Head to head: Ltx 2.3 Quality vs Seedance 2.5 Reference to Video
Ltx 2.3 Quality vs Seedance 2.5 Reference to Video
By Ryan Merket · Published · Updated
A four-scene comparison tests how well each model handles a dusk lighting change, a puffin’s cliff dive, latte art, and herons in a mangrove creek. The results reveal different strengths in motion, composition, and prompt detail.
Seedance 2.5 Reference to Video earns more of the individual task calls, especially on the puffin, latte-art, and mangrove scenes. It more often carries the puffin from the ledge into flight, forms a recognizable rosetta during a continuous pour, and renders convincing herons and mangrove roots. Those wins come with misses: the puffin can start small or unclear, the latte-art angle is sometimes less overhead, and the mangrove scenes omit much of the requested flock behavior or end on a distracting crop. Ltx 2.3 Quality has a real counterpoint in the lighting tests. In one, its wider locked composition makes the sunset fading into blue dusk—and the lamp turning on—more legible. But Seedance handles that transition more smoothly in other runs, while Ltx sometimes shifts abruptly or never turns on the lamp. This is a close, prompt-specific split, not a consistent advantage for either model. The aggregate scores favor Seedance, 30.7 to 23.4, but the reported confidence is only 50% that either model is genuinely better. The task-level pattern is worth noting; it is not enough to turn a sample tie into a dependable overall ranking. **Final call: Tie. These models are effectively even here: Seedance more often nails the requested action and subject detail, while Ltx can be stronger on composition and a clearly staged lighting transition.**
Lighting transition
A living room at dusk as the sun sets over a few seconds: the warm light visibly dims and shifts to cool blue while a lamp flickers on, the transition smooth and believable, locked-off frame, 16:9.
OpenAI: GPT-6 Luna: Model A more clearly shows the sunset through the windows fading into cool blue while the lamp turns on, all in a consistent locked-off wide composition. Model B has an appealing warm-to-cool shift, but its tighter framing makes the sunset transition less explicit. (Second judge pass, order swapped — scores are the average of both: Model A shows a clear progression from sunset through cool dusk to night, with the living-room lamp turning on as the daylight fades. Model B also has a strong warm-t | Anthropic: Claude Fable 5.1: Both deliver a locked-off warm-to-cool dusk tra…
Puffin Cliff Dive Run
A single continuous 16:9 shot begins on a wind-battered basalt ledge and whip-pans into a low tracking move beside an Atlantic puffin sprinting awkwardly but furiously toward the cliff edge before launching into a fast sea-skimming dive above churning indigo water, wings beating at high speed with believable motion blur and a strong sense of acceleration as spray streaks past and the background cliffs smear from momentum; late golden-hour sunlight flashes on the beak and wave tops, the camera banks slightly to stay with the bird in one unbroken shot, and the mood is exhilarating, urgent, and salt-sharp.
OpenAI: GPT-6 Luna: Model A more clearly shows the full run-to-launch progression, with the puffin, cliff backdrop, and spray remaining legible in a cinematic golden-hour scene. Model B conveys a fast bank and sea-skimming flight, but the bird becomes very small and the opening frame is noticeably less clear. (Second judge pass, order swapped — scores are the average of both: Model A more clearly shows the puffin running on the ledge, launching, and skimming low over the water with spray, matching the requested pro | Anthropic: Claude Fable 5.1: Model A delivers the full arc — awkward sprint…
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 forms a clear, finely layered rosetta-like pattern with a continuous, natural-looking pour, while Model A mostly produces concentric rings rather than a rosetta. Model A is more overhead, but Model B’s stronger latte-art result outweighs its slightly lower camera angle. (Second judge pass, order swapped — scores are the average of both: Model B develops a clean, detailed rosetta-like pattern in a close overhead view, with a coherent pour across the frames. Model A pours smoothly but form | Anthropic: Claude Fable 5.1: Model B actually produces a clean, recognizable…
Mangrove Heron Tangle
A single continuous 16:9 shot glides sideways at knee height along a copper-tide mangrove creek at blue dawn as eleven tricolored herons and four white ibises forage independently across slick roots and shallow ripples—one bird hops to avoid a fiddler crab, two briefly cross paths in opposite directions, another flares its wings to land on a leaning root, and a juvenile darts through the foreground—while every neck, leg, reflection, and wake remains distinct with no merging or warping; the camera makes a gentle handheld push-in near the end, cool misty backlight catches droplets and wet bark, and the mood is tense, watchful, and alive with layered natural motion.
OpenAI: GPT-6 Luna: Model A includes a crab and a wing-flaring bird, but its highly stylized, sparse scene and awkward reflections undermine the naturalistic prompt. Model B has convincing mangrove roots, herons, and a continuous camera glide, though it shows too few birds and ends with a distracting crop. (Second judge pass, order swapped — scores are the average of both: Model B better captures the dense mangrove creek and tricolored herons, with a low lateral camera move and a late push-in, though it misses the | Anthropic: Claude Fable 5.1: Model B delivers a true mangrove-root creek with…
Matchup powered by OpenRouter.