Bernini-R Edit Video never really gets on the board here. The aggregate score gap — 33.8 to 27.8 — is substantial, and the statistical read is even more decisive: Gemini Omni Flash wins with 94% confidence. That lines up with the task count too: Gemini takes 3 wins, Bernini takes 0, with 1 tie.
The clearest separation is in camera motion control. Bernini produces a plausible ramen-restaurant scene, but the requested smooth 180-degree orbit mostly reads like a restrained pan or slide. Gemini, by contrast, delivers the actual shot: stronger parallax, a more continuous constant-height orbit, and a more cinematic sense of moving around the bowl rather than merely reframing it.
Gemini also wins where video models usually crack: temporal consistency and crowd motion. In the engineer-on-catwalk task, Bernini looks good in isolated moments but fumbles key prompt details — wrong-hand toolbox carry, unclear valve interaction, inconsistent staging. Gemini is simply more coherent across the sequence and tracks the requested camera progression better. In the Tokyo scramble scene, the difference is even more practical: Gemini gives you a recognizable scramble crossing with denser, more independent pedestrian flow, while Bernini feels lighter, less specific, and less convincing as a busy urban crowd.
The only non-loss for Bernini is the atmosphere & mood in motion task, and even that ends as a tie after the second pass. That says something important: Bernini can compete on mood, lighting, and industrial texture. But in this matchup, Gemini keeps winning the more consequential argument — not just whether a clip looks good in a frame, but whether it actually executes the shot, the action, and the scene dynamics the prompt asked for.
Final call: Gemini Omni Flash is the clear winner. Bernini-R Edit Video has style, but Gemini is the more reliable and better-directed video model.
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 27.9 to Gemini Omni Flash's 33.8.
1. Camera motion control
A single smooth orbit (roughly 180 degrees) around a steaming bowl of ramen on a wooden table, the camera circling at a constant speed and height in one continuous move, warm restaurant light, 16:9.
Winner: Gemini Omni Flash — Model A shows a plausible restaurant setting and steady framing, but the camera movement appears limited and the bowl composition changes only modestly, falling short of a clear smooth 180-degree orbit. Model B better conveys a continuous orbit at consistent height with stronger parallax and more cinematic presentation, while maintaining good temporal coherence and overall visual appeal. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested smooth constant-height orbit around a steaming ramen bowl, with clear lateral viewpoint change, strong restaurant ambiance, and appealing cinematic detail. Model A is more stable and realistic-looking in some respects, but the camera movement appears closer to a mild pan/slide than a pronounced 180-degree orbit, and the scene is less faithful to the warm ramen-restaurant aesthetic.)
2. Atmosphere & mood in motion
A 12-second continuous shot in a vast midnight sheet-metal stamping hall after rain, where an automated gantry crane glides above rows of dormant presses carrying a single glowing amber furnace core the size of a bathtub while a small inspection trolley rolls below on wet rails; the camera begins low beside puddles reflecting blue emergency strips, then slowly dollies forward and tilts upward into a wide orbit around the moving crane, so the mood evolves from hushed and serene to quietly awe-struck as the core’s warm light blooms across mist, catches suspended dust, and gradually wakes silver highlights on the machinery without any scene cut; pacing should feel deliberate and breath-like, with soft hydraulic motion and expanding pools of light shaping the emotion, 16:9
Winner: Tie — Model A better matches the prompt’s evolving mood and camera language: it shows the wet stamping hall, dormant presses, gantry-carried glowing core, inspection trolley, and a gradual upward/orbiting perspective with cohesive lighting bloom. Model B has strong atmosphere and includes the trolley and crane, but the furnace core appears overly fiery and less like a contained glowing amber mass, making the tone and prompt details feel less precise and slightly less temporally coherent. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested emotional progression and camera language: it starts low by reflective puddles with the inspection trolley visible, then reveals the overhead crane and warm core blooming through mist in a coherent, cinematic way. Model A has strong industrial atmosphere and appealing lighting, but the carried object reads more like a large ladle than a bathtub-sized furnace core, and the sampled framing emphasizes side-tracking close views over the specified low-start-to-wide-orbit evolution.)
3. Temporal consistency
A 9-second continuous shot inside the copper-walled calibration bay of the decommissioned Kestrel-9 turbine lab: a single maintenance engineer with a shaved head, a crescent scar above the left eyebrow, and a mustard-yellow insulated coverall with a stitched patch reading VERA-17 and one missing reflective stripe on the right forearm walks steadily along a grated catwalk while carrying a dented teal toolbox in the left hand and tightening a silver valve wheel with the right; the camera performs a slow dramatic orbit from front-left to rear-right at chest height, staying close enough that the engineer’s face, clothing details, toolbox dents, and gait must remain perfectly identical from first frame to last with zero morphing or flicker; harsh sodium-vapor work lights rake across drifting steam and polished steel, creating a focused, procedural mood, 16:9
Winner: Gemini Omni Flash — Model A has a strong copper-bay look and keeps the shaved head and teal toolbox fairly stable, but it misses key prompt actions and details: the toolbox is in the wrong hand, the valve-tightening action is unclear, and the catwalk/close-orbit staging is inconsistent. Model B better matches the grated catwalk, valve interaction, left-hand toolbox carry, and front-left to rear-right camera progression, with steadier identity and cleaner motion overall, though some clothing details and exact facial markings are less faithfully preserved. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt’s catwalk setting, close orbiting progression, and toolbox-carrying engineer, with stronger continuity in face, clothing, and environment across frames; its main miss is that the valve-tightening action and some specified clothing details are not clearly maintained. Model A has a striking copper-bay look and visible scar, but it departs more from the prompt with a different corridor layout, inconsistent action/pose, and weaker temporal continuity in how the engineer interacts with objects and moves through the space.)
4. 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: Gemini Omni Flash — Model B better matches the prompt with a clearly recognizable Tokyo scramble crossing, a larger crowd moving in multiple directions, and stronger overall scene realism. Model A has decent independent pedestrian motion, but it feels less like a busy scramble crossing and shows a few less convincing interactions and composition choices. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt with a recognizable Tokyo-style scramble crossing, denser multidirectional pedestrian flow, and stronger sense that individuals move independently without obvious merging. Model A is cleaner and fairly consistent, but it shows a less busy scene and includes some less natural crowd interactions and weaker scramble-crossing specificity.)
See every prompt and the full side-by-side outputs in the interactive Head-to-Head.