简单易懂的Minimax H3 工作流/Simple Workflow for Minimax H3 for RTX3060 12G
1.1版本增加了Director工作台，更换了turbo lora已解决兼容问题。同时附带了文生视频/图生视频LLM辅助提示词，帮助您将一个想法通过LLM转化为H3提示词。
Version 1.1 introduces the Director workspace and replaces Turbo LoRa to resolve compatibility issues. It also includes LLM prompt templates for text-to-video and image-to-video generation, helping you convert an idea into an H3 prompt using an LLM.
需要以下
- Director github：[Director](https://github.com/AIMixer/ComfyUI_MiniMaxH3_Director)
- Turbo Lora：[Kijai/MiniMax-H3_comfy](https://huggingface.co/Kijai/MiniMax-H3_comfy/blob/main/loras/minimax_h3_fl2v_lightx2v_turbo_4step_v0.1_comfy.safetensors)
- kijai/ComfyUI-KJNodes：[KJNodes](https://github.com/kijai/ComfyUI-KJNodes)
Description
更换为Easy节点，更换了效果更好的turbo lora，LLM提示词也进行了优化。
We switched to the Easy node, adopted the more effective Turbo LoRa, and optimized the LLM prompts.
FAQ
Comments (5)
Very distorted results in 540p.
Is it actually not work properly with 12Gb
All of my examples were generated at 540p.
The workflow doesn't seem to work right for me. My CPU isn't the newest, but i have a 4070, and my generation times are nothing like the generation times named in the workflow. The workflow says it takes like 2 minutes on the lowest resolution. For me it take up to an hour for the first run, and then the next run says "Your GPU ran out of Memory, try a smaller size" or something like that, and then the next run takes a good 10 minutes. And that's at 3 seconds length.
Why does it take so long and thinks i don't have enough memory?
The generation time for Wan and other stuff like that always seemed accurate to me. This looks like the 3060 is 20 times faster than my 4070...
I guess you have an RTX 4070 8GB. If that's the case, it's actually the limitation of the VRAM rather than the GPU performance. This workflow requires at least an RTX 3060 12GB, and the key point is the 12GB of VRAM.