LTX-2.5 Open-Sources Native Multishot Video Generation for Character Consistency

JafarNajafov · x · 2026-08-12

LTX-2.5 introduces Native multishot generation, handling multiple shots as a continuous sequence to maintain character, setting, and voice consistency across cuts. This solves the consistency issue found in traditional AI video workflows that treat every shot as a separate generation.

The model also upgrades prompt understanding using a custom Gemma 4 12B backbone and a dedicated E2B prompt enhancer. A new duration head analyzes prompts before diffusion to predict and allocate appropriate video lengths for complex scenes.

Being open source, builders can adapt it into pipelines for short films, product videos, music videos, and advertising concepts.

Related event: Lightricks Releases Open-Source Video Model LTX-2.5(16 posts)→

Original post →

More from Multimodal

Multimodal channel →