🎨 Accumulated on #minimaxH3 — Part 29
The 29th part of a collection of VAE, LoRA, and ComfyUI tools for the open 33B video model MiniMax H3, which generates video and native stereo sound, has been released in the 'Neuronaut' channel.
🌍 In a month, an ecosystem on par with early Stable Diffusion has formed around the open H3 weights: its own VAEs, 4-step turbo-LoRA, quantizations, and wrappers. Lightweight VAEs and the QuantFunc INT4 quantization lower the entry barrier from data center GPUs to consumer cards.
👤 With ComfyUI and 8–12 GB of VRAM, you can already get a working pipeline: a 360° scene flythrough in 4 steps, a clean single frame up to 8 MP without banding, or a 4-bit version of the model. Each tool is a live link with a model, node, and workflow. The only numerical metric in the collection is the 23.7 dB PSNR of the quantization — it is frame-based and says nothing about video smoothness or sound.
Source 1: https://github.com/shootthesound/ComfyUI-Fizgig-H3-Still Source 2: https://huggingface.co/speach1sdef178/MiniMax-H3-X2-Detail-VAE