4]. Between blade passes the outfit stays still and sharp. Four rigid blades and one hub the whole time. Natural daylight, photorealistic fabric, no text.
camera movement, zoom, pan, tilt, cut, second fan, extra fan blades, bent or melting blades, outfit changing while visible, clothes morphing, two outfits mixed, face change, different person, hairstyle change, extra limbs, extra people, text, captions, logo, watermark
No negative-prompt box in your tool? Add the list at the end of the prompt as "Avoid: …".
Most broken results come from changes that do not line up with a blade. Split the clip into equal parts, one per look, and write those seconds into the prompt. A blade should be over the body at each change, so slow fans and four blades suit short clips.
| Clip length | Looks | Change at (seconds) | Each look on screen |
|---|---|---|---|
| 8s | 3 | 2.67 / 5.33 | 2.67 s each |
| 8s | 4 | 2 / 4 / 6 | 2 s each |
| 10s | 3 | 3.33 / 6.67 | 3.33 s each |
| 10s | 4 | 2.5 / 5 / 7.5 | 2.5 s each |
| 15s | 3 | 5 / 10 | 5 s each |
| 15s | 4 | 3.75 / 7.5 / 11.25 | 3.75 s each |
This generator does not split evenly: it follows its reference shot, where the person rolls onto the rug in the first 1.5 seconds and the looks change at about 5.4, 9 and 12.9 s of the 15-second clip, each one behind a blade.
If a result goes wrong, check the matching line before you change anything else.
| Line in the prompt | What it prevents |
|---|---|
| "The camera is fixed to the ceiling… one continuous shot" | Zooms, pans and cuts that break the illusion |
| "just below the lens, so each blade sweeps across the entire frame" | A small fan at the top of the frame that never covers the body, the most common failure |
| "exactly four matte blades… four rigid blades and one hub" | Extra, bent or melting blades, and a second fan appearing |
| "only while a blade fully hides the body" | Clothes morphing before the blade arrives |
| "0–2.5 s [look 1]; 2.5–5 s [look 2]…" | Looks out of order, skipped or changing twice |
| "Between blade passes the outfit stays still and sharp" | Two outfits blending into one |
| "Keep the person's face, hairstyle, skin tone and body identical…" | A slightly different face after each change |
| "lies on their back… whole body in frame" | Cropped feet, invented shoes and people sitting up |
Paste the four looks into the brackets of the prompt above. The first five also exist as one-click packs in the generator; the last four are variations for other tools.
Swap "light rug" for "dark wooden floor scattered with autumn leaves" and add "warm candle-like light".
Ask for "low golden afternoon sun from the window" so the rug and knitwear look warm.
Sequins and satin reflect light: add "soft, even lighting" so the fabric does not flicker.
Long dresses need room: keep "whole body in frame" and say the hem stays on the rug.
The safest first test: plain fabrics keep their shape best on cheap models.
Attach two photos and write "two people lie side by side, the person from image 1 on the left and the person from image 2 on the right; both outfits change behind the same blade; never swap or blend their faces". The generator here makes one person per video.
Here the room changes instead of the clothes: replace "Only the clothes change" with "Only the room and the outfit change to match the era".
Keep one outfit and write "Only the hairstyle and make-up change"; the face itself must stay the same person.
The person and the clothes stay; only the rug, props and light change on each pass.
The same idea needs different wording on each model family. These are the differences that matter for this trend, from building this generator on each of them.
| Model | How to pass your photo | Watch out for | Here (15 s, 480p or lowest) |
|---|---|---|---|
| Seedance 2.0 / 2.5 (Dreamina, CapCut) | Reference mode: name the photo in the prompt as @Image1 (Seedance 2.5 writes @图片1) and any motion clip as @Video1. | Seedance 2.5 may refuse real-person photos unless they pass the provider's check; Mini is cheap but softer on faces. | 117 credits |
| Kling | Image-to-video with your photo as the start frame, so an overhead photo of you lying down works far better than a selfie. | Put the negative list in its negative-prompt box; keep four looks in 10 seconds at most. | Not offered here |
| Wan 3.0 | Reference images (up to ten) plus an optional reference clip of up to 15 seconds; outfit photos can be added as extra images. | Follows a reference clip's fan and timing closely, which is why it is the default here. | 67 credits |
| MiniMax H3 (Hailuo) | Reference images (up to nine) and a reference clip; output at 768P or 2K. | Always writes its own audio; here it is replaced by the clip's music track. | 116 credits |
About two minutes of input, then a few minutes of waiting. No ceiling fan, tripod or editing app.
01
A clear, front-facing photo of one person in good light. A full-body photo helps the model get your build right.
02
Everyday, Autumn, Night out, Halloween or Formal gives four styled looks. Or choose My outfit photos and add four pictures of clothes in the order they should appear.
03
Wan 3.0 is the default; Seedance 2.0 Mini at 480p is the cheapest test. Keep With music, or pick Silent to add a trending sound on TikTok.
04
Download the MP4, then post it with the platform's AI-generated label switched on.
Every 15-second video, for every model and resolution, from the same function the generator charges. The outfit source and the sound do not change the price.
| Model · resolution | Credits | About (monthly plan) |
|---|---|---|
| Wan 3.0 · 480p | 67 | $2.49 |
| Wan 3.0 · 720p | 133 | $4.94 |
| Wan 3.0 · 1080p | 266 | $9.89 |
| Seedance 2.0 Mini · 480p | 23 | $0.86 |
| Seedance 2.0 Mini · 720p | 48 | $1.78 |
| Seedance 2.0 Fast · 480p | 48 | $1.78 |
| Seedance 2.0 Fast · 720p | 104 | $3.87 |
| Seedance 2.0 · 480p | 81 | $3.01 |
| Seedance 2.0 · 720p | 174 | $6.47 |
| Seedance 2.0 · 1080p | 436 | $16.21 |
| Seedance 2.5 · 480p | 117 | $4.35 |
| Seedance 2.5 · 720p | 262 | $9.74 |
| Seedance 2.5 · 1080p | 464 | $17.25 |
| MiniMax H3 · 720p | 116 | $4.31 |
| MiniMax H3 · 1080p | 185 | $6.88 |
Dollar figures use the cheapest monthly-plan credit ($0.037 per credit). That is a credit rate, not a per-video purchase. The $5 trial pack (24 credits) covers one Seedance 2.0 Mini video at 480p (23 credits). The Starter plan (240 credits for $9 a month) covers 3 Wan 3.0 videos at 480p.
| Question | Answer |
|---|---|
| Free credits | New accounts may get 5 signup credits; the cheapest video here is 23 credits. |
| Cheapest way in | The $5 trial pack (24 credits) covers one 23-credit video (Seedance 2.0 Mini, 480p). |
| Watermark | A moving site watermark on accounts without a purchase; any plan or pack removes it, also on videos already made. |
| Download | MP4 download for every finished video, with or without the mark. |
| Failed video | Credits come back automatically, also when Seedance 2.5's photo check refuses a photo. |
| 720p and 1080p | Priced per tier in the table; there is no separate HD export fee. |
Test a new photo or outfit set on Seedance 2.0 Mini at 480p, then make the one you post at 720p or higher. See plans and credit packs
This is the clip the generator follows, so it is the look you get: the roll onto the rug, the steel three-blade fan, the four blade-timed changes and the same music. It was made by Kavel AI, not on this site; the link goes to its source. Other public ceiling fan videos use different rooms, fans or songs, so they are not shown here.
Kavel AI's showcase clip, made from this exact reference shot: she rolls off the sofa, then a yellow sundress becomes a leather jacket, a white crop top with cream trousers and a white shirt set, each behind a blade.
Source: Kavel AIYour video keeps this camera, fan and timing, with your face and the four looks you choose.
They share the overhead fan shot, not the method. The filmed version (and most CapCut fan templates) joins several real takes at the moment a blade covers you. The AI version generates the whole clip from one photo, so the outfits can be anything, but the model has to keep your face. On TikTok both sit under #ceilingfantransition.
There is no single official sound. Duran Duran's "Girls on Film" is a common pick, and many creators use whatever is trending. Videos from this generator come with the reference clip's own music; choose Silent to add a song from TikTok's or Instagram's library instead.
The model was not told when to change. Write each change time into the prompt (see the blade-timing table), say the outfit changes only while a blade fully hides the body, and make the fan sit just below the lens so a blade really does cover you. This generator follows a reference shot whose changes already sit behind the blades.
State the blade count twice ("exactly four rigid blades and one hub") and lock the camera ("fixed to the ceiling, one continuous shot"), then add camera movement, zoom, cut, extra fan blades, bent blades and second fan to the negative prompt. Short clips with fewer looks also drift less.
For keeping one face across four looks, Seedance 2.5 and Wan 3.0 with a reference clip do best; Kling works well when your start photo is already taken from above. On this page Wan 3.0 is the default at 67 credits (480p), Seedance 2.0 Mini is the cheapest at 23 credits and Seedance 2.5 keeps the most detail.
Not with this generator, which makes one person per video. In a tool that takes several reference photos, use the couples prompt above: two people side by side, image 1 on the left and image 2 on the right, both outfits changing behind the same blade, and a line that forbids swapping or blending faces.
Use a template if you want the trend look quickly: the fan, camera and timing are already solved. Write your own prompt when you need something a template does not offer, such as two people, a room or hair change, or a particular model. The generator here sits in between: built-in shot and timing, with your own outfit photos if you want them.
Pick a theme pack or add your own outfits; the fan, the camera and the timing are already done.
More templates: Raindance AI · Tecktonik AI Dance · Wipe Me Down Dance AI · APT AI Video · Alors on Danse AI · Replace Person in Video · All templates
