Head to head: Bernini-R Reference Edit Video vs Marey Realism V1.5
Bernini-R Reference Edit Video vs Marey Realism V1.5
By Ryan Merket · Published
This matchup pits prompt-rich cinematic rendering against stronger physical and temporal coherence across crowds, domestic motion, macro action, and buoyancy tests.
The 28.2–28.5 aggregate split is negligible. Marey Realism V1.5 has the higher raw score, but with only limited confidence that either model is genuinely better, this is statistical noise—not a defensible victory. Marey made its clearest case in the laundry-room tests, winning both by actually putting the tabby in motion across the dryer while preserving the rain, steam, curtain, hanging laundry, and melancholy atmosphere. It also produced the more convincing dense Tokyo scramble in one crowd test and the more complete espresso setup in one macro test, including details Bernini-R sometimes dropped, such as the copper spoon and steam. Bernini-R answered with a clean sweep of the physics-realism tests. Its cube penetrated, rose, and settled with more believable buoyancy, while Marey’s persistent splash columns, shifting cube appearance, or continued sinking weakened the illusion. Bernini-R also took the alternate crowd and espresso tests through cleaner pedestrian separation, stronger frame-to-frame coherence, and more natural splash behavior—even when its prompt coverage was less complete. The pattern is sharp: choose Marey when scene density, atmosphere, and explicit prompt elements matter most; choose Bernini-R when motion integrity and physical plausibility are the priority. **Final call: too close to call—a genuine tie. Marey’s 0.3-point edge does not clear the statistical bar for a winner.**
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.
OpenAI: GPT-5.6 Sol Pro: Model B better captures the scale, density, elevated viewpoint, and recognizable Tokyo scramble-crossing atmosphere, with coherent crowd flow across frames. Model A preserves cleaner individual pedestrians with less visible blur or overlap, but the crossing feels comparatively sparse and less convincingly busy. (Second judge pass, order swapped — scores are the average of both: Model B more strongly fulfills the busy Tokyo scramble-crossing prompt, with a dense multidirectional crowd, stable set | Anthropic: Claude Fable 5.1: Model A renders a moderately busy Tokyo cr…
Laundry room draft
A single continuous 16:9 shot in a modest home laundry room during a rainy afternoon: a tabby cat pads along the lid of a running dryer while a pale yellow curtain by the half-open window billows, damp maple leaves outside tremble, a forgotten shirt sleeve hanging from a rack sways subtly, condensation threads down the windowpane, and a mug on the sill sends up thin steam, with every background motion natural and uninterrupted; the camera performs a slow waist-high tracking move to the right past the cat, lit by soft overcast daylight and the low amber bulb above the machines, giving the scene a cozy, lived-in, gently melancholy mood.
OpenAI: GPT-5.6 Sol Pro: Model B better fulfills the central action with the tabby visibly padding across the dryer beneath a rainy, condensed window, while also including the curtain, hanging sleeve, and steaming mug. Model A is more photorealistic and temporally stable, but the cat remains on the windowsill rather than the dryer and shows little discernible movement. (Second judge pass, order swapped — scores are the average of both: Model B more fully realizes the requested scene, with the cat visibly padding across | Anthropic: Claude Fable 5.1: Model B delivers far more of the prompt's l…
Espresso bead burst
A single continuous 16:9 shot in a quiet apartment kitchen at blue dawn: as a copper teaspoon drops one sugar cube into a small cobalt espresso cup, a crown of coffee and tiny amber droplets erupts upward in convincing slow motion, each bead sharply resolved with fluid frame-to-frame motion while faint steam curls from the cup; the camera makes a gentle intimate close-up push-in with a slight left-to-right arc around the saucer, catching cool window light and a warm stove glow on the splash, creating a hushed, precise, almost meditative mood.
OpenAI: GPT-5.6 Sol Pro: Model B better captures the cobalt cup, copper teaspoon, sharply resolved amber droplets, faint steam, and dramatic cool-warm lighting, with a convincing splash progression despite the sugar cube being unclear. Model A has a coherent kitchen setting but omits the spoon and cube in the sampled frames, and its oversized pale splash transitions into a less natural central mound. (Second judge pass, order swapped — scores are the average of both: Model B delivers the sharper, more cinematic amber be | Anthropic: Claude Fable 5.1: Model B delivers the copper spoon, cobalt…
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.
OpenAI: GPT-5.6 Sol Pro: Model A shows a more believable submerged-to-floating progression, stable cube geometry, and naturally distributed fizz, though the initial splash is somewhat foamy. Model B is more cinematic, but the oversized persistent splash column and shifting cube appearance undermine buoyancy realism and temporal consistency. (Second judge pass, order swapped — scores are the average of both: Model A more clearly depicts a coherent splash, submergence, buoyant rise, and settling near the surface, with per | Anthropic: Claude Fable 5.1: Model A looks like genuine footage: the fo…
Matchup powered by OpenRouter.