LTX 2.3 Rack Focus Test | ComfyUI Built-in Template [Prompt Included]

A detailed performance test of the new LTX 2.3 video generation model reveals significant improvements in handling complex cinematic techniques. Using a ComfyUI template, the tester generated a 7-second, 1920x1088 pixel clip featuring a challenging 'rack focus' transition from a detailed white-and-gold android to a distant desert landscape. The test, run on a high-end rig with an NVIDIA RTX 4090 GPU, showed a cold-start generation time of 413 seconds, which dropped by roughly 30% to 289 seconds on a warm start once model weights were cached in VRAM, highlighting the model's substantial computational demands for high-resolution, temporally consistent video.

The key finding was LTX 2.3's ability to maintain structural integrity during the depth-of-field shift. The prompt specifically tested whether fine mechanical details on the android's ear and neck would degrade or 'melt' into a blur—a common failure in previous AI video models. The results were promising: the rack focus effect was fluid, and the details held their solid form even as they entered the out-of-focus bokeh region. The tester noted an absence of the 'temporal shimmering' or 'pixel soup' artifacts that often plague AI-generated video transitions, marking a step forward in the model's coherence and stability for professional-grade visual effects.