LTX 2.3 horizontal example (1920x1088)

An independent benchmark of Lightricks' LTX-2.3 text-to-video AI model reveals the substantial hardware requirements for generating high-resolution video content. Using a modified 48GB Chinese RTX 4090 GPU and 128GB of DDR5 RAM, the tester generated a 5-second, 1920x1088 resolution clip in 192 seconds, with the system consuming 46GB of VRAM and 81GB of system RAM during the process. The test used a detailed prompt describing a woman walking through Times Square at night, demonstrating the model's capability to handle complex urban scenes with specific character details and atmospheric elements. However, the generation failed completely when attempting vertical video formats, indicating format limitations in this early version.

The technical specifications highlight the current frontier of consumer-grade AI video generation, where even short clips require workstation-level hardware. The 10-second generation took 337-370 seconds with similar resource usage, showing near-linear scaling. The audio quality was noted as unchanged from previous versions, suggesting audio generation remains a secondary focus. These benchmarks provide crucial data points for creators evaluating whether to invest in high-end hardware for AI video workflows, with the tester humorously suggesting viewers might need to "sell your kidneys" for professional-grade GPUs like the RTX 6000 Ada. The test represents real-world performance data outside controlled corporate demos, showing both the impressive capabilities and current limitations of accessible AI video generation.