Loqman × Jul, a vertical promo video made entirely with AI

In late September 2026, Jul released an unfinished track and invited his listeners to write the second verse. Loqman, a rapper friend, wrote his and cobbled together a first video with a consumer generation tool: a landscape picture on a phone screen, a watermark, characters whose faces change mid-shot. He asked me to remake something clean. It is an offered video: there was no quote and no payment, and this file does not present it as a service.

Work as on a real shoot, without a camera. A production bible sets the scene (a studio with an orange backdrop, a single hanging microphone), the two characters and their outfits, then breaks the track into fifteen shots aligned on the beat grid: every cut lands on a beat. For each shot, a still image is generated and approved first, then animated into video. The sung shots then go through a lip-sync model driven by the track’s isolated vocal, so that the mouth actually follows the words. Every version of the edit was reviewed by the artist, and his feedback guided twelve successive versions.

• A 62-second vertical edit at 24 frames per second, with an opening title and an end card that invites viewers to vote and states that the video was made with AI.
• Fifteen shots written in advance: wide shot, close-ups, the microphone handover, an overhead shot, back to the duo for the last verse.
• The track’s vocal separated from the instrumental, then cut shot by shot to drive the lip movement.
• An automatic crop around the right face when both artists share the frame, then recomposition into the original picture: the model animates only the right person.
• Checks that are measured rather than felt: frame count of the edit, cuts verified on the grid, mouth opening compared with the vocal envelope on every shot.
• Every take kept: each shot retains its backup variants, and the edit is rebuilt from a single decision list.