Head to head: AnimateDiff vs Marey Realism V1.5
This matchup wasn’t subtle: Marey Realism V1.5 swept all four tasks and did it with a decisive statistical margin. AnimateDiff had flashes of visual appeal, but across prompt adherence, motion, and scene construction, Marey was the model that more reliably delivered the shot that was actually asked for.
By RuntimeWire · Published

Marey Realism V1.5 takes this one cleanly. The aggregate score gap — 33.3 to 17.5 — is already lopsided, and the statistical read makes it plain: 97% confidence, decisive. Just as important, the task board is a shutout: Marey wins 4 of 4, AnimateDiff wins 0.
What stands out is not just image quality, but obedience to the brief. In physics realism, Marey was judged as the version that actually looked like a tall glass of soda with fizz, splash, and believable ice-cube behavior under soft kitchen light; AnimateDiff’s result drifted toward generic clear liquid and muddier action. In the terrier tennis-ball prompt, the same pattern repeated: Marey better preserved the teal rooftop, violet ball, metal bowl, bandana, and the sense of an energetic chase, while AnimateDiff mostly defaulted to a dog-walking shot and dropped the key props.
The temporal consistency result is especially telling. AnimateDiff wasn’t bad here — it was described as attractive and fairly stable — but Marey more convincingly held onto the man’s face, raincoat, umbrella, and toward-camera motion in the rainy tracking shot. That’s the difference between a model that can generate a nice clip and one that can sustain a specific shot direction over time.
The misty estuary heron task seals the editorial verdict. Marey at least delivered the estuary mood, boardwalk, fog, and water detail, even if the heron’s motion was limited. AnimateDiff, by contrast, was dinged for missing the core subject entirely in the main pass. And while the order-swapped notes show some judge instability on individual prompts, they don’t change the big picture: across the full evaluation, Marey was the model that more often landed the requested subject, setting, and action together.
Final call: Marey Realism V1.5 is the clear winner. AnimateDiff can produce appealing imagery, but in this head-to-head it was the less dependable model; Marey was simply better at turning prompts into the specific videos users asked for.
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. AnimateDiff scored 17.5 to Marey Realism V1.5's 33.3.
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: Marey Realism V1.5 — Model B matches the prompt much better with a tall glass of soda, visible fizz, splash, and believable bobbing/settling of a single ice cube under soft kitchen lighting. Model A is visually detailed but appears more like underwater ice in clear liquid, with weak soda cues and poor adherence to the requested action and buoyancy behavior. (Order-swapped judge pass: Model A clearly matches the prompt with a single ice cube in a tall glass of soda, showing a plausible splash, bobbing near the surface, and rising fizz under warm kitchen light. Model B looks like multiple ice chunks in clear liquid rather than soda, with weak event progression and poor adherence to the requested action and setting.)
2. Temporal consistency
A man in a yellow raincoat walking toward camera down a rainy street; his face, coat, and umbrella must stay perfectly consistent with no morphing or flicker from the first frame to the last, steady tracking shot, 16:9.
Winner: Marey Realism V1.5 — Model B better matches the prompt with a clear steady tracking shot of a man walking toward camera in a yellow raincoat under an umbrella, while maintaining stronger facial and wardrobe consistency across frames. Model A is also consistent and attractive, but the subject appears more side-turned and less clearly advancing toward camera, with softer facial definition and slightly less convincing motion progression. (Order-swapped judge pass: Model A better matches the prompt with a steady 16:9 tracking shot of a man in a yellow raincoat walking toward camera, and his face, coat, and umbrella remain more consistent across frames. Model B has pleasing rainy-street imagery, but it is portrait-framed rather than 16:9 and shows softer facial consistency and less direct toward-camera motion.)
3. Misty estuary heron walk
At blue-hour dawn on a narrow cedar boardwalk above the Brindlemere estuary, a speckled grey heron takes slow deliberate steps past clumps of salt grass while the whole setting stays alive: low fog drifts in ribbons over the water, distant reeds sway irregularly in the breeze, tiny ripples spread and overlap around half-submerged posts, and faint steam rises from a moored crab pot; the camera makes a smooth gliding gimbal push forward at ankle height, tracking beside the bird in one continuous shot, cool silver light with peach highlights reflecting off the water, calm hushed mood, 16:9
Winner: Marey Realism V1.5 — Model B matches the misty blue-hour estuary mood and includes the heron on a boardwalk with pleasing water, fog, and reeds, though the bird appears mostly stationary rather than walking beside a tracking camera. Model A is temporally stable but misses the core subject entirely, showing only a boardwalk landscape without the heron or the specified close tracking action. (Order-swapped judge pass: Model A includes the heron, misty estuary mood, dawn lighting, and water detail with strong visual appeal, though it misses the specified walking/tracking action and boardwalk context. Model B shows a boardwalk and estuary but omits the heron entirely and feels more like a static aerial push than the requested ankle-height tracking shot.)
4. Terrier tennis ball rebound
On a sunlit rooftop agility yard with faded teal rubber flooring, a wiry cinnamon terrier sprints after a scuffed violet tennis ball that drops from offscreen, strikes the surface, rebounds lower on each bounce, then glances off a shallow metal water bowl and skitters into the dog's paws as a loose bandana on its neck flutters from its speed; the camera performs a low handheld-style lateral tracking move that keeps pace with the chase in one continuous shot, crisp late-afternoon light casting sharp shadows, playful energetic mood, 16:9
Winner: Marey Realism V1.5 — Model B matches the prompt far better by clearly showing the rooftop teal flooring, violet tennis ball, metal water bowl, bandana, and energetic chase with stronger cinematic framing. Model A is temporally stable but largely misses the key action and props, showing mostly a dog walking toward camera without the ball or bowl. (Order-swapped judge pass: Model A matches the rooftop teal flooring, terrier, violet ball, metal bowl, bandana, and energetic chase much more closely, with stronger cinematic lighting and implied tracking motion. Model B is temporally stable but misses key prompt elements like the ball, bowl, chase action, and handheld lateral tracking, and the dog design is less faithful.)
See every prompt and the full side-by-side outputs in the interactive Head-to-Head.