Bernini-R Edit Video takes this one cleanly. The aggregate gap is meaningful — 32.3 to 27.9 — and the statistical read is stronger still: a 94% confidence win, with Bernini-R claiming 3 task wins to 0, plus one tie. That’s not a vibes victory; it’s a real performance edge.
The clearest separation came on prompts that punish weak temporal reasoning. In Macaw Behind Laundry, Bernini-R was simply more competent: the bird starts where it should, passes behind the laundry with coherent identity and scale, and re-emerges to land on the mint-green cage stand. Marey produced attractive rooftop imagery, but it lost the thread exactly where the prompt got difficult — through occlusion and re-entry. That’s the kind of miss that matters.
Bernini-R also came out ahead on Subject action, where the judges ultimately favored its more believable hand motion and latte-art progression, and on Physics realism, where the averaged result landed in its favor for a more coherent drop-splash-bob sequence. Those weren’t unanimous first-glance blowouts — the swapped-order second passes show some genuine ambiguity — but the combined scoring still points the same way: Bernini-R is more reliable when a scene has to evolve correctly over time, not just look good in isolated moments.
Marey Realism V1.5’s best case is obvious: it can be striking, moody, and occasionally more cinematic. That showed up most in Terrier Barrel Bounce, which ended as a tie after the judges split on whether to reward prompt detail and setup fidelity or atmosphere and composition. But that tie is also the story of Marey’s ceiling and floor: when the brief gets messy, Bernini-R keeps the action legible, while Marey too often settles for pretty approximation.
Final call: Bernini-R Edit Video is the better video model here, and it wins for the right reason — stronger prompt fidelity under motion, occlusion, and physical progression, with enough statistical separation to call this a clear result.
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 32.3 to Marey Realism V1.5's 27.9.
1. Subject action
A barista's hands pouring latte art: the milk stream forms a clean rosetta in the crema with natural, fluid wrist motion, no cuts, overhead close-up, soft café light, 16:9.
Winner: Bernini-R Edit Video — Model A better matches the prompt with an overhead close-up, soft café lighting, and a believable progression of the milk stream forming a clean rosetta with fluid hand motion. Model B is visually striking but less faithful in camera angle and style, and its latte art progression appears more static and less naturally animated across frames. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt with a clear overhead close-up of hands pouring and a recognizable rosetta forming in the crema, while maintaining strong visual appeal and continuity across frames. Model A has smoother-looking progression and good fidelity, but the angle is less overhead and the latte art reads more like a tulip/stacked leaf than a clean rosetta.)
2. Macaw Behind Laundry
One continuous 16:9 shot in a cramped rooftop courtyard in late-afternoon sun: a blue-and-gold macaw named Tico waddles along a mosaic parapet, then launches into a short glide that carries him behind a row of wind-whipped hanging laundry—striped sheets, a red soccer jersey with the number 27, and two clipped pillowcases—before he re-emerges on the far side and lands on a dented mint-green cage stand, unchanged in size, coloring, and pose continuity; the camera begins in a medium-wide profile and makes a gentle handheld sidestep tracking move to the right with a slight upward tilt, warm golden light and fluttering shadows giving a playful, domestic mood; the bird must remain coherent through the occlusion as fabric overlaps, flaps, and briefly hides him, no cuts.
Winner: Bernini-R Edit Video — Model A follows the prompt much more closely: the macaw starts on the mosaic parapet, passes coherently behind wind-blown laundry including the red 27 jersey, and re-emerges to land on the mint-green cage stand with strong continuity. Model B has appealing warm rooftop imagery, but the bird’s action is less coherent through occlusion and it fails to clearly show the required re-emergence and landing continuity. (Second judge pass, order swapped — scores are the average of both: Model A follows the prompt much more closely: the macaw starts on the parapet, passes behind wind-blown laundry including the red 27 jersey, and re-emerges to land on the mint-green cage stand with coherent scale and identity. Model B has appealing warm light, but the bird disappears behind the laundry without a clear re-emergence or landing continuity, weakening both adherence and temporal coherence.)
3. Physics realism
A close-up of a single ice cube dropped into a tall glass of soda: it falls, splashes, bobs, and settles with believable buoyancy and fizzing bubbles rising naturally, soft kitchen light, 16:9.
Winner: Bernini-R Edit Video — Model B better matches the prompt’s believable buoyancy and natural fizzing, with clearer rising bubbles, a convincing bob near the surface, and stronger visual appeal. Model A has a plausible splash entry, but the ice cube appears to sink and settle too low for a single cube, making the buoyancy feel less realistic despite decent consistency. (Second judge pass, order swapped — scores are the average of both: Model A better matches the prompt’s physics: the ice cube clearly drops into the soda, creates a plausible splash and surface disturbance, then appears to bob and settle with believable buoyancy. Model B is visually striking, but the temporal progression looks inconsistent and the cube/splash interaction feels less physically coherent, with the cube seeming to hover near the rim rather than naturally submerge and resurface.)
4. Terrier Barrel Bounce
A single continuous 16:9 shot at a windswept coastal cider yard at blue-hour dusk: a rusty 20-liter oak tasting barrel slips from a plank and rolls off, then tumbles and bounces down three uneven granite steps while a wiry brindle terrier in a saffron raincoat sprints after it, splashing through shallow puddles; the camera starts low beside the plank and smoothly dollies backward while panning to keep both dog and barrel centered, with cold harbor light, a flickering sodium lamp, and wet stone reflections creating a tense, lively mood; the barrel’s weight, spin, impacts, brief airborne arcs, and the terrier’s paw traction and body recoil must all look physically believable under gravity and momentum, no cuts.
Winner: Tie — Model A matches the prompt more closely with a brindle terrier in a saffron raincoat, visible granite steps, harbor setting, and a more plausible chase dynamic between dog and barrel. Model B is moodier and attractive, but it misses key prompt details like the coat color and step action, and the barrel-dog interaction looks less physically grounded and temporally coherent. (Second judge pass, order swapped — scores are the average of both: Model B better matches the blue-hour coastal mood, wet stone reflections, and the terrier chasing a wooden barrel on granite steps, with stronger cinematic composition and more believable contact between dog, puddles, and barrel. Model A captures the general setup but misses key prompt details like the brindle terrier and rusty oak look, and its motion/staging across frames feels less physically grounded and less polished.)
See every prompt and the full side-by-side outputs in the interactive Head-to-Head.