1Start a widescreen project
Pick Widescreen first: the shape decides how every picture is framed.
Editor → Start a New Video
Shape: Widescreen (16:9, the shape of a TV or YouTube player)FreeA tall picture cropped to widescreen loses most of itself.
A pop singer, singing and dancing through one night in the city, from golden hour on a rooftop to sunrise on the same roof — and none of it happened. There was no singer, no city and no camera. It all came from one AI-generated picture, made on OpenMuse, step by step below.
Every prompt exactly as it was typed, in the order we made it. Copy any one and change the person, the places or the song.
What it costs
One £50 Pro pack covers the whole film — 6,000 tokens.
8 steps
Start a widescreen video, then make the two things everything else is made from.
Pick Widescreen first: the shape decides how every picture is framed.
Editor → Start a New Video
Shape: Widescreen (16:9, the shape of a TV or YouTube player)FreeA tall picture cropped to widescreen loses most of itself.
Frame it waist-up and name nothing below the waist: every scene in the film starts from this picture.
Create with AI → Create an Image
Image model: Nano Banana ProSize: 2K (about 2,000 pixels wide)Shape: 16:9Reference: none96 tokens32 tokens a picture. We spent 96 on three tries.
Describe someone you would watch for two and a half minutes: the face in detail, one outfit, a plain backdrop, nobody else in the frame.
It took three tries. The first two named her trousers and trainers, and both came back full length with a small face: the model shows whatever you name. The one that worked puts the framing first and stops at the waist. The trousers and trainers come back in the scene prompts.
Using your own photo? Everything after this step works the same.
The one picture
No reference picture
Waist-up portrait photograph, framed from her waist to just above the top of her head, of a pop singer: a Filipina woman in her mid-twenties, an oval face with high rounded cheekbones and a small pointed chin, glossy black hair cut in a sharp chin-length bob with a blunt fringe, warm dark-brown eyes, and a small beauty mark high on her left cheekbone. She wears a cropped silver sequin jacket with the sleeves pushed up to the elbow over a white ribbed tank top tucked into high-waisted black trousers, and small gold hoop earrings. Her face is large and sharp in the frame. She stands square to the camera and looks straight into the lens with a relaxed, confident half-smile, lips closed, centred in the frame with open space on both sides. A plain seamless studio backdrop in warm tangerine orange fills the whole frame edge to edge, with no lights, stands, reflectors or studio equipment visible anywhere. One large soft key light from the front left and a gentle rim light catching the edge of her hair and the sequins. Real photographic skin with visible fine texture and pores, healthy and clear, never airbrushed or plastic. She is alone; there are no other people anywhere in the frame. Photographic style, shallow depth of field. No text, no letters and no logos anywhere in the frame.
Write a brief, not lyrics, and let the switches set the length: ours came back at 2:35.
Your Song → Write Lyrics, then Create Your Music
Lyrics: Verse & chorusGenre: PopMood: HappyLength: MinimalTempo: Fast64 tokens2 tokens for the lyrics, 30 for the song. We spent 64: we made the song twice.
No lyrics to write. The brief says what the song is about, which part of the story each section tells, and two words the chorus must use; the app writes the words.
Our first song ran 3:51 (Length Standard, Tempo Medium, a bridge). Length Minimal, Tempo Fast and "No bridge." in the brief brought it to 2:35.
On Create Your Music, say "solo female lead vocal, no backing vocals" outright: one voice that matches her face is one voice for her mouth to follow (lip-sync).
1 of 2The brief, typed into "What's your song about?"
An upbeat pop song sung by a young woman about one perfect night out in the city, a night so good that nobody will ever believe it happened — she sings it with a wink, as if it was too good to be real. Verse 1 is the rooftop at golden hour and the street as the shop lights come on; verse 2 is dancing alone in the all-night laundromat; the last chorus is the same rooftop at sunrise. No bridge. The chorus is a playful hook about the night that never happened, and the chorus's first line uses the words "never happened". Keep it one voice: no lines that call a crowd in. Keep every line concrete and visual; no lines about dreams, journeys or shining.
2 of 2The sound, typed into Sound details
Disco-pop at 124 BPM in a major key — the hook is a rubbery octave-jumping bass line under a four-on-the-floor kick, with a sparkling glockenspiel doubling the chorus melody note for note; offbeat open hi-hats, handclaps on 2 and 4, bright clean funk guitar chops. Solo female lead vocal, bright and playful, no duet, no male voice, no choir, no backing vocals, no gang shouts, no "hey" chants. Quick conversational phrasing, words close together, driving in eighths, never dragging, no long held notes. Short intro, no long instrumental breaks. Punchy, dry, modern mix.
Fourteen scenes from the one picture, a moving clip of each, and five takes of her singing.
Fourteen places, one face: each scene is a new picture made from the one picture, checked before you pay for it to move.
Create with AI → Edit a Photo
Image model: Nano Banana ProSize: 2KShape: 16:9Reference 1: the one picture, every timeReference 2: an earlier scene, for the place (6 of 14)608 tokens32 tokens a picture, 14 kept. We spent 608 on 19 tries.
A reference is a picture the model copies from; Edit a Photo takes two, in slots 1 and 2. Three rules kept her the same person in fourteen places:
Slot 1 is always the one picture
Never a scene made from it. Copies of copies drift: the face changes a little each time until she is someone else.
Slot 2 is only for the place
Put the earlier scene in slot 2 and say so in words: "Use the second reference only for the place; take nothing about the woman or the light from it." That scene has her face too.
Write the outfit out in full
In every prompt. A reference only carries what it shows, and the one picture stops at the chest.
Check the picture before it moves
A picture costs 32 tokens; a moving clip 75 or 150. Our rejects (fake letters on a neon sign, a real shop sign through a window, a stranger behind her) were caught as pictures, not clips.
Rooftop close-up, golden hour · 04
Became the verse-one take
From the waist up on the same rooftop at golden hour, facing the camera straight on, centred with generous headroom, the low sun behind her making a warm rim around her hair, the skyline wide and out of focus behind her.
Use the woman in the first reference photo: keep her face, her hair and her features exactly as they are in that photo. Use the second reference only for the place (its buildings, skyline and surfaces); take nothing about the woman or the light from it. From the waist up on the same rooftop at golden hour, facing the camera straight on, centred with generous headroom, the low sun behind her making a warm rim around her hair, the skyline wide and out of focus behind her. She is a Filipina woman in her mid-twenties, an oval face with high rounded cheekbones and a small pointed chin, glossy black hair cut in a sharp chin-length bob with a blunt fringe, warm dark-brown eyes, and a small beauty mark high on her left cheekbone. She wears a cropped silver sequin jacket with the sleeves pushed up to the elbow over a white ribbed tank top, and small gold hoop earrings. Her lips are softly closed, her eyebrows relaxed, the warmth in her eyes. Real photographic skin with visible fine texture and pores, healthy and clear, never airbrushed or plastic. She is alone; there are no other people anywhere in the frame. Photographic style, shallow depth of field. No text, no letters and no logos anywhere in the frame.Every scene becomes a moving clip. Count the dance, "twice every second", and keep every mouth still.
Create with AI → Make a Photo Move
Video model: Kling v2.5 ProSingle-photo clips: Motion Light Play, Face Allow changes, Subject GenericGentle gestures: off for the counted dancesLast shot: Join Two Photos, no fine-tuning1,650 tokens15 tokens a second. 14 clips, 110 seconds: 1,650 tokens. None was made twice; one was cut from the film.
Light Play adds only a gentle shift in the light, so your words decide what moves. Allow changes and Generic leave out the sentences that hold her face and body still, which fight a dance. Join Two Photos makes one clip that travels from one picture to another.
Count the dance
Turn Gentle gestures off (it adds a line asking for gentle movement) and give the tempo as a count: "twice every second". Compare the studio sway and the studio dance below. The clip can't hear the song; the cut puts the moves on the beat.
Watch every mouth
Every single-photo prompt ends by holding her face, outfit and mouth; only the singing takes move her mouth to words. The rooftop wind clip moved her lips anyway, so we cut it.
End where you started
The last shot travels from the studio dance picture back to the one picture, so the film ends on the frame it began with.
1 of 4Studio sway · 02a · 10s
Slow sway: no count, Gentle gestures on
She dances on the spot: shoulders bouncing, hips swaying, small steps side to side; the camera holds still.
She dances on the spot: shoulders bouncing, hips swaying, small steps side to side; the camera holds still. Her face and features stay exactly the same and she wears the same outfit throughout; the background holds its shape. She keeps the same open smile throughout, her mouth still; she is not singing or talking.2 of 4Studio dance · 02b · 10s
Counted, Gentle gestures off
She dances fast and punchy: she bounces on the balls of her feet twice every second, her shoulders shimmying and her hips swinging with each bounce, both arms held out wide at her sides, the sequins flashing; the camera holds still.
She dances fast and punchy: she bounces on the balls of her feet twice every second, her shoulders shimmying and her hips swinging with each bounce, both arms held out wide at her sides, the sequins flashing; the camera holds still. Her face and features stay exactly the same and she wears the same outfit throughout; the background holds its shape. She keeps the same open smile throughout, her mouth still; she is not singing or talking.3 of 4Rooftop wind · 03 · 10s
The warm wind lifts her hair and the edge of her jacket; the camera pushes in slowly; the skyline stays still.
The warm wind lifts her hair and the edge of her jacket; the camera pushes in slowly; the skyline stays still. Her face and features stay exactly the same and she wears the same outfit throughout; the background holds its shape. Her lips stay closed; she is not singing or talking.4 of 4Back into the one picture · Z · 5s
Join Two Photos: no fine-tuning
She stops dancing, turns square to the camera and settles into a still, relaxed stance with a closed-lip half-smile as the camera pushes in to frame her from the chest up; the orange backdrop holds.
A take is one 20-second clip of her singing a chosen stretch of the song. One take per sung part: two verses, two choruses, the outro.
Create with AI → Sing My Photo
Take length: 20 secondsPicture: a close-up, lips closed600 tokens6 tokens a second. Five 20-second takes: 600 tokens. Each was ready in about a minute.
Pick a close-up with her lips closed, then drag the 20-second highlight over the part she sings; it snaps to the start of a line.
Ask for energy as a counted move, as with the dances. All five takes used the same direction:
She sings straight to the camera with a playful wink, nodding her head on every beat, her shoulders moving with the rhythm.
1 of 5Verse one
Made from 0:01.9–0:21.9 of the song · in the film: one shot, 0:02.1–0:21.0
2 of 5First chorus
Made from 0:24.9–0:44.9 · plays at 0:25.2 and 0:40.3
3 of 5Verse two
Made from 0:54.8–1:14.8 · plays at 0:54.8 and 1:09.8
4 of 5Second chorus
Made from 1:21.2–1:41.2 · plays at 1:21.4 and 1:34.5
5 of 5Outro
Made from 1:47.2–2:07.2 · plays at 1:47.2 and 2:02.0
Cut every shot to the song, then make the video.
Put every singing shot back over the words it was sung to. It is arithmetic, not dragging.
Shape Your Video
Clip sound: mutedEvery join: Hard cutFreeFree.
The timeline is the row of shots in the order they play: 30 here, for 2:35.
Keep dance shots to two to five seconds. AI arms and legs wobble over longer stretches, and a cut on the beat hides it.
Set every join to Hard cut: one shot ends, the next begins, no fade. Add clips puts a 1-second fade on each join, and a fade pulls everything after it earlier, out of sync.
Type each shot's Duration from the Seconds column below. Dragging is never exact.
For a singing shot, set Start at to its song time minus the moment its take was made from.
Check it in Organize → Line up with the sound: it shows which words play over each clip.
Worked example
The first chorus plays its second piece at 0:40.3. That take was made from 0:24.9, so its Start at is 15.4 seconds.
Cut on the beat
We measured the song at 128 beats a minute, not the 124 the sound details asked for, and cut every shot on a beat.
Swipe to see more →
| Song time | Shot | Seconds | Where |
|---|---|---|---|
| 0:00.0 | The one picture, moving | 2.1 | studio |
| 0:02.1 | Sings verse one | 18.9 | rooftop, golden hour |
| 0:21.0 | Street walk | 4.2 | street, blue hour |
| 0:25.2 | Sings the first chorus | 7.5 | studio |
| 0:32.7 | Shop-window glance | 3.8 | shop window |
| 0:36.5 | Street walk | 3.8 | street |
| 0:40.3 | First chorus, continued | 4.7 | studio |
| 0:44.9 | Studio dance | 3.6 | studio |
| 0:48.6 | The one picture, moving | 2.9 | studio |
| 0:51.5 | Studio dance | 3.3 | studio |
| 0:54.8 | Sings verse two | 7.5 | neon street |
| 1:02.2 | Laundromat aisle dance | 3.7 | laundromat |
| 1:06.0 | Drumming on the washing machine | 3.8 | laundromat |
| 1:09.8 | Verse two, continued | 4.6 | neon street |
| 1:14.4 | Laundromat bounce | 3.8 | laundromat |
| 1:18.2 | Laundromat aisle dance | 3.3 | laundromat |
| 1:21.4 | Sings the second chorus | 7.5 | studio |
| 1:28.9 | Underpass dance | 2.8 | underpass |
| 1:31.7 | Laundromat aisle dance | 2.8 | laundromat |
| 1:34.5 | Second chorus, continued | 6.5 | studio |
| 1:41.0 | Underpass dance | 1.9 | underpass |
| 1:42.9 | Night bus, eyes closing | 4.3 | night bus |
| 1:47.2 | Sings the outro | 7.4 | rooftop, sunrise |
| 1:54.6 | Sunrise dance | 4.6 | rooftop, sunrise |
| 1:59.2 | Studio dance | 2.8 | studio |
| 2:02.0 | Outro, continued | 5.1 | rooftop, sunrise |
| 2:07.1 | Sunrise dance | 4.9 | rooftop, sunrise |
| 2:12.0 | Walk to the roof edge | 10.0 | rooftop, sunrise |
| 2:22.0 | Studio sway | 9.0 | studio |
| 2:31.0 | Back into the one picture | 4.6 | studio |
The price shows before you press anything: 156 tokens for 2:35.
Make Your Video
Size: 1920×1080 (Full HD)No watermark with any token pack312 tokens1 token a second: 156 tokens. We spent 312: we made it twice.
It took a little over two minutes.
We made it twice. In the first, the rooftop wind clip moved her lips out of time with the song; we took it out and let verse one play unbroken, 19 seconds from 0:02.1.
| Item | Tokens | Pounds |
|---|---|---|
| The one picture3 tries × 32 | 96 | under £1 |
| Lyrics2 × 2 | 4 | under £1 |
| The song2 × 30 (the first ran 3:51) | 60 | under £1 |
| Scene pictures19 tries × 32, 14 kept | 608 | about £5 |
| Moving clips110 seconds × 15 | 1,650 | about £14 |
| Singing takes100 seconds × 6 | 600 | about £5 |
| Render2 × 156 | 312 | about £3 |
| Total | 3,330 | about £28 |
Three-quarters of the cost is motion and singing, so fewer clip seconds is the lever that matters.
Every scene was made from the original picture, never from a scene made from it. Copies of copies drift; the original does not.
To match an earlier scene, put it in slot 2 and say in words to take only the place from it. That scene has her face in it too, and a second face is a second chance to drift.
"Twice every second" made a dance; "dances on the spot" made a slow sway. And turn off Gentle gestures — it asks for gentle.
Every other clip says her lips stay closed. One moved them anyway, and a mouth out of time with the song looks broken. Watch every mouth.
People expect an AI music video to look like a music video: the singer actually singing, actually dancing, in more than one place, cut to the beat. What they usually get is a slideshow of pictures with a song behind it, because keeping one face the same across thirty shots, making every shot move, and putting the mouth on the words are three separate problems. So we made one from a single picture and wrote down every prompt. Nothing on the timeline is a still, the same woman is in every shot, and whenever she sings on screen, her mouth is on the words.
Some moments this fits — there are more.
A video for a song before there is a budget for a shoot.
Show a label, a client or a friend what the video could be.
A made-up singer who stays the same person in every shot.
Start from a picture of yourself and sing your own song.
Describe what the song is about; the words and the music are made for you.
Widescreen 1920x1080, music licensed for commercial use, no watermark with any token pack.
One real project shows how the pieces fit faster than any tour.
A few small things make a noticeable difference.
Describe trainers and the model shows trainers, whatever the framing says. Put the framing first and leave out anything below the frame.
AI text comes out as gibberish. Ask for blank sign boards and neon as shapes of light, and say "no text, no letters and no logos".
Two to five seconds, cut on the beat. AI arms and legs wobble over longer stretches, and the cut hides it.
A fade overlaps two clips and pulls everything after it earlier, so every singing shot after it falls out of sync.
Sing My Photo opens the mouth itself. A closed, relaxed mouth in the picture gives it the cleanest start.
Make the one picture at 16:9, then the song from a short brief. Set Length and Tempo for a 2–3 minute song: Length Minimal, Tempo Fast and "No bridge." in the brief made ours 2:35.
Make each scene from the one picture, look at it, then make it move. Pictures are cheap and clips are not, so review before you animate.
One Sing My Photo take per sung part of the song. Put each on the timeline at its own song time, with hard cuts on the beat, then make the video.
The full method behind one real 2:19 film — including what it actually cost.
Read moreA 2:44 road-trip film, one couple, fifteen shots — and nobody left the house.
Read moreOne photo, and the face in it sings every word.
Read moreYour photo in a neon ring that pulses with your song.
Read more