I recently stumbled upon an interesting combination of WAN 2.2 and Light 2.2 V2I Loras that produces videos with impressive prompt adherence without sacrificing movement. The secret lies in combining the new Light 2.2 V2I Loras with the 2.1 V2I Loras. I experimented with different strengths, adding the 2.2 Lora at 1 and the 2.1 Lora at 3 strength for high noise, and the 2.2 Lora at 1 and the 2.1 Lora at 0.25 strength for low noise. The results were astonishing! Not only did I achieve great prompt adherence, but the videos also generated quickly thanks to kijai’s sampler using the flowmatch_distill scheduler with 4 steps. If you’re struggling with video generation, I highly recommend giving this combination a try. You might be surprised at the results you get.
Note: The original title had a typo, and I meant to say I2V models, not V2I. Thank you for reading, and I hope this helps!