Head to head: Kandinsky5 Pro vs LTX 2.5 Image to Video Pro

Kandinsky5 Pro never really gets on the board here. On aggregate, LTX 2.5 Image to Video Pro leads 32.6 to 28.3, takes 3 of 4 tasks outright, and the statistical read is decisive: a 94% confidence win. That’s not a vibes-based edge; it’s a clear result.

The pattern is consistent across the losses. In Camera motion control, LTX is plainly better at the requested constant-height orbit around the ramen, with real lateral viewpoint change rather than the slight drift Kandinsky offers. In Single continuous shot, LTX sells the entrance-to-altar cathedral glide as an actual traveling shot, with stronger forward motion and richer dust-and-light atmosphere, while Kandinsky feels narrower and more static. And in Marshlight Shift, LTX is again the more obedient model: the gentle orbit reads, the stepping behavior is more visible, and the warm-to-cool lighting transition actually develops across the clip instead of barely registering.

The only non-loss for Kandinsky is Reedfall Splash, and even that ends as a tie because the judges split on what mattered more. One pass preferred LTX’s mood, water interaction, and bog-level cinematography; the other favored Kandinsky’s steadier continuity, peat-bog setup, and more believable framing. But neither model truly nailed the shell’s tumble-bounce-splash beat, so this wasn’t a hidden win for Kandinsky so much as a shared miss with different strengths.

The editorial takeaway is simple: Kandinsky5 Pro can produce coherent, attractive shots, but too often it defaults to restrained motion and incomplete action when the prompt asks for choreography. LTX 2.5 Image to Video Pro is the more reliable director. It moves the camera when asked, carries subjects through time more convincingly, and better translates prompt structure into visible on-screen events.

Final call: LTX 2.5 Image to Video Pro wins clearly. Kandinsky5 Pro looks competent, but LTX is the model that more consistently turns a still image into an actual shot.

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. To cancel position bias, every task was judged twice — once in each presentation order — and every number reported here, including the headline totals, is the average of both passes. Kandinsky5 Pro scored 28.3 to LTX 2.5 Image to Video Pro's 32.6.

1. Reedfall Splash

One continuous 16:9 shot in a shallow peat bog at dawn: a moss-flecked Eurasian otter wriggles onto a tilted reed raft, dislodging a fist-sized blue freshwater mussel shell that tumbles off the slick edge, hits a lower mat of reeds, bounces once, then drops into tea-colored water with a believable splash and expanding rings while the otter startles back; the camera tracks in low and close, then subtly cranes downward to follow the shell’s fall, catching droplets, reed flex, and the shell’s changing speed under gravity; cool misty light, hushed but lively mood, no cuts.

Winner: Tie — Model B better matches the prompt’s low, close bog-level cinematography and the shell/fall setup, with stronger atmosphere and more believable water interaction; Model A is coherent but misses the key shell action, keeping the blue object perched while the otter simply enters the water. Model B still appears to simplify or omit the explicit bounce and splash-follow camera beat, but overall it adheres more closely and looks more polished. (Second judge pass, order swapped — scores are the average of both: Model A better matches the peat-bog setup, reed raft, and startled otter behavior with steadier visual continuity and more believable framing, though it still misses the shell’s full tumble-and-splash action. Model B has moodier cinematography, but the shell reads more like a glowing orb, the action progression is less clear, and temporal continuity is weaker.)

2. 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.

Winner: LTX 2.5 Image to Video Pro — Model B better matches the requested smooth orbit, showing clearer lateral viewpoint change around the ramen while maintaining steady height and strong visual clarity. Model A has appealing warm lighting and steam, but the camera movement appears minimal relative to the requested roughly 180-degree orbit, making it less adherent despite decent consistency. (Second judge pass, order swapped — scores are the average of both: Model B better matches the requested smooth constant-height orbit around the ramen, with clear lateral viewpoint change across frames while keeping the bowl stable and visually appealing. Model A has strong steam and mood, but the camera appears mostly static with only slight drift, so it adheres less well to the specified 180-degree circling move.)

3. 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.

Winner: LTX 2.5 Image to Video Pro — Model A matches the candlelit cathedral glide well and appears temporally stable, but Model B better conveys the full entrance-to-altar move in a single continuous shot, with stronger visible dust/light beams and a more dramatic sense of forward motion. Model B is also more visually striking overall, though slightly brighter and less purely candlelit than Model A. (Second judge pass, order swapped — scores are the average of both: Model B better matches the prompt with a clear entrance-to-altar glide in a grand candlelit cathedral, strong warm volumetric light, and convincing forward motion across frames. Model A is also coherent and attractive, but it feels darker and more static, with less evident dust/light atmosphere and a narrower sense of traveling through the full cathedral space.)

4. Marshlight Shift

One continuous 16:9 shot at the edge of a brackish marsh: a piebald heron stands on a crooked driftwood stump, then slowly steps forward and snaps at silver insects as the camera makes a gentle half-orbit from left to right at waist height; halfway through, a dense wedge of storm cloud slides over the late-afternoon sun so the scene transitions smoothly from warm amber glow to cool muted gray, the heron’s feathers losing their rim light while ripples on the black water and seed heads of cordgrass dim naturally in real time; breezy, tense, rain-coming mood, no cuts.

Winner: LTX 2.5 Image to Video Pro — Model B better matches the continuous marsh-edge shot with a gentle orbit, visible stepping behavior, and a convincing warm-to-cool lighting transition across the clip, while maintaining stronger visual richness. Model A has a solid stormy marsh setup and stable heron rendering, but it underdelivers on the specified action and camera movement, feeling more static and less expressive overall. (Second judge pass, order swapped — scores are the average of both: Model B better matches the marsh-edge setup, crooked driftwood perch, and especially the smooth warm-to-cool lighting transition with natural dimming across water and grasses; it also suggests more camera movement consistent with a gentle orbit. Model A is visually pleasing and stable, but it looks more static, the perch and bird read less like the specified piebald heron in a brackish marsh, and the prompt’s stepping/snapping action and evolving light shift are less evident.)

See every prompt and the full side-by-side outputs in the interactive Head-to-Head.