Head to head: Bernini-R Edit Video vs Seedance 2 Image to Video

The topline number says it plainly: 28.6 vs 31.4 is not a meaningful separation here. With only 69% confidence that either model is genuinely better, this is not a result that supports crowning a winner. In editorial terms, these two are effectively even.

The split in strengths is unusually clean. Bernini-R Edit Video takes the physics-realism task because it gets the ice cube behavior closer to what the prompt asked for: the cube disturbs the soda, rises, bobs, and settles with more believable buoyancy while the fizz stays readable. Seedance’s version is polished, but the cube lingering near the bottom undercuts both realism and adherence.

Seedance 2 Image to Video, though, is the better prompt follower on camera logic and scene choreography. It wins the pearl-diver test by more faithfully staging the ascent, wrist-check, fin details, and overall underwater progression. It also wins the single-continuous-shot task with the clearer entrance-to-altar glide; Bernini-R’s cathedral clip looks handsome, but it feels like it starts in the wrong place and drifts laterally instead of committing to the requested move.

The kelp-tea-plume task is the reason this matchup refuses to resolve cleanly. One judge pass favored Seedance for atmosphere, lighting, and a stronger intimate underwater composition; the swapped-order pass favored Bernini-R for actually including more of the prompt-specific mechanics, including the strapped samovar setup, clearer plume behavior, bubbles, and the orange squat lobster. That split verdict is a good summary of the whole contest: one model often looks more cinematically composed, the other often nails the physical or object-level brief more literally.

Final call: too close to call. Bernini-R Edit Video and Seedance 2 Image to Video are effectively tied here — Bernini-R has the edge in physical plausibility, Seedance has the edge in shot execution and prompt staging, and the aggregate gap is too small to honestly turn into a win.

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 28.6 to Seedance 2 Image to Video's 31.4.

1. 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 A better matches the requested physics: the ice cube appears to fall into the soda, create surface disturbance, then rise and settle with plausible buoyancy while fizz remains visible. Model B is visually clean, but the cube seems to sink and stay near the bottom with less believable bobbing behavior for ice in soda, reducing realism and prompt adherence. (Second judge pass, order swapped — scores are the average of both: Model A better matches the requested physics: the ice cube appears to rise and bob near the top with natural fizz and believable soda motion, while maintaining a consistent close-up kitchen look. Model B is visually clean, but the cube sinks and stays at the bottom rather than showing realistic buoyancy and settling behavior for ice in soda.)

2. Pearl diver consistency

A short continuous shot of a single pearl diver named Ilya Morren ascending a mooring line from a turquoise sinkhole to the sunlit surface, turning once to check his depth gauge and then swimming directly past camera; he must remain perfectly identical throughout the clip with the same narrow face, small notch in his left eyebrow, short copper-blond hair floating under a clear low-volume mask, faded plum rash guard with a hand-stitched white manta emblem over the heart, one teal fin and one black fin, and a scratched brass wrist compass on his right arm; the camera performs a smooth backward rise while keeping him in medium close-up, with dappled noon light, suspended silt, and a calm focused mood, 16:9

Winner: Seedance 2 Image to Video — Model B follows the prompt more closely with a clearer ascent on the mooring line, visible depth-check action, mismatched teal/black fins, and stronger underwater cinematography, while keeping the diver fairly consistent. Model A has decent close-up continuity and the right general wardrobe, but misses key details like the fins and compass placement/appearance, and its action staging is less faithful to the requested shot progression. (Second judge pass, order swapped — scores are the average of both: Model B follows the prompt much more closely: the diver ascends a mooring line in a coherent underwater shot, checks his wrist instrument, and preserves key wardrobe details including the mismatched teal and black fins and plum rash guard. Model A has decent facial continuity and surface lighting, but it misses major prompt elements such as the sinkhole ascent framing, fin visibility, correct emblem/compass details, and overall motion/camera setup consistency.)

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: Seedance 2 Image to Video — Model B matches the prompt more closely with a clear forward glide from the entrance down the cathedral nave toward the altar in a convincing single continuous take, with strong dust-filled light and stable composition. Model A is visually rich and candlelit, but it appears more like a lateral or drifting view near the altar rather than a sustained entrance-to-altar move, so its prompt adherence and motion clarity are weaker. (Second judge pass, order swapped — scores are the average of both: Model B matches the prompt closely with a clear single forward glide from the cathedral entrance toward the altar, maintaining strong spatial continuity, warm candlelight, and visible dust-filled light beams. Model A is visually rich and consistent, but it appears to stay near the altar area with more lateral drift than an entrance-to-altar progression, so it adheres less well to the requested camera path.)

4. Kelp tea plume

A short continuous underwater clip inside a narrow basalt arch on the Faroe shelf: a dented brass samovar strapped to a scientific marker unexpectedly tips and releases a dense amber plume of tea into icy seawater while silver air bubbles stream upward, the liquid ribbons curl around swaying kelp fronds and split realistically against the rock, and a tiny orange squat lobster scuttles across the foreground; the camera makes a slow intimate close-up tracking move left-to-right and slightly forward, following the plume as it disperses, with cold blue-green shafts of morning light, crisp particulate detail, and a hushed eerie mood, 16:9

Winner: Tie — Model B better matches the prompt’s intimate underwater close-up inside a basalt arch, with a clearly dented brass samovar strapped to a marker, convincing tea plume dispersion, kelp interaction, bubbles, and stronger cinematic lighting. Model A captures the orange squat lobster and a plausible plume, but the vessel reads less like the specified samovar setup and the composition feels less faithful and polished overall. (Second judge pass, order swapped — scores are the average of both: Model A matches more of the prompt-specific action: the samovar appears strapped to a marker, the amber tea plume streams and disperses through the narrow basalt passage with visible bubbles, and the orange squat lobster is clearly present in the foreground. Model B is atmospheric and attractive, but it misses the lobster, shows a less convincing strapped setup, and the plume behaves more like a soft cloud than liquid ribbons splitting realistically against rock.)

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