V2 Update / V2 更新

English: V2 adds synchronized LightX2V Turbo 4-step / 8-step switching and two independently selectable modes: Loop Video or First-and-Last-Frame Video. It now uses MiniMaxH3DualClockSamplerT8 and MiniMaxH3AVDecodeT8 from comfyui-minimax-h3-audio-T8 for improved native audio. RTX 5090 benchmark: about 47 seconds for a 5-second, ~560p-class video. Use CUDA 13.0. The four sample videos demonstrate the 8-step loop mode.

中文：V2 新增 LightX2V Turbo 4步/8步同步切换，并可独立切换循环视频或普通首尾帧视频。采样与解码改用 comfyui-minimax-h3-audio-T8 的 MiniMaxH3DualClockSamplerT8 与 MiniMaxH3AVDecodeT8，原生音频效果更好。RTX 5090 实测：5秒、约560p档约47秒；请使用 CUDA 13.0。四段样片均演示 8步+循环模式。

LightX2V Turbo LoRA: https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main

在线体验 / Try it online: https://www.runninghub.ai/post/2086870255426711553/?inviteCode=rh-v1067

V1

A MiniMax H3 first/last-frame image-to-video workflow accelerated to 4 sampling steps with the MiniMax H3 Turbo v4 LoRA and Turbo Sampler. It generates 24fps video with native stereo audio. The default 5-second duration is aligned to H3’s valid frame grid, producing 124 frames / about 5.17 seconds.

MiniMax H3 首尾帧图生视频 4步加速工作流，使用 Turbo v4 LoRA 与 Turbo Sampler，将采样降至4步。支持分别输入首帧和尾帧，生成24fps并带原生立体声音频的视频。默认5秒，自动对齐为124帧、约5.17秒。

Original 4-step LoRA and nodes / 原4步加速 LoRA 与配套节点: https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora