MiniMax-H3 Singularity · GGUF

Pick the quant that fits your GPU, then set up ComfyUI. Companion tool for Abiray/MiniMax-H3-Singularity-GGUF.

joint audio-video diffusion text-to-video · image-to-video ComfyUI

1What GPU are you running?

Sizing assumes the diffusion model stays resident while ComfyUI swaps the text encoder out, plus ~1.2 GB for the two VAEs, latents and activations with tiled decode.

2All quants

Sizes load live from the Hugging Face API. Click any row to switch the download commands below.

3Download

4Everything else ComfyUI needs

The GGUF above is only the diffusion model. You also need a text encoder and two VAEs, and the ComfyUI-GGUF custom node to load GGUF unets.

5Speeding it up

Pair the model with the minimax_h3_ref2v_turbo_4step_v0.1 LoRA for 4-step inference instead of the full schedule — the single biggest wall-clock win, and it lowers peak VRAM too since fewer steps means fewer cached activations.