Canton Tower FPV Drone · Racing Up the Tower Along a Red Arrow

GPT Image 2 + Seedance 2.0Drone AerialOne TakeFirst-Person POVUrbanImage to VideoVerticalSeedance 2.0GPT Image 215s
An image-to-video trick: draw a red arrow path on a night shot of Canton Tower, then have the model race along it in first person, hugging the tower as it climbs, before pulling back into a full aerial view.
PROMPT · Video Prompt
Create with AI
Please remove the red line and arrow. The camera, at extreme first-person speed, faithfully follows the red arrow's path as it rushes forward. Cinematic in feel. It starts with a low-angle street-level view reaching the base of the building, combining ambient light, reflections and realistic material detail, then shoots sharply upward along the base of the tower. While climbing and passing through the center of the building, it captures the expressions and movements of people relaxing on the lawn below, then turns back to look at the full aerial view of the building. The image is richly detailed, realistic and strongly atmospheric. One continuous take, with no additional cuts or splicing.
✍️ Editor’s Notes

The core technique here is 'image-to-video plus a path annotation': a red arrow is drawn on the reference image to mark the camera path, then the prompt opens with a dedicated line — 'please remove the red line and arrow' — telling the model to erase the annotation tool and keep only the resulting camera move. That's the standard move for working from an annotated reference image. The camera move itself is written in three beats: a low ground-level approach, a steep climb hugging the tower, and a turn into an aerial overview at the end — each beat paired with a concrete visual requirement (light/reflections, material detail, capturing people's expressions) instead of a vague 'epic aerial shot.' The closing line, 'one continuous take, with no additional cuts or splicing,' locks the narrative into a single long shot so the model can't break the rhythm with its own cuts.

Want tighter control? Editor-expanded reference version (not the original prompt):
The shot can be broken into four beats: 1) Ground segment (0–3s): a low camera position close to the sidewalk, with a slight handheld-style wobble, as the tower slowly enters frame from the top; 2) Steep climb segment (3–8s): speed spikes sharply, the tower facade's LED texture shows subtle motion blur at this speed, and wind noise builds with it; 3) Pass-through segment (8–11s): the camera skims through the tower's latticed structure, light flares flickering rapidly across the lens; 4) Turn-back finale (11–15s): the camera decelerates and slowly turns, the score shifting from an urgent drumbeat to a long, distant synth tone, settling on a wide shot of the tower against the city skyline.
Create with AI
Related Works
׋›
↑