Head to head: Happy Horse 1.1 Image to Video vs LTX 2.3 Video Fast
Happy Horse 1.1 Image to Video vs LTX 2.3 Video Fast
By Ryan Merket · Published
This matchup tests whether stronger prompt adherence and shot coherence can outweigh atmosphere and fluidity across demanding underwater, tracking, and camera-control scenes.
Happy Horse posts the higher aggregate score, 29.9 to 23.6, but that margin does not survive the uncertainty in the evaluation. With only limited confidence that either model is genuinely better, these results describe different strengths—not a defensible overall winner. Happy Horse was generally better at preserving the requested content. It kept the raincoat, face, and umbrella steadier in the tracking sequence; rendered the lanternfish bead fracture, magenta-edged slivers, and orange roe more legibly; and produced the more coherent ramen shot. Its camera control was still imperfect—the supposed 180-degree orbit often read as a push-in or changed height—but LTX suffered more obvious topping deformation and barely established the requested move. The manta test shows why the aggregate cannot settle the contest. Happy Horse more reliably checked the prompt’s species list, while LTX produced the more fluid, cinematic rush in another pass, with stronger depth, sunrise atmosphere, and forward motion. That split captures the broader trade-off: Happy Horse favors literal coverage and object continuity; LTX can look more naturally cinematic, but is less dependable about preserving every specified element. **Final call: tie. Happy Horse has the better raw score and more task-level nods, but at limited confidence the models are effectively even here—too close to call.**
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.
OpenAI: GPT-5.6 Sol Pro: Model A maintains a notably consistent face, yellow coat, and umbrella across the sequence while presenting a polished, steady-looking rainy tracking shot. Model B also adheres well, but the umbrella silhouette and framing vary more noticeably, and the face is less clearly preserved under the hood and deep shadow. (Second judge pass, order swapped — scores are the average of both: Model A maintains the man's face, bright yellow coat, and umbrella more consistently while showing a convincing forw | Anthropic: Claude Fable 5.1: Model A keeps the man's face, yellow coat,…
Manta Cleaning Rush
Crowd & multi-subject motion — One continuous 16:9 shot over a volcanic coral cleaning station at first light: three reef manta rays circle at different depths while a dozen electric-blue cleaner wrasse dart in and out, two Moorish idols weave across the foreground, a green sea turtle glides diagonally beneath them, and a loose cloud of tiny silver sweepers parts and rejoins around the larger bodies, all moving independently and plausibly without merging or warping; the camera performs a fluid gliding gimbal shot descending slowly forward between coral heads, warm sunrise shafts mixing with cool underwater haze, for a serene but bustling mood.
OpenAI: GPT-5.6 Sol Pro: Model A adheres much more closely by clearly depicting three mantas at varied depths, blue cleaner fish, Moorish idols, a turtle, and schooling silver fish with strong continuity across the forward glide. Model B is more cinematic and naturally atmospheric, but omits or fails to distinguish several required species and shows an excess of manta-like rays. (Second judge pass, order swapped — scores are the average of both: Model A adheres more closely to the specified ensemble, clearly depicting t | Anthropic: Claude Fable 5.1: Model A ticks more literal boxes (Moorish…
Lanternfish Bubble Burst
Slow-motion detail — One continuous 16:9 shot inside a twilight-blue kelp nursery tank: a striped juvenile lanternfish snaps at a suspended glass feeding bead, and the bead fractures into glittering slivers just as a trapped air bubble erupts through a veil of orange roe, every shard, bubble skin, and fin twitch rendered in convincing high-speed slow motion with crisp fluid frame-to-frame detail; the camera makes a gentle right-to-left gliding macro push-in with a slight upward arc, cool caustic light from above and a single magenta inspection lamp edging the fragments, creating a hushed, uncanny scientific mood.
OpenAI: GPT-5.6 Sol Pro: Model A closely depicts the striped fish, suspended glass bead, radiating slivers, orange roe, cool tank lighting, and magenta fragment edging with polished slow-motion imagery, though the event progression is slightly ambiguous. Model B has credible aquatic bubbles and lighting but largely omits the glass-bead fracture and clear roe-burst interaction, while the fish quickly leaves the composition. (Second judge pass, order swapped — scores are the average of both: Model A clearly depicts the su | Anthropic: Claude Fable 5.1: Model A delivers the core narrative — a st…
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.
OpenAI: GPT-5.6 Sol Pro: Model A is visually polished and temporally consistent with smooth framing changes, though the move reads more like a push-in than the requested 180-degree orbit. Model B shows limited convincing orbital movement and substantial topping deformation between frames, reducing motion quality and coherence. (Second judge pass, order swapped — scores are the average of both: Model A has stronger restaurant atmosphere, clearer steam, more convincing background parallax, and substantially better visual | Anthropic: Claude Fable 5.1: Model A renders a gorgeous, steaming ramen…
Matchup powered by OpenRouter.