Head to head: AnimateDiff Turbo vs Happy Horse 1.1 Image to Video
AnimateDiff Turbo vs Happy Horse 1.1 Image to Video
One model showed up to the brief; the other mostly showed style. Across all four tasks, Happy Horse 1.1 Image to Video delivered the more usable video model by a wide and statistically decisive margin.
This one is not a nail-biter. **Happy Horse 1.1 Image to Video wins decisively**, taking all four tasks and posting a **33.8 to 7.2** aggregate score advantage. The statistical read is even harsher: **97% confidence** that this is the real ordering, not noise. The pattern is simple: Model B more often produced actual prompt-following video, while AnimateDiff Turbo too often collapsed into attractive but irrelevant imagery. In **glassblower continuity**, that gap was brutal—Happy Horse delivered a coherent workshop action sequence, while AnimateDiff looked like a static stylized portrait that barely engaged with the assignment. In **night-shift fish market flow**, the story repeated: B gave a recognizable market aisle with multiple actors and plausible concurrent motion; A again drifted into an unrelated character shot. The more technical tests reinforced the same verdict. In **camera motion control**, Happy Horse was the one that actually executed the requested orbit around the ramen bowl with meaningful viewpoint change. AnimateDiff remained mostly still, which is a fatal miss on a camera-move prompt. And in **physics realism**, B was credited with the more believable soda, ice-cube entry, splash, buoyancy, and fizz behavior, while A’s liquid and splash logic were judged inconsistent. Yes, the order-swapped judge notes show some instability in individual writeups, so this was not a case where every observer saw the exact same failure mode every time. But the aggregate outcome is still unambiguous: **0 task wins for AnimateDiff, 4 for Happy Horse, 0 ties**. When one model repeatedly delivers scene structure, motion, and prompt adherence while the other keeps defaulting to stylization without execution, the editorial verdict is easy. **Final call: Happy Horse 1.1 Image to Video is the clear winner. AnimateDiff Turbo may generate striking frames, but on this test it was not a serious competitor as a controllable video model.**
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.
Model B clearly matches the prompt with a tall glass of soda in a kitchen, a single ice cube entering, splashing, then floating with plausible buoyancy and visible fizz. Model A is visually striking but fails the prompt and physics realism with an implausible blue liquid, unclear ice behavior, and inconsistent splash geometry. (Order-swapped judge pass: Model A clearly matches the prompt with a tall glass of soda, a single ice cube that splashes, rises with plausible buoyancy, and visible fizz in a softly lit kitchen. Model B deviates strongly from the prompt with an incorrect glass/drink appearance and less believable, less informative physics despite decent visual style.)
Glassblower continuity
Temporal consistency — In a single continuous 8-second shot, a middle-aged glassblower named Mirek Voss in a cobalt-blue canvas apron with a stitched white crane on the chest, round amber safety glasses, a soot mark on his left cheek, and a silver wedding band on his right hand steadily turns a glowing pear-shaped glass vessel on a blowpipe while walking three steps from the furnace to the marver table, never changing appearance, clothing, or identity for a moment; the camera makes a slow fluid gliding gimbal arc from his front-left to directly beside him at waist height, keeping him centered as heat shimmer ripples in the workshop air; warm orange furnace light mixes with cool skylight from high windows, creating a focused, industrious mood, 16:9
Model B clearly depicts a continuous glassblowing action in a workshop with coherent movement from furnace toward the table and strong visual consistency. Model A is essentially a static stylized portrait that misses most prompt details, action, and camera motion. (Order-swapped judge pass: Model A closely matches the prompt with a realistic middle-aged glassblower, correct workshop setting, glowing pear-shaped vessel on a blowpipe, and coherent movement across frames. Model B is a static stylized portrait of a young person with incorrect identity, setting, action, and no meaningful motion or shot continuity.)
Night-shift fish market flow
Crowd & multi-subject motion — One continuous 10-second shot inside the pre-dawn auction lane of Brine Hall Market, where seven seafood workers in distinct gear move independently and plausibly at once: a woman in lemon-yellow waterproof bibs drags a blue pallet jack toward camera, a man in a red knit cap hoists a silver hake crate onto a dolly, two clerks in gray smocks weave past each other carrying clipboards, a buyer in a pine-green raincoat sidesteps a puddle while raising his phone flashlight, and two porters in mismatched rubber boots roll separate carts in opposite directions without colliding or merging; the camera performs a smooth forward gimbal glide down the wet aisle with a slight rightward drift, passing hanging number signs and melting ice; harsh fluorescent ceiling lights reflect in the puddles with faint magenta dawn leaking through the loading-bay door, creating a brisk, high-pressure working mood, 16:9
Model B clearly depicts a fish-market aisle with multiple workers, wet reflective flooring, numbered signs, and plausible independent motion close to the prompt, while Model A is an unrelated stylized close-up of a single character with essentially no prompt match. B is not perfect because some specified subjects/actions are missing or simplified, but it is far superior overall. (Order-swapped judge pass: Model A clearly depicts a fish-market aisle with multiple workers, wet reflective flooring, signage, carts, and plausible concurrent activity that broadly matches the prompt. Model B is an unrelated static animated portrait with no market-lane crowd motion or requested camera/action details.)
Camera motion control
A single smooth orbit (roughly 180 degrees) around a steaming bowl of ramen on a wooden table, the camera circling at a constant speed and height in one continuous move, warm restaurant light, 16:9.
Model B clearly matches the prompt with a realistic steaming ramen bowl in warm restaurant lighting and a smooth lateral orbit that changes viewpoint substantially while maintaining height. Model A is stylized and nearly static, with minimal camera movement and weak evidence of a continuous 180-degree orbit. (Order-swapped judge pass: Model A clearly depicts a realistic steaming bowl of ramen on a wooden table with warm restaurant lighting and shows meaningful camera orbit progression across frames. Model B is stylized/illustrated, barely changes viewpoint, and does not convincingly realize the requested smooth 180-degree camera move or restaurant-table scene.)
Matchup powered by OpenRouter.