Head to head: LTX 2.3 Video Fast vs Luma Ray 3.2 Image to Video
LTX 2.3 Video Fast vs Luma Ray 3.2 Image to Video
By Ryan Merket · Published · Updated
This matchup tests two video models on identity preservation, continuous camera movement, temporal consistency, and a changing-light loading-dock scene. The comparison focuses on whether each clip follows the brief—not just whether it looks polished.
LTX 2.3 Video Fast scored **29.3** and Luma Ray 3.2 Image to Video scored **26.1**, but the saved result is a **tie**. Across the 12 task outcomes, LTX took 7 wins to Luma’s 2, with 3 ties; editorially, this is **Too close to call (not statistical)**, not a clear overall win for either model. LTX’s clearest advantage was specificity and continuity: it more often held onto Lenka’s defining details at the watchmaker’s bench and kept the yellow-coated figure and umbrella legible across the rainy-street clips. It also tended to make the cathedral approach read as a steady, candlelit glide. Luma had its strongest moments in the loading dock and one aisle run, where its atmosphere and scene composition were compelling, though the dock results split over whether it better conveyed the lighting sweep or the requested setting and cargo. The scores are not a basis for claiming statistical superiority. Across three successful repeated trials, the mean paired score difference (LTX minus Luma) was 3.266666666666667; the **Student-t 95% confidence interval** for that mean difference was -4.679015204568563 to 11.212348537901896. That interval is reported as an interval, not as a probability of winning or a claim about general capability. **Final call: tie—too close to call on this saved task sample.** LTX was the more frequent task winner, but Luma’s wins and the mixed dock and aisle results keep the overall verdict from being a decisive one.
Watchmaker bench identity lock
A short continuous 16:9 shot inside a tiny repair studio called Mirek & Sons Horology: a watchmaker with a distinctive copper streak in her black braid, round smoke-tinted glasses, a forest-green apron with a stitched name patch reading "Lenka", and a small crescent scar above her left eyebrow sits at a cluttered bench delicately turning a brass pocket-watch movement with tweezers while tiny screws glint on blue paper trays; the camera performs a slow dramatic orbit from front-right to nearly profile, pushing in slightly as she leans closer, swaps tools, and exhales in concentration under a cool overhead task lamp and dim amber room light; her face, hairstyle, glasses, apron, hands, and all clothing details must remain perfectly consistent from first frame to last with no morphing, flicker, or identity drift, creating a calm, exacting mood with no cuts.
OpenAI: GPT-6 Luna: Model A closely matches the specified identity and details, including the copper-accented braid, glasses, Lenka apron patch, scar, watch work, and blue trays, while maintaining a coherent gradual push-in toward profile. Model B has a pleasing workshop mood and continuous action, but omits or obscures key identity and clothing details, especially the forest-green apron and name patch. (Second judge pass, order swapped — scores are the average of both: Model A much more clearly matches the identif | Anthropic: Claude Fable 5.1: Model A nails the specified details (black brai…
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-6 Luna: Model A more clearly conveys a continuous forward glide from the entrance along a candle-lined aisle toward the altar, with coherent spatial progression. Model B is especially atmospheric, but its sparse candles and slight change in viewing direction make the requested approach less direct. (Second judge pass, order swapped — scores are the average of both: Model A more clearly begins at the entrance and glides down the central aisle toward the altar, with strong candlelight and consistent frami | Anthropic: Claude Fable 5.1: Both deliver a smooth single push-in from the d…
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-6 Luna: Model B shows a clearer, natural approach toward the camera with convincing rainy-street atmosphere, while keeping the raincoat and umbrella visually stable. Model A is also consistent and polished, but its motion is less evident and the close framing makes the walk harder to read. (Second judge pass, order swapped — scores are the average of both: Model B shows a clear, continuous approach with visible rain and a cinematic wet street, while keeping the yellow coat and umbrella recognizable thro | Anthropic: Claude Fable 5.1: Model A keeps the man's face, coat and umbrella…
Loading dock shadow sweep
A short continuous 16:9 shot at a specialty tea warehouse loading dock: a forklift operator in a faded saffron safety vest slowly backs a pallet of silver canisters marked "Lot 47B" toward a blue roll-up door while the camera makes a smooth sideways dolly from left to right at waist height, staying close enough to catch the operator glancing over one shoulder; midway through the shot, a low thundercloud slides over the afternoon sun so the bright warm dock light gradually dims into cool muted shade, reflections on the canisters soften, then a strip of amber motion-sensor lights under the awning clicks on one by one, restoring a believable mixed lighting across the concrete; the mood should feel tense but routine, with no cuts.
OpenAI: GPT-6 Luna: Model B more clearly shows the forklift and pallet approaching the blue roll-up door, with cool dock light and warm awning lights creating the requested mixed-lighting feel. Model A has attractive imagery and a noticeable shift in lighting, but the operator and forklift largely disappear from view and the action is less clearly tied to the prompt. (Second judge pass, order swapped — scores are the average of both: Model B better matches the blue roll-up door, silver canisters, and cool, overcast | Anthropic: Claude Fable 5.1: Model A delivers the key beats—legible 'Lot 47B…
Matchup powered by OpenRouter.