Head to head: Bernini-R Edit Video vs Happy Horse

Happy Horse takes this head-to-head cleanly. It posts a 34.8 vs 26.6 aggregate advantage, wins 3 of 4 tasks with one tie, and the statistical read is decisive: 94% confidence that this is the real ordering, not noise. Bernini-R Edit Video never actually wins a task.

The pattern is straightforward. On Crowd motion, Happy Horse is the one that actually delivers a Tokyo-style scramble crossing: denser foot traffic, more convincing multi-directional pedestrian movement, and a scene that reads as the prompt rather than a generic intersection. Bernini-R is stable and tidy, but too sparse and too plain to sell the brief.

On Moray Behind Pylon and Kelp Lab Dawn Drift, the gap gets wider because prompt fidelity matters and Bernini-R repeatedly drops key details. Happy Horse gives you the moray eel, the orange tag, the pylon occlusion, and the underwater pier structure with believable continuity. In the kelp-lab scene, it’s again the only model really honoring the half-submerged dawn viewpoint, the diver beneath the surface, the marked buoys, and the environmental specificity. Bernini-R’s outputs are often cleaner than average, but they keep simplifying or misreading the scene.

The only place Bernini-R avoids a loss is Single continuous shot, and even there it doesn’t convert. One judge preferred its cleaner entrance-to-altar glide; the other preferred Happy Horse’s richer cathedral atmosphere and stronger light treatment. A tie is the right outcome, and it underscores the broader story: Bernini-R can be competent when the task is mostly about maintaining a stable camera move, but Happy Horse is the model with the stronger visual imagination and better prompt adherence.

Final call: Happy Horse is the clear winner. It is more faithful to complex scene instructions, better at preserving specific objects and staging through motion, and simply produces more convincing video across this set.

How they were tested

We ran 4 fresh video tasks, generated on the fly for this matchup so neither model could prepare in advance, and had gpt-5.4 score each one. To cancel position bias, every task was judged twice — once in each presentation order — and every number reported here, including the headline totals, is the average of both passes. Bernini-R Edit Video scored 26.6 to Happy Horse's 34.8.

1. Crowd motion

A busy Tokyo scramble crossing seen from above, dozens of pedestrians crossing in different directions, each moving independently without merging or warping into one another, overcast daylight, 16:9.

Winner: Happy Horse — Model B better matches the prompt by depicting a busier Tokyo scramble crossing with many pedestrians moving in multiple directions while maintaining clearer independent trajectories and stronger overall scene realism. Model A is clean and temporally stable, but it feels less crowded and less distinctly like the iconic Tokyo scramble, making it weaker on prompt adherence and overall impact. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt with a recognizable busy Tokyo-style scramble crossing, higher pedestrian density, and more convincing independent crowd motion across frames. Model A is cleaner and fairly consistent, but it shows a simpler intersection with fewer people and less of the dense multi-directional scramble specified in the prompt.)

2. Moray Behind Pylon

A short continuous 16:9 underwater shot beneath the Pier 14 service deck in clear green water: the camera slowly pushes forward and slightly pans right, tracking a cobalt-striped moray eel carrying a bright orange maintenance tag in its mouth as it swims from left to right, passes completely behind a barnacle-encrusted concrete pylon for a full moment, then emerges on the other side unchanged with the same tag and stripe pattern, continuing smoothly past dangling mussel ropes and a blinking pressure gauge; shafts of late-afternoon sunlight tremble through the surface above, suspended silt glitters, and the mood is tense but mesmerizing, no cuts.

Winner: Happy Horse — Model B matches the prompt much better: it clearly shows a moray eel with an orange tag moving left to right through a detailed pier structure with barnacled pylon, ropes, gauge, and strong underwater lighting, while maintaining a more believable eel form and continuity. Model A includes the pylon occlusion and gauge, but the creature looks more like a stylized fish than a moray eel, the tag placement changes noticeably, and the overall scene is less convincing and polished. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt more closely: the eel carries the orange tag consistently, passes behind a barnacle-covered pylon, then re-emerges unchanged while the underwater pier details, ropes, gauge, and lighting all read clearly and cinematically. Model A captures the basic eel-and-pylon idea but looks less like a moray, misses much of the specified pier-deck detail and occlusion staging, and appears more stylized and less temporally grounded.)

3. Single continuous shot

One unbroken take gliding slowly through a candlelit cathedral from the entrance toward the altar, no cuts, jumps, or transitions, dust and warm light in the air, 16:9.

Winner: Tie — Model A better matches the prompt’s slow unbroken glide from the entrance toward the altar, with consistent forward motion down a candlelit central aisle and warm dusty atmosphere. Model B is visually striking and temporally stable, but its framing feels more like a centered push-in already near the nave and emphasizes cool daylight beams over the requested candlelit entrance-to-altar progression. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt with a majestic candlelit cathedral nave, visible dust-filled warm light beams, and a convincing slow forward glide toward the altar in a stable unbroken composition. Model A also follows the brief and has smooth forward motion, but its hazier, lower-contrast look and less striking light treatment make it feel less visually rich and slightly less aligned with the specified warm dust-and-light atmosphere.)

4. Kelp Lab Dawn Drift

A short continuous 16:9 shot inside a half-submerged coastal research lagoon at dawn: the camera makes a slow sideways dolly along a rust-red floating walkway while a diver in a faded chartreuse drysuit kicks gently beneath the surface beside tethered sensor buoys marked "B7"; all around, the environment stays alive with natural background motion—long ribbons of bull kelp sway at different tempos, silver ripples skate across the waterline, tiny air bubbles rise from a cracked hose, a pale steam plume curls from a compact desalination unit on the dock, and high cirrus clouds drift steadily in the pink-blue sky reflected on the water; cool early light, quiet curious mood, no cuts.

Winner: Happy Horse — Model B matches the prompt far more closely with a half-submerged lagoon viewpoint, a diver in a chartreuse drysuit beneath the surface beside clearly marked B7 buoys, rust-red floating walkway, hose, kelp, and strong dawn atmosphere; its frames also suggest smoother continuous lateral motion and better visual coherence. Model A is cleaner than average but misses key prompt elements, especially the underwater/half-submerged perspective and the diver being beneath the surface, and its buoy markings and scene details are less faithful. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt much more closely: it presents a half-submerged dawn lagoon view with a sideways drift along the rust-red walkway, a diver in a chartreuse suit beneath the surface, visible B7 buoys, reflective waterline detail, and stronger environmental richness. Model A has pleasing color and a stable wide shot, but it misses key prompt elements by placing the figure on top of the walkway rather than underwater, showing incorrect buoy markings, and lacking the specified half-submerged intimate lagoon perspective and underwater kelp activity.)

See every prompt and the full side-by-side outputs in the interactive Head-to-Head.