This is the latest MiniMax H3 I2V Image-to-Video 1280 resolution accelerated workflow
The results are quite stunning

Model

ConvRot Hadamard rotation quantization technology (not pruning) model: minimax_h3_fl2va_int8_convrot.safetensors

Acceleration Module

Official 4-step acceleration LoRA


10s 1280 HD video output takes about 5 minutes.

This fl2va model supports text-to-video, single-image reference, and first-last-frame video generation.

Note: Not uploading an image means text-to-image, or disable the image box.