Name the light
Write 'she folds the letter and looks up', not 'she feels sad'. A visible action gives the model something to animate inside five seconds.
Describe a scene and a camera move, and get a short romantic clip built for a story.
Create a Love Video








Prompt Guide
Model
Reference Image
Who is in frame, where they are and what changes: 'two adults fold a letter at a kitchen table, one looks up'. Keep it under 25 words.
Five seconds for a single gesture, ten for a gesture plus a reaction. The workspace renders 1280x720 in 16:9 by default.
16:9 for YouTube and site banners, 9:16 for Reels, Shorts and TikTok. The same prompt can be rendered twice, so keep both files.
Keep the prompt and change only the light or the camera move. Changing three things at once makes it impossible to tell what fixed the shot.
Ten starting points across four moods. Open one to see the prompt, then change the place, the light or the gesture.
Two fictional adults walk barefoot where the tide has just pulled back, camera low behind them, reflections in the wet sand, dusk light, 16:9, no text.
Try This StyleTwo fictional adults share one umbrella on a rainy city street at night, neon reflections underfoot, camera tilts up from shoes to faces, 16:9, no text.
Try This StyleTwo adults turn in a slow first dance under a chandelier, warm chandelier light, dress fabric swinging, camera orbiting gently, 16:9, no text.
Try This StyleStart from a portrait of two people standing close, animate a small head turn, a blink and a smile, keep the background still, 16:9, no text.
Try This StyleA faded photograph on a table lifts into motion: the two people in it turn toward each other and laugh, the edges of the print still visible, warm grain, 16:9, no text.
Try This StyleClose shot of hands unfolding a letter by candlelight, the flame flickering, a wax seal breaking, shallow focus, warm shadows, 16:9, no text.
Try This StyleTwo adults walk across an empty bridge at blue hour, city lights behind them, camera pulling back slowly, cool palette with warm highlights, 16:9, no text.
Try This StyleTwo people say goodbye on a train platform at dusk, one steps back as the doors close, camera holds steady, station lights, 16:9, no text.
Try This StyleTwo adults dance in a small kitchen while dinner cooks, one spins the other under their arm, warm overhead light, steam, handheld camera, 16:9, no text.
Try This StyleTwo adults sit on a balcony with coffee at sunrise, one reaches over to tuck hair behind the other's ear, city waking up below, 16:9, no text.
Try This StyleWrite 'she folds the letter and looks up', not 'she feels sad'. A visible action gives the model something to animate inside five seconds.
A slow push-in or a single orbit. Two moves in one five-second render is what makes a clip look like a slideshow of stills.
Window light, tungsten bulbs, low sun behind the couple. Naming the source fixes the colour and stops the model from inventing a spotlight.
Five seconds holds one gesture; ten fits a turn and a reaction. Choose before you generate, because the length changes how tight the framing should be.
Text alone gives you two new people every render. Upload one photo when the same couple has to appear across several clips.
9:16 crops the sides, so keep the couple centred and let the camera move up or down instead of sideways.
I used a photo of my grandparents on their anniversary and asked for a slow turn toward each other. The first render kept their faces but swung the camera too fast, so I dropped the orbit and held the frame; the second take was usable in a 9:16 card.
My first prompt was too literary - 'a love that lasted forty years' gave me two strangers staring at nothing. Rewriting it as 'two adults repot a plant on a balcony, one hands over a trowel' finally produced a clip I could cut.
Ten seconds was too long for the shot I needed: the model added a second gesture and broke continuity. Five seconds of one hand reaching for the other was enough for the wedding teaser.
A short romantic clip from a written scene or a photo: a dance, a walk in the rain, a letter being read. Each render is 5 or 10 seconds at 1280x720, in 16:9 or 9:16, ready to cut into a reel or a card.
No. Text alone is enough for a scene with new people. Upload a photo when the same two faces have to appear in more than one clip, or when the clip is a gift for someone specific.
Not in one render. Generate several 5 or 10 second shots and edit them together; keeping one camera move per shot is what makes the joins invisible.
Only when you supply a reference photo. Without one, each render invents a new pair, even when the prompt is word-for-word identical.
The file is yours under the plan you generated it on. Do not upload a photo of someone else without their permission, and do not use a real person's likeness in an ad.