Head to head: Ltx 2.3 Quality vs MiniMax H3 Reference to Video
Ltx 2.3 Quality vs MiniMax H3 Reference to Video
By Ryan Merket · Published
This matchup tests whether prompt coverage and occasional spectacle can outweigh temporal coherence, physical credibility, and cinematic control across crowded action, particle effects, fluid simulation, and continuous camera movement.
MiniMax H3 Reference to Video wins decisively: **30.8 to 23.9**, with **10 task wins against Ltx 2.3 Quality’s one**, plus one tie. The limited confidence verdict confirms that this is a clear, repeatable advantage rather than a narrow scoring break. MiniMax’s edge was consistency in motion. Its cathedral sequences held architecture together through slow entrance-to-altar glides, while its ice-cube tests produced more credible impact, splash, buoyant return, and settling. In the canal scenes, it generally delivered cleaner independent movement, stronger wet blue-hour atmosphere, and more coherent runners, skaters, strollers, and marshals—even when the baton exchange or canal itself was not always explicit. Ltx could still hit individual prompt details harder. It earned its lone win by better showing the canal, dense relay crowd, low tracking angle, neon reflections, stroller, and marshals, and it shared the tie when its dramatic chalk burst and loose markers offset weaker continuity. But those strengths repeatedly came with merged participants, synthetic lighting, unstable fixtures, distorted fluid geometry, or outright temporal failures such as a cyclist disappearing and leaving an upright bicycle behind. **Final call: MiniMax H3 Reference to Video is the clear winner. It is substantially more dependable when a shot must remain physically plausible, spatially coherent, and cinematic from one frame to the next.**
Canal relay crossing
A single continuous 16:9 shot at blue-hour after a light rain follows a low gimbal camera gliding diagonally along a canal-side promenade as nine mixed-age relay runners in mismatched club colors weave through a crowded nighttime charity baton exchange, with two inline skaters cutting past, a stroller jogger angling around a chalk arrow, three volunteer marshals waving LED batons, and spectators leaning over wet barricades; every person moves independently and plausibly without tangling or merging as shoes splash through shallow puddles, neon shop signs reflect on the slick pavement, and the mood feels urgent, electric, and slightly chaotic under sodium streetlights and cool storefront glow.
OpenAI: GPT-5.6 Sol Pro: Model A better preserves the larger runner group and includes the stroller jogger, but omits the requested skaters and gives a less evident canal setting. Model B more convincingly depicts the canal-side route, wet blue-hour lighting, skater motion, marshals, and energetic crowd activity with stronger visual fidelity, though the runner count and baton exchange remain unclear. (Second judge pass, order swapped — scores are the average of both: Model B better captures the requested urgent, crowded | Anthropic: Claude Fable 5.1: Model B delivers a more photoreal, cohesiv…
Velodrome chalk plume
A single continuous 16:9 shot inside a dim industrial velodrome begins with a sweeping side-tracking camera pacing a lone track cyclist entering a steep bank, then arcing closer as the rider snaps into a hard skid-stop drill on the infield apron, sending a dense burst of pale training chalk and tire dust curling into layered vortices that trail, spread, and settle naturally in the wake of the bike while loose paper lane markers flutter and a few distant cyclists blur past high on the boards; harsh overhead mercury lamps carve bright rims through the airborne particles, and the mood is gritty, tense, and hyper-focused.
OpenAI: GPT-5.6 Sol Pro: Model A includes the loose markers and a dramatic chalk burst, but the plume becomes an implausible, disconnected dust column and the framing changes feel less continuous. Model B delivers more coherent tracking, lighting, atmosphere, and natural particle settling, though the skid action and paper markers are less explicit. (Second judge pass, order swapped — scores are the average of both: Model A more clearly depicts the side-tracked cyclist, scattered paper markers, and a forceful chalk plume | Anthropic: Claude Fable 5.1: Model A delivers the dense chalk plume and…
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 B shows a convincing single ice cube entering the soda, displacing liquid, rising with plausible buoyancy, and settling amid fine fizz, while maintaining a polished close-up composition. Model A has an unnaturally hollow-looking cube, oversized bubbles appearing outside the glass, and inconsistent fluid and object geometry that undermine physical realism. (Second judge pass, order swapped — scores are the average of both: Model B shows a convincing splash followed by coherent bobbing and s | Anthropic: Claude Fable 5.1: Model A shows a stylized splash with bubbl…
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-5.6 Sol Pro: Model B presents a more convincingly slow, continuous entrance-to-altar glide with exceptionally stable architecture, atmospheric light, and refined cathedral detail. Model A also shows clear forward progression, but stronger apparent acceleration, slight pew/fixture instability, and harsher highlights reduce realism and temporal polish. (Second judge pass, order swapped — scores are the average of both: Model B more convincingly depicts a slow, unbroken entrance-to-altar glide with consistent a | Anthropic: Claude Fable 5.1: Model B begins literally at the entrance d…
Matchup powered by OpenRouter.