A genre-spanning internal benchmark testing the limits of every available AI video model.
Synopsis
A macro pour of whiskey opens the benchmark. The camera lifts into a frontier saloon where a long-haired gunslinger in a studded black stetson and leather duster laughs across the gambling table, then the cards turn. A man is launched airborne over the felt, a revolver fires through smoke, men grapple at the iron lamppost. The gunslinger walks out into the night, takes a shotgun from his saddle and mounts the black stallion as fire glows behind the saloon doors. The town becomes a chase: posse closing on foot, a building detonating in a storm of debris and embers, riders firing through the dust at full gallop. The sequence pulls out wide to a moonlit plain where he runs his horse alongside a steam locomotive lit at the horizon, then closes head-on on the train itself - and ends on the gunslinger pulled up onto its side, gloved hand locked to the iron rail.
MM Showcase is not a film. It is an internal stress test - one protagonist carried through every escalating condition that breaks AI video models. Low-light interior brawls, smoke and muzzle flash, the human-animal contact point at the mount, leg articulation at full gallop, large-scale practical fire, atmospheric night with two competing light sources, mid-action object interaction at speed. Each shot was engineered to isolate a single failure mode. The result became the production standard - every model selection at the studio for the following year was traced back to a decision made on this benchmark.
Category
Internal Benchmark
Year
2025
Runtime
1 min 18 sec
Reach
Internal production standard
The Brief
Build a single-genre internal benchmark to stress-test AI video models across the hardest conditions in a frontier western - low-light interiors, kinetic violence, human-animal contact, large-scale practical fire, and atmospheric night - using one protagonist held across every escalation.
The Approach
One character, one genre, every failure mode. A long-haired gunslinger in a studded black stetson and leather duster carried from a saloon brawl through revolver fire, a night mount on a black stallion, a posse chase, a practical-fire detonation, full-gallop mounted gunfire, and a moonlit train-board climax. Each shot engineered to isolate a single capability - face retention, kinetic distortion, animal articulation, practical fire, dual-source night lighting, and mid-action object interaction at speed.
The Result
Became the internal production standard. The continuous saloon-to-train arc proved character lock could survive a full action sequence without prompt rebuilding between cuts, and the calibration loop the test established was carried into every major production that followed.
The Gunslinger
Long dark hair, studded black leather stetson, weathered leather duster over a deep red shirt. Carried unchanged across saloon interior, night street, full gallop, and the moving train. The face retention anchor for the entire benchmark.
The Black Stallion
Obsidian-black horse with bridle and saddle, used as an environmental anchor against the dusty western set. Tested for human-animal contact at the mount, leg articulation at full gallop, and tracking parity across the moonlit plain.
Saloon Set
Interior frontier saloon - felt-topped gambling table, scattered cards, oil lamps and tungsten wall sconces, exposed wood beams. Built as a controlled low-light environment to test kinetic distortion (airborne body, revolver discharge, grapple at the lamppost) without open sky degrading the frame.
Main Street at Night
Two-story timber facades, gas lamps, hitching rails, raised boardwalks, blue night sky overhead. The exterior chase environment - tested for crowd choreography on foot and horseback, practical-fire detonation of a building, and dust-and-muzzle-flash at full gallop.
Steam Locomotive
Period-accurate steam engine head-on on a single track at night, single bright headlamp, vertical steam plume against deep blue sky. The final benchmark sequence - cold moonlight against warm headlamp and firebox glow, and the gunslinger's gloved hand locked to the iron side rail at speed.
Extreme Macro / Static / Tungsten Key
Extreme macro on a tumbler of whiskey. Amber liquid catches a single warm key light, glass distorts the saloon behind it. The benchmark opens on a still object before a single human or animal is asked of the model - a deliberate baseline for refraction, surface, and color depth.
Production Note
Glass and amber liquid are an honest test of model fidelity at rest. If refraction shifts or the meniscus drifts between frames, the foundation is unstable and nothing downstream is trustworthy. The shot is held longer than a film cut would dictate, specifically as a proof.
01.01Medium / Static / Low Tungsten
Interior frontier saloon, low tungsten and oil-lamp light. The gunslinger sits across from a laughing companion at a felt-topped gambling table, glasses and a beer bottle in frame. The mood pivots - a hand falls on his shoulder, men crowd around him, the room contracts. Character retention at conversational distance, with multiple period faces in soft focus behind.
Production Note
Multi-character interiors with overlapping hats, beards, and period costume are where most AI models lose identity lock on the protagonist. The benchmark holds the gunslinger in clear key light and lets background figures soften deliberately, proving the lock holds against visual competition.
02.0102.02
02.03Medium Wide / Handheld / Tungsten
The cards turn to violence. A body sails airborne across the gambling table, a revolver fires from inside the duster with a thick plume of muzzle smoke, men grapple at the cast-iron lamppost. The scene tests kinetic distortion at close range - airborne human in arc, smoke against tungsten, multi-body collision in low light.
Production Note
Generating an airborne body and a discharging revolver in adjacent shots is a worst-case kinetic test. The cuts arrive before any single trajectory has to fully resolve - the production used this beat specifically to identify which model handled smoke, recoil, and impact without collapsing into mush.
03.01
03.02
03.03Low Angle / Slow Motion / Practical Firelight
Exterior. The gunslinger walks out of the saloon into the night street, a sawn-off shotgun in his off-hand, and mounts the black stallion in slow motion. Warm firelight glows from inside the doorway, gas lamps line the boardwalk, posse figures spill into the frame from the saloon. The benchmark shot for human-animal contact at the mount.
Production Note
The contact point between rider and saddle at the moment of mount is where most models fail - they can render either the human or the horse cleanly, but the hand-on-pommel and foot-in-stirrup interaction at speed is the load-bearing detail. The shot was framed to hold both contact points simultaneously as the proof.
04.0104.0204.03Wide / Tracking / Practical Fire
The chase ignites. The gunslinger gallops down the night street as armed posse runs in from behind. A second gunman fires a revolver from inside a structure, then the saloon porch detonates in a sepia storm of debris, embers, and lantern fragments. Practical fire and full-gallop motion in adjacent shots.
Production Note
Holding character lock through an explosion frame is the structural test of the entire benchmark - if the gunslinger reads as the same person before and after the detonation, the character anchor has survived environmental chaos. The debris was let to dominate the frame deliberately to force the test.
05.01
05.02
05.03
05.04
05.05Tracking / Side Angle / Cool Night Key
Night gallop down the main street. A bearded rider fires a revolver near-camera with a heavy muzzle flash, a second mounted figure fires from deep in the dust behind him, and the gunslinger rides past lit shopfronts on the black stallion. Pure kinetic test - mounted gunfire, smoke, dust, and full gallop in the same frame.
Production Note
Mounted gunfire with muzzle flash at full gallop combines four hard problems at once: rider-horse continuity, recoil distortion, additive light against ambient night, and dust as a moving volumetric. The production used this beat specifically to separate the models that could carry compound motion from those that could not.
06.01
06.02
06.03
06.04Wide Tracking into Head-On / Moonlit Night
The town releases. Wide moonlit plain, a single-track railroad cuts the foreground, the gunslinger and a second rider run their horses alongside a distant steam locomotive lit at the horizon. The camera then closes head-on on the locomotive itself - single bright headlamp, vertical steam plume against deep blue night.
Production Note
Atmospheric night with two competing light sources - cold moon and warm locomotive - is the most demanding lighting condition in the test. The production used this beat specifically to determine which model would handle long-form night material going forward.
07.0107.02Side Profile / Tracking / Mixed Firebox and Moonlight
Side angle on the moving locomotive. The gunslinger is pulled up onto the iron sidestep, gloved hand locked to a vertical handrail, hat low against the steam, firebox glow on his face. The benchmark closes on mid-action object interaction at speed - human contact with hard period metalwork while both train and camera are in motion.
Production Note
A gloved hand gripping a moving iron rail at speed is the closing exam of the benchmark - if the contact point holds, the model can be trusted with action object interaction in long-form material. The shot is held to the last frame specifically as the closing proof.
08.01
08.02