LTX 2.5 shows reduced audio-reactivity compared to 2.3

ART-ficial-Ignorance · reddit · 2026-08-25

The author found LTX 2.5 performs worse than 2.3 for audio-reactive video generation workflows.

Key Differences

LTX 2.5: Tends to hold the first frame static until an obvious beat arrives, making the opening of shots feel dead.
LTX 2.3: With the Audio-Reactive LoRA, produces subtle motion from the very beginning (fog shifts, surfaces breathe), making the shot feel alive.

Workflow Details

Hardware: RTX 4070.
First/Last Frame Generation: The song is cut into scenes where the start frame of Scene 2 becomes the last frame target for Scene 1. This creates a chain of transformations.
Visuals: Morphing and transitions happen inside the model rather than via editing effects.

Conclusion

Despite 2.5 being faster, the author prefers LTX 2.3 + Audio-Reactive LoRA for music videos due to better early-shot responsiveness.

More from Multimodal