The task
This is our own project; the brand did not commission anything. We wanted to test on ourselves whether a spot at the level of a sports brand can be built without a camera, actors or a location, so we gave ourselves a brief in the Nike spirit: “One guy. One city. No rules.” Dawn, a basketball court in fog, the roofs of Soviet-era apartment blocks, parkour. 9:16 for TikTok and Reels, about 30 seconds.
The hero was described down to the details, because otherwise he changes from scene to scene: stocky, athletic, buzz cut, black oversized tee, pink long sleeve underneath, olive knee-length cargo shorts, black high socks. The sneaker: black upper, pink sole, white logo on the tongue. Light was fixed as a rule too: the sun on the horizon in front, light hitting the face, shadows going back, the hero always moving toward the sun. Background: five- to nine-storey panel blocks, roofs of flat dark roofing felt.
Why, and what the idea was
The hardest part of AI video is not a pretty frame but the same hero and the same product across a dozen shots in a row. So the main idea of the project was not the plot but the order of work: references first, scenes second. No scene was generated from a verbal description; every one got images: the face up close, the full body, the sneaker from the needed side.
The second decision was about the edit: the first shot is not the first in the script but the strongest one. The video opens with the finale, the hero from behind with arms spread over the city, and only then come the laces, the court and the dribbling.
How it went
| When | What happened |
|---|---|
| Stage 1 | Concept, hero description, light rules, a scene-by-scene script: emerging from fog, a first-person run, the jump and landing, a rooftop run, the finale, two sneaker inserts, an end card |
| Stage 2 | 13 references in Nano Banana: face, full body with the ball, sneaker from the side, sole, top view for the first-person scene, key frames for the jump, the landing, the run, the sole impact, the roof edge, the portrait and the view from behind |
| Stage 3 | Scene generation: emerging from fog from a start frame, the first-person run from text, the jump and landing as one 10-second scene, the finale at 8 seconds, inserts at 3 to 5 seconds, the sneaker card at 5 seconds |
| Stage 4 | Found a mistake in every prompt: “black joggers” instead of cargo shorts. Rewrote them all, regenerated the scenes where the sneaker changed design |
| Stage 5 | A sound map of 9 effects with timings: rooftop wind, laces, dribbling on a wet court, the jump and landing, a bass hit. Voice track and a Suno beat |
| Stage 6 | Edit in CapCut: 3-second cuts, the Nike title in Bebas Neue, music ducked under the voice with keyframes. 31 seconds |
Tools and method
Frames: Nano Banana. Scenes with the hero: Seedance in Omni Reference mode, which takes up to 12 references at once. Scenes with a ready key frame were animated as I2V. Scenes without the hero, such as the first-person run and the sneaker card, were made from text. Kling was kept for simple motion and atmospheric details.
Every prompt carried lock lines: one for the face, one for the clothes, one for the sneaker. Plus three mandatory phrases: smooth motion with no stops, zero other people in the frame, the same man as in the references. Before each scene we counted its beats and took the duration from that: one beat is 3 to 5 seconds, a jump with a landing is 8 to 10.
Sound: ElevenLabs effects set to 8 to 10 seconds, because by default the generator gives 1 to 2 seconds. Voice in ElevenLabs, three short lines, each as a separate file so they are easy to place in the edit. Music from Suno for “dark phonk, 130 BPM, instrumental only”. In the mix the music sits at 30 to 40 percent, 15 under the voice.
Result
The video is 31 seconds, 9:16. 0-3 s: the hero from behind, arms spread, the city below. 3-6 s: laces up close. 6-18 s: full body with the ball and dribbling from three angles on a wet court among apartment blocks. 18-21 s: the toe of the sneaker hangs over the parapet. 21-24 s: a first-person rooftop run toward the sun. 24-27 s: the jump between buildings. 27-30 s: the finale, the face in dawn light. 30-31 s: the sneaker against the sky and the Nike title.
What we accepted with reservations. The sole-impact insert: gravel instead of concrete and someone else’s logo on the sneaker, but the motion and the dust came out strong, so it stayed. The first-person run: Seedance makes it slow, but the shot with the sneakers on the edge at dawn is atmospheric. The finale: the camera’s entry from the front did not work, but the view from behind with the city to the horizon became the best shot of the video and went into the hook. For the rooftop run no new scene was generated: a ready clip turned up that fit.
Mistakes and what we took from them
