MiniMax H3 Audio Fix and 4 step Turbo LoRA for Low VRAM GPUs
The last 48 hours have been wild with ComfyUI Nightly updates, leaving me in a vicious loop of constant updating and testing with MiniMax H3. The good news? Some major issues are finally fixed, including that distorted audio bug!
Even better news: thanks to the 4-step Turbo LoRA, we can now run relatively fast iterations even on low VRAM GPUs, like the RTX 3060 12GB. Also, for anyone who isn't just focused on human-centric scenes, MiniMax H3 is absolutely fantastic for niche areas like technical renders and architecture. If you're into this stuff, hit me up in the comments and share your work!
The last two days have been super intense. Because of the latest ComfyUI Nightly updates, I had to recreate several scenes for my YT channel since the audio was incredibly important. If you're still experimenting with MiniMax H3 and hitting a wall, just drop your questions below. I’ll either answer them myself, or someone from the ComfyUI team will jump in. Stay tuned, I'm just finishing up some new showcases for YT and will share a few tips soon!
Quick reminder: don't forget to update ComfyUI to the absolute latest version. Even better, grab a fresh portable version of ComfyUI just for MiniMax H3 experiments. Pro tip: use Turbo LoRAs, the updated experimental VAE by kijai (great work, btw), and definitely check out my WFs and switch your KSampler to Euler. That should fix any distortion and give you flawless audio.
Cheers!
MiniMax H3 Extravaganza - ComfyUI PlayGround RTX 3060 12GB
minimax_h3_turbo_4step_ckpt850.safetensors at 5 Steps 960x544
https://www.youtube.com/@ComfyUIPlayground
Cheers!