LTX 2.5 is now on WanGP
WanGP now supports LTX2.5, so I tried it out.
- System Requirements: It's WanGP so you can do this with low VRAM. NVIDIA or AMD GPU.
- Generation Speed: 10 sec 480p video takes around 3 minutes to generate on A4500.
Some examples (generated with text-to-video):
Prompt:
A hyper-realistic casual phone recording captured by a bystander walking along the Embarcadero in San Francisco on an ordinary overcast afternoon. The footage must feel exactly like a real smartphone video someone unexpectedly recorded in public, not a cinematic scene. The location should be immediately recognizable as real San Francisco: Embarcadero waterfront, palm trees, piers, bike lane, Muni street activity, pedestrians, light traffic, and a faint view of the Bay Bridge in the distance. The clip begins mid-walk with slightly messy handheld movement, natural footstep motion, and the camera not fully aimed yet. The operator notices a large California sea lion lying awkwardly on the sidewalk and instinctively turns the phone toward it with a slightly delayed reaction. A couple pedestrians nearby are already stopping and looking confused, giving it space. The camera keeps moving forward, a little shaky and uncertain, revealing another sea lion near a bike rack, then another by the curb. Someone nearby says, “Why are there sea lions up here?” in a genuinely confused tone. The operator keeps walking and then quickly pans around the corner, revealing an intersection unexpectedly occupied by dozens of large sea lions spread across the street, some lying on the asphalt, some slowly waddling between stopped cars, some barking. A Muni bus is partially blocked. Police have loosely closed off the intersection and are standing back while pedestrians on the sidewalk film and react with disbelief. The sea lions must look totally real in size, texture, and movement, with natural weight on the pavement and realistic interaction with the environment. The video should have authentic smartphone qualities: vertical or casual horizontal framing, slight hand shake, uneven panning, brief autofocus breathing, mild exposure adjustment, occasional motion blur, imperfect composition, ordinary phone audio with distant traffic, sea-lion barking, gulls, footsteps, wind, and overlapping confused human chatter. No cinematic camera moves, no polished staging, no dramatic soundtrack, no color grading, and no visual effects. The entire clip should feel like a genuine viral bystander video of a bizarre but plausible real incident in downtown San Francisco.
