Head to head: Boreal vs Wan v2.2 5B
Boreal vs Wan v2.2 5B
By Ryan Merket · Published · Updated
Boreal and Wan v2.2 5B face off on four demanding video tasks, from latte art to a moving cathedral shot. The comparison turns on a real trade-off: precision in staged action versus a stronger showing on a specific, recognizable subject action.
This is a sample tie, not a narrow win dressed up as a verdict. Boreal scores 26.3 to Wan v2.2 5B’s 24.8, but the evaluation puts only limited confidence behind either model being genuinely better. That gap is too small to call a winner. The clearest split is in the coffee pours. Wan more often produces a recognizable rosetta in a close view of a cup, while Boreal tends toward concentric rings, a wider vessel, or—in one case—a spill. If latte art is the test, Wan has the more convincing results. Boreal is stronger on the other recurring challenge: making the requested action and camera movement legible. Its heel shots more faithfully show the riveted magenta shoe striking the copper tray and sending water and petals upward; its cathedral clips actually glide toward the altar, where Wan’s are often nearly static. The ribbon-knot task is less decisive: Boreal better evokes the backstage setting and camera drift, but neither model reliably shows the ribbon being tied or the marking turn. These are meaningful differences in what each model gets right, not evidence that one is reliably ahead overall. **Final call: tie. Boreal and Wan v2.2 5B are effectively even here—choose Wan for latte-art detail, Boreal for action fidelity and continuous camera movement.**
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.
OpenAI: GPT-6 Luna: Model B has a tighter, more overhead close-up of a standard coffee cup, and the continuous pour creates a more recognizable latte-art pattern, though neither video clearly forms a clean rosetta. Model A is also visually coherent, but its wider bowl and concentric-ring pattern are less aligned with the prompt. (Second judge pass, order swapped — scores are the average of both: Model B is closer to the requested close-up of pouring into a cup, with a coherent stream and warm café lighting, though | Anthropic: Claude Fable 5.1: Model B shows a more convincing overhead close-u…
Backstage ribbon knot
In one continuous 16:9 shot, a contemporary dance student in a scuffed silver rehearsal costume crouches backstage and quickly threads then ties a pale blue satin ribbon around her worn ballet flat before rising into a small marking turn, while the camera pushes in handheld from waist height over her right shoulder to a close view of her fingers, then drifts up with her as she stands, preserving natural fluid body mechanics and precise hand gesture detail; lit by half-open stage work lights and a thin strip of amber footlight leak, the mood is focused, nervous, and intimate.
OpenAI: GPT-6 Luna: Model B more clearly follows the ribbon-handling-to-standing progression and maintains a consistent backstage setup, though neither entry clearly shows a completed tie or marking turn. Model A has more cinematic lighting and a closer camera drift, but the ribbon action and hand detail are less legible. (Second judge pass, order swapped — scores are the average of both: Model B better matches the dark backstage setting, amber footlight, silver rehearsal costume, and handling of the pale blue ribb | Anthropic: Claude Fable 5.1: Model A delivers the handheld push-in and drift…
Rivet heel burst
In a single continuous 16:9 slow-motion shot, a flamenco soloist on a rain-dark rooftop stage snaps a rivet-studded magenta heel into a shallow copper tray, sending a crown of bronze-tinted water, glitter, and three loose rose petals upward as the camera makes a tight low-angle arc from her left ankle toward the front of the tray, capturing every droplet, petal wobble, and vibrating fringe detail with crisp fluid frame-to-frame motion; lit by warm string bulbs and a cold cyan sign spill, the mood is fierce and electric.
OpenAI: GPT-6 Luna: Model A more closely matches the rivet-studded magenta heel and tight, low-angle rooftop setup, with a readable splash-and-lift sequence. Model B has a more dramatic upward burst, but its shoe lacks the visible studs and the framing feels more static. (Second judge pass, order swapped — scores are the average of both: Model B better captures the rooftop setting, cyan sign and warm string bulbs, and delivers a more dramatic upward burst of water and petals from the copper tray. Model A shows the | Anthropic: Claude Fable 5.1: Model A delivers a grounded low-angle shot of a…
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.
OpenAI: GPT-6 Luna: Model A visibly glides forward through the cathedral, gradually revealing the altar in a continuous, coherent movement. Model B has beautiful candlelit haze and a strong altar composition, but its sampled frames appear nearly static rather than showing the requested approach. (Second judge pass, order swapped — scores are the average of both: Model B has a strong candlelit cathedral composition with visible haze, but its sampled frames are nearly identical and do not convey the requested glide. | Anthropic: Claude Fable 5.1: Model A delivers an actual continuous glide forw…
Matchup powered by OpenRouter.