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: <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main">https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main</a>在线体验 / Try it online: <a target="_blank" rel="ugc" href="https://www.runninghub.ai/post/2086870255426711553/?inviteCode=rh-v1067">https://www.runninghub.ai/post/2086870255426711553/?inviteCode=rh-v1067</a>V1A 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: <a target="_blank" rel="ugc" href="https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora">https://huggingface.co/larryvrh/MiniMax-H3-Turbo-Lora</a>

MiniMax H3 4-Step Accelerated First-and-Last-Frame Video | minimaxh3-4步加速首尾帧视频

V2 更新 / V2 Update中文：本版本新增两组一键切换：① LightX2V Turbo 4步/8步 LoRA 与对应采样步数同步切换；② 循环视频/普通首尾帧视频切换。循环模式将第一张图同时作为首尾帧；普通首尾帧模式使用两张不同图。采样与解码改用 comfyui-minimax-h3-audio-T8 的 MiniMaxH3DualClockSamplerT8 与 MiniMaxH3AVDecodeT8，音频表现更好。RTX 5090 单次实测：5秒、约560p档约47秒；实际速度因分辨率与环境而异。请使用 CUDA 13.0。English: Adds synchronized LightX2V Turbo 4-step/8-step LoRA and sampling-step switching, plus Loop Video / First-and-Last-Frame Video modes. Loop mode reuses the first image as both endpoints; first/last-frame mode uses two separate images. The workflow now uses MiniMaxH3DualClockSamplerT8 and MiniMaxH3AVDecodeT8 from comfyui-minimax-h3-audio-T8 for improved native audio output. RTX 5090 single-run benchmark: about 47 seconds for a 5-second, ~560p-class video; performance varies. Use CUDA 13.0.LightX2V LoRA:<a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main">https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main</a>在线体验 / Try it online:<a target="_blank" rel="ugc" href="https://www.runninghub.ai/post/2086870255426711553/?inviteCode=rh-v1067">https://www.runninghub.ai/post/2086870255426711553/?inviteCode=rh-v1067</a>

Files with SHA256: 672a3d89b9aeb8044b4a6be3b40a1892025ff98385c91b3ce6fa38a591c9dcd7