Head to head: Luma Ray 3.2 Image to Video vs Pika Image to Video (v2.2)
Luma Ray 3.2 Image to Video vs Pika Image to Video (v2.2)
By Ryan Merket · Published
This matchup tests whether stronger prompt execution can outweigh cleaner continuity across lighting changes, animal action, and cinematic camera movement. The results expose sharply different strengths—and recurring failure modes—in both models.
Pika posts the higher aggregate score, 26.7 to Luma’s 24.8, but that 1.9-point gap is not statistically persuasive. The confidence that either model is genuinely better is only 50%, making the topline result an effective tie rather than a defensible Pika victory. Pika’s clearest advantage is directed visual progression. It swept the lighting-transition tests by making the warm-to-cool dusk shift and lamp activation unmistakable, and it generally handled the continuous church glide better, with stronger forward movement, architectural continuity, haze, and candlelit atmosphere. The tradeoff was occasional overstatement: cyan-heavy grading, lamps or props appearing abruptly, and motion that satisfied the prompt without always looking physically credible. Luma was more competitive when object continuity and chaotic action mattered. It took two of the three terrier trampoline trials, better preserving the dog, crate, trampoline, watering can, and rebound sequence while Pika produced multiplying cans, disappearing animals, or reversed-looking motion. The spaniel tests were less decisive: Pika more often conveyed the actual shake and spray, while Luma sometimes maintained a clearer subject but substituted running splashes for the requested full-body shake—and even let camera equipment intrude into frame. **Final call: too close to call. Pika has the nominal score edge and better prompt legibility in lighting and camera-movement tasks; Luma can be more coherent in complicated physical sequences. At limited confidence, these models are effectively even here.**
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-5.6 Sol Pro: Model B more clearly depicts the requested warm-to-cool dusk transition in a stable locked-off room, though the final cyan cast is overly intense and the lamp glow becomes less apparent. Model A is aesthetically moody, but begins already cool and dark, and the lamp fixture appears abruptly rather than merely flickering on, weakening temporal consistency and prompt adherence. (Second judge pass, order swapped — scores are the average of both: Model B more clearly depicts the requested warm-to-coo | Anthropic: Claude Fable 5.1: Model A delivers a believable, cinematic l…
Spaniel dock shake spray
A single continuous 16:9 shot on a misty dawn lakeside dock: a soaked chestnut cocker spaniel bursts up from the water clutching a neon yellow training dummy, scrambles onto the wet cedar planks, and performs a full-body shake that sends concentric sprays, suspended droplets, and fine mist arcing outward while ripples slap the dock posts and a scatter of tiny mayflies hovers above the surface; the camera makes a gentle handheld push-in from knee height, tracking the dog’s head and shoulders as the water particles and lake surface behave naturally across the clip, lit by cool blue early-morning light with a calm, intimate mood.
OpenAI: GPT-5.6 Sol Pro: Model B more clearly depicts the full-body shake with an outward arc of droplets and has stronger scenic composition, natural lighting, and subject consistency, though the dog resembles a retriever and is already on the dock. Model A better shows the water-to-dock scramble, but the visible camera operator, deformed dummy, rain-like particles, and lack of a clear shake reduce fidelity and aesthetics. (Second judge pass, order swapped — scores are the average of both: Model B better captures the m | Anthropic: Claude Fable 5.1: Model B delivers the core shake with arcin…
Single continuous shot
One unbroken take gliding slowly through a candlelit cathedral from the entrance toward the altar, no cuts, jumps, or transitions, dust and warm light in the air, 16:9.
OpenAI: GPT-5.6 Sol Pro: Model B more clearly progresses from the entrance toward the altar while maintaining strong architectural continuity, warm candlelight, atmospheric haze, and excellent visual fidelity. Model A is also cohesive and aesthetically strong, but its forward movement is less pronounced and the framing feels less explicitly entrance-originated. (Second judge pass, order swapped — scores are the average of both: Model B more clearly depicts a continuous forward glide from the entrance toward a prominent | Anthropic: Claude Fable 5.1: Model B delivers a clear, unbroken glide fr…
Terrier trampoline crate drop
A single continuous 16:9 shot at golden-hour in a cluttered community-garden tool shed: a wiry brindle terrier wearing a faded teal bandana springs off a low overturned seed crate onto a sagging mini trampoline, accidentally knocking a dented aluminum watering can from a shelf; the can tumbles, hits the trampoline edge, and ricochets to the dusty floor while the dog lands, rebounds once, skids, and regains balance with believable weight, momentum, and gravity, sending a few seed packets fluttering; the camera starts at paw level and slowly dollies sideways to keep both dog and falling can framed, with warm slanting sunlight through plank gaps, long shadows, and a comic-but-natural mood.
OpenAI: GPT-5.6 Sol Pro: Model A delivers a more coherent rebound-and-skid sequence with warmer shed atmosphere, though the watering-can event is absent and the dog distorts in the final frame. Model B includes more requested props and the crate setup, but the oversized flying can, disappearing dog, and reversed-looking action severely undermine believable motion and continuity. (Second judge pass, order swapped — scores are the average of both: Model B includes more of the specified crate, trampoline, watering can, see | Anthropic: Claude Fable 5.1: Model A nails the golden-hour cluttered sh…
Matchup powered by OpenRouter.