Nick Ball and Brandon Figueroa face one another as modern gladiators in a sunlit colosseum.

AI commercial · Boxing

TALL ORDER: Bringing Real Fighters into a Photoreal AI Colosseum

An open brief became a gladiatorial boxing film—and taught me to generate for the story, not simply for the spectacle.

By Gabriele Stifani 5 minute read

“Boxers are the new gladiators.”

That was the central idea behind TALL ORDER: Nick Ball and Brandon Figueroa collide in a colosseum like gladiators, fighting for one world title. The proposition was simple. Making it believable was not.

AI commercial · 33 seconds Watch TALL ORDER

The film came through Vertical Haus for Queensberry Promotions and DAZN Boxing. It began with a drafted idea from the director, and my responsibility within the wider production was the hands-on AI visual build: character development, look development, image and motion generation, and the final picture edit.

Mike Staniforth’s account tells the detailed story of the wider five-day production and its journey to broadcast. Here is what I learned while creating the fighters, building their world, shaping the sequence and developing the workflow behind it.

Real fighters, real proportions

Most of the AI studies I was doing showed close-ups: one character speaking, perhaps another replying, often with people invented purely for the exercise. Those studies can show what a model does, but they avoid a harder problem. The people in TALL ORDER could not be random creations. They were Nick Ball and Brandon Figueroa, and boxing audiences already knew their faces, bodies and presence.

I began by making realistic character sheets for both fighters. I worked through facial structure, build, hair, profile, expression, wardrobe and lighting before taking any shot into motion. One impressive close-up was useless if the next shot no longer looked like the same person.

Their real height contrast was especially important. The difference is clearly visible, so it had to remain truthful—neither exaggerated for spectacle nor flattened to make generation easier. In full-body frames, face-offs and moving sequences, the two fighters had to occupy the image according to their real relative proportions.

Nick Ball and Brandon Figueroa face each other in the colosseum, with Live on DAZN and Queensberry campaign branding between them.
The final DAZN and Queensberry campaign image: the fighters’ real height contrast remains clear while both retain equal visual weight.

That contrast naturally introduced a David-and-Goliath tension in scale, but not in role. The familiar myth tells us who the hero is and who stands in his way. I did not want the film to make that decision for the audience.

I was not building one hero against a villain. I was building two heroes.

Neither fighter could be reduced to “the giant” or “the underdog.” Both needed weight, dignity and their own sense of arrival. The image could create tension from their differences while leaving viewers free to decide whom they supported.

Watermarked character-development sheet for Nick Ball, showing full-body front and back views and close facial studies.
Nick Ball character development: likeness, wardrobe and proportions studied across full-body and close facial views.
Watermarked character-development sheet for Brandon Figueroa, showing facial studies and full-body front and back views.
Brandon Figueroa character development: likeness, wardrobe and proportions studied across full-body and close facial views. These reduced-resolution, watermarked previews are shown to discourage unauthorized reuse.

An open brief with a hard deadline

The wider production moved from concept to a 30-second international DAZN broadcast in under five days. As Mike Staniforth writes, the film needed “fight-night scale”, but “volume is not progress”: “approve the world, approve the shots, protect the edit.”

For that world, I drew inspiration from the cinematography of Gladiator: hard light cutting through dust, warm stone, deep shadow, backlit figures and monumental scale. I was not trying to reproduce a particular shot. I wanted the colosseum to give boxing’s existing rituals—the entrance, the stare-down, the crowd and the moment before impact—a mythic visual language.

High-angle Gladiator reference frame of a lone gladiator crossing the arena sand, partly obscured by foreground architecture. Gladiator reference frame from a shadowed entrance, with three silhouetted figures looking into the bright, crowded arena. Low-key Gladiator reference close-up of Maximus with his eyes closed before the arena action.
Reference frames from Gladiator (2000), studied for arena geography, silhouetted entrances and intimate low-key portrait lighting.
Nine-frame look-development contact sheet with fighter close-ups, torch-lit stone cells, barred gates, sandalled arena entrances and a cheering colosseum crowd.
The look-development sheet tested hard light, stone corridors, barred gates, arena scale and crowd energy before the sequence moved into animation.

Generating for the edit

At first, I did not know exactly how the idea would compress into its final broadcast length. I built contact sheets of possible moments: arena wides, entrances, faces, wardrobe details, confrontations and impact beats. Comparing them side by side helped me choose what belonged together and recognise what was missing.

Some images I thought I needed became repetitive in the edit. Others exposed a missing transition or suggested a stronger order. A large collection of attractive images was not yet a film; the sequence still needed geography, escalation, contrast, emotion and rhythm.

The music supplied to me became another guide. Its rises, pauses and impacts suggested when to reveal the arena, when to stay with a fighter and when to accelerate towards the collision. Because I was also cutting the final picture, generation and editing became a feedback loop: the images shaped the edit, and the edit told me what to generate next.

DaVinci Resolve timeline for TALL ORDER, with generated colosseum shots, layered video clips, audio tracks and a fight frame in the viewer.
Editing was part of generation: the Resolve timeline exposed missing coverage, repetition and transitions, then sent me back to generate what the cut still needed.

From social-first to live DAZN coverage

From my side of the production, TALL ORDER began as a social-first commercial. Then it aired during DAZN’s live international coverage of Ball vs. Figueroa.

I do not need to call it a global first to understand what that meant for me. It was the first time an end-to-end AI visual workflow I had led reached a live international sports broadcast. Seeing work built through character sheets, contact sheets, generation and picture editing hold its place inside a major fight-night presentation was one of the proudest moments of my early AI work.

The film continued to travel through Queensberry Promotions’ official Instagram account. As of 6 August 2026, the reel showed 6,131 likes, 101 comments and 99 reposts—a dated snapshot of the audience engagement around the work.

Watch the official campaign reel from Queensberry Promotions

Official Queensberry Promotions campaign-reel cover showing Nick Ball and Brandon Figueroa facing one another in the colosseum.
The official campaign reel as published by Queensberry Promotions. Watch it on Instagram.

What stayed with me

Before the arrival of Seedance 2.0, keeping characters, locations and proportions consistent was not easy. Kling 3.0 did a strong job, especially when I generated a sequence from a first frame or used start-and-end-frame techniques. It became a continuous edit in my head: every generation was already being judged by how it could enter, connect to and leave the next shot.

Kling alone did not solve every continuity problem. I also edited images in Photoshop to preserve character details, location elements and the relative proportions of the fighters before sending them back into the generation workflow. The process became a loop between FLORA, Kling 3.0, Photoshop and the edit.

FLORA workflow canvas for TALL ORDER showing fighter reference frames connected to alternate generated entrance and face-off sequences.
Inside the FLORA workflow: source frames branch into alternate entrances and face-offs while continuity, scale and sequence are judged together.

The models will keep changing. The deeper lesson will remain.

AI did not remove the need for judgement. It made selection, continuity, scale, coverage and editing even more important. The workflow did not simply help me create what I wanted; it revealed what the film needed.

TALL ORDER was one of the most gratifying projects I had worked on. It taught me that speed is only useful when every image still serves the characters and the story.

Watch and read