Head to head: Happy Horse vs LTX Video-0.9.7 13B Distilled
Happy Horse vs LTX Video-0.9.7 13B Distilled
By Ryan Merket · Published
One model consistently understood the assignment; the other mostly delivered cleaner-looking approximations. Across every judged task, Happy Horse beat LTX Video-0.9.7 13B Distilled on prompt fidelity, scene logic, and motion behavior.
This one isn’t a nail-biter. **Happy Horse wins outright, and the numbers back it up**: 35.5 to 26.0 overall, a clean 8-0 task sweep, and a 95% confidence verdict. When a matchup lands this decisively, the interesting question isn’t who won — it’s why the loser never found a lane. The pattern was brutally consistent: Happy Horse actually rendered the scenes it was asked for. In **Kelp Lab Drift**, it delivered the flooded abandoned marine lab the prompt described — kelp outside shattered windows, suspended silt, fish, rusted racks, bubbling gauges, and a believable sideways underwater glide. LTX’s version was smoother and cleaner in places, but it kept missing the point, reading more like a dry, sunlit corridor than an eerie submerged lab. The same thing happened in **Harbor Buoy Bounce**. Happy Horse nailed the chipped orange buoy, taut rope leash, rust-streaked breakwater setting, dawn mood, and the crucial half-submerged camera behavior. LTX produced attractive water and stable imagery, but it flattened the setup into something generic and stylized, with less convincing buoy-water interaction and little of the prompt’s specific physical staging. On **Crowd motion** and **Temporal consistency**, Happy Horse widened the gap rather than merely holding serve. Its Tokyo scramble crossing had actual density, multidirectional pedestrian flow, and better preservation of separate individuals without obvious merging. Its rainy-street tracking shot kept the man’s face, yellow raincoat, and black umbrella coherent across frames while preserving the intended atmosphere. LTX was often visually tidy, but too sparse, too glossy, and too willing to substitute a clean approximation for the requested scene. **Final call: Happy Horse is the better video model here by a clear margin — not because it’s flashier, but because it is far more reliable at turning detailed prompts into believable moving images.**
Kelp Lab Drift
A short continuous shot in the flooded remains of the abandoned Nereid-4 marine botany lab, now open to the sea: long ribbons of giant kelp trail through shattered window frames, tiny silver fry pulse in loose schools, silt lifts in faint curls from the tiled floor, and threads of sunlight from the late afternoon surface shimmer across rusted instrument racks while a thin stream of trapped air pearls upward from a cracked pressure gauge; the camera performs a slow fluid gliding gimbal move sideways through the corridor, passing a barnacle-covered desk as all the background motion continues naturally and unbroken; the lighting is green-gold and dappled, and the mood is eerie but calm, 16:9
Model A matches the flooded abandoned marine lab prompt much more closely, with kelp outside shattered windows, suspended silt, fish, rusted equipment, bubbles from a gauge, and a believable sideways glide through the corridor. Model B has a cleaner and more stable lateral move, but it misses key underwater atmosphere and ecological details, looking more like a dry sunlit hallway than an eerie submerged lab. (Second judge pass, order swapped — scores are the average of both: Model A matches the prompt much more closely, showing a flooded abandoned lab with kelp, fish, silt, rusted racks, pressure gauges, rising air bubbles, and a coherent sideways glide under green-gold underwater light. Model B has decent smooth motion and clean imagery, but it feels too dry and pristine, missing key underwater details and the eerie marine-lab atmosphere described.)
Harbor Buoy Bounce
A short continuous shot at dawn inside a turquoise tidal pool beside the rust-streaked breakwater of Port Cadris: a chipped orange inspection buoy, about the size of a basketball, is dropped from a wet basalt ledge into the pool, hits the water, bobs hard, ricochets off the rock wall, and rebounds with smaller and smaller oscillations while droplets arc and fall back under gravity and the taut rope leash snaps it slightly sideways; the camera makes a slow fluid gliding gimbal push from just above the waterline to half-submerged, tracking the buoy’s motion without cuts; cold pink sunrise light glints on the ripples, and the mood is hushed, precise, and observational, 16:9
Model A matches the specified harbor setting, chipped orange inspection buoy, rope leash, half-submerged camera angle, and observational dawn mood much more closely, with convincing splash-to-bob progression and stable geometry. Model B has pleasing color and readable ripples, but it misses the breakwater/basalt pool setup, lacks the requested camera behavior and ricochet feel, and looks more stylized and less physically grounded. (Second judge pass, order swapped — scores are the average of both: Model A matches the prompt more closely with a rust-streaked breakwater setting, a chipped orange buoy on a taut rope, and a camera move that reaches a convincing half-submerged perspective while preserving coherent buoy-water interaction. Model B is attractive and temporally stable, but it misses key specifics like the breakwater/basalt context, the ricocheted rebound behavior, and the requested half-submerged tracking shot.)
Crowd motion
A busy Tokyo scramble crossing seen from above, dozens of pedestrians crossing in different directions, each moving independently without merging or warping into one another, overcast daylight, 16:9.
Model A matches the prompt much better with a clear overhead view of a Tokyo-style scramble crossing, many pedestrians moving in multiple directions, and believable independent spacing without obvious merging. Model B has cleaner graphic contrast, but it shows far fewer pedestrians, a less convincing Tokyo setting, and more stylized, less natural motion and crowd behavior. (Second judge pass, order swapped — scores are the average of both: Model A matches the prompt much better with a true overhead Tokyo scramble crossing, many pedestrians moving in multiple directions, and cleaner independent trajectories with strong scene coherence. Model B has decent visual clarity, but it shows a simpler single intersection with sparse foot traffic and less convincing scramble-crossing behavior, so it adheres less well to the prompt and feels less dynamic.)
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.
Model A matches the prompt closely with a rainy urban street, steady approach toward camera, and strong consistency in the man's face, yellow raincoat, and black umbrella across frames. Model B is visually clean but deviates from the prompt with a different umbrella color/style and a less convincing rainy tracking-shot feel, though its subject remains fairly stable. (Second judge pass, order swapped — scores are the average of both: Model A better matches the prompt with a steadier tracking shot and stronger temporal consistency in the man's face, raincoat, and umbrella across frames. Model B is bright and appealing, but the umbrella framing and facial details shift more noticeably and the overall motion feels slightly less coherent.)
Matchup powered by OpenRouter.