MachineDelusions

/

LTX-2_Image2Video_Adapter_LoRa

like 40

Thank you for releasing this high-rank (Rank 256) adapter for LTX-Video 2! It looks very promising for I2V workflows.

I have a question regarding its versatility: While this LoRA is specifically designed to bridge the gap for Image-to-Video (I2V), how does it perform in pure Text-to-Video (T2V) scenarios (i.e., without a reference image input)?

Specifically:

Does it offer any improvements in motion coherence or visual stability for T2V generations?

Or would its high-rank adaptation potentially interfere with the model’s ability to follow text prompts when no image embedding is provided?

I’m currently building workflows in ComfyUI and would love to know if I should keep this LoRA active as a "general enhancer" or strictly reserve it for I2V tasks. Thanks in advance!