Initial release of Kinetic Sakuga for MiniMax H3.
A style LoRA trained on action animation sequences, focusing on dynamic poses, fast movement, close-range combat, motion trails, and impact effects.
Trigger word: kinetic_sakuga2
Recommended starting strength: 1.0
For initial testing, use 20 sampling steps with Turbo disabled. Describe the action clearly after the trigger word. Results may vary depending on the scene, resolution, and base model.
ComfyUI example workflows for T2V, I2V, and R2V are included. Upload your own input images before using the I2V or R2V workflows.
Training: 3,000 steps, rank 32, alpha 32. Text encoder training was disabled.
