Clip made with Seedance 2.0 from this exact prompt. Hover to play, or tap play on a phone.

What each part of the prompt does

One person, one action
The rower is the only person with a job: pull the oar and glide. Giving a single figure a single repeated motion keeps hands and the oar believable for five seconds.
Lanterns at three depths
Eave lanterns on the buildings, a boat full of them in the foreground and floating ones on the water fill the frame in layers, which gives the shot depth without any fast camera work.
Reflections named
"Long reflections shimmer" tells the model the water is dark and smooth. That is where most of the light in a dusk river scene really comes from.
Sky colour behind the bridge
Placing the pink-violet glow behind the bridge puts the brightest part of the sky where the push-in ends, so the camera move has a destination.

Variations to try

Same idea, one thing changed. These have no clip yet.

  • Vertical, boat close-up

    A narrow wooden boat stacked with glowing paper lanterns on a dark river at dusk, ochre shophouses with silk lanterns on the far bank. Shot 1: close shot at water level, the boat glides past the lens, candle flames flickering inside the lanterns. Shot 2: the camera turns to follow the boat as it moves toward a small roofed footbridge. Setting and light: deep blue sky, warm lantern light reflected on the water. Style: cinematic, warm and romantic, shallow depth of field. Sound: water lapping, the splash of an oar. Duration: 5 seconds, vertical 9:16.

  • Releasing a lantern

    Two hands lower a single paper lantern with a lit candle onto a dark river at dusk, old ochre shophouses and a small roofed footbridge behind. Shot 1: close shot, the lantern touches the water and starts to drift away. Shot 2: the camera holds as the lantern joins dozens of others floating toward the bridge. Setting and light: deep blue sky with a pink glow, warm lantern light on the water. Style: cinematic, quiet and tender, shallow depth of field. Sound: a soft ripple, distant voices. Duration: 5 seconds.

If the result isn't right

  • If the lanterns race downstream, write "lanterns drift very slowly, almost still"; a moving boat makes the model speed up everything on the water.
  • Crowds on the promenade can turn into a blur of half-formed people. Leave them out, or ask for "a few people far away, out of focus".
  • Keep the town unnamed. A generic old trading town gives you the look while avoiding landmark details the model may get wrong.
  • When the clip is for a festival or event, add your own date and place as text in the edit and leave the lanterns plain; written characters on generated lanterns come out garbled.