Skip to content
Images

MiniMax-H3-Image-NVFP4

fast image generation up to 16.8 MP in ComfyUI

2 min readupdated

This is a 4-bit quantized version of MiniMaxAI/MiniMax-H3-Image built specifically for ComfyUI. It handles both text-to-image generation and reference-based image editing at high resolutions (up to 4096×4096) in just 6 steps. Based on Comfy-Org's 40.2 GB pruned bf16 checkpoint, 200 linear layers were converted to NVFP4 (4-bit quantization for the NVIDIA Blackwell architecture) without fine-tuning. This drops the file size to 12.5 GB, letting the model and its text encoder fit comfortably inside the RTX 5090's 32 GB of VRAM while running 2.1–2.4× faster than the original.

Key features

  • Compact footprint: reduced from 40.23 GB to 12.54 GB by converting 200 layers to NVFP4 via ComfyUI's built-in quantizer.
  • High speed: on an RTX 5090, generating a 4 MP frame (2048×2048 or 2720×1536) takes ~3 seconds, while 16.8 MP (4096×4096) takes ~16 seconds.
  • Native integration: loads directly with standard UNETLoader and CLIPLoader nodes in ComfyUI—no custom nodes required.
  • Reference support: works for both text-to-image generation and reference-guided image editing.

What you can do

  • Generate detailed images and posters with legible text at 4K and up to 16.8 megapixels without long wait times.
  • Edit scenes using reference images (such as relighting a room from day to night in ~5.7 seconds).
  • Run the heavy MiniMax H3-Image model locally on a single 32 GB VRAM GPU instead of 96 GB server accelerators.

Example

Place minimax_h3_image_pruned_nvfp4.safetensors into ComfyUI/models/diffusion_models/ and load it using UNETLoader (weight_dtype = default). Use the following KSampler settings for generation:

steps: 6
cfg: 1.0
sampler: euler
scheduler: simple
denonise: 1.0
positive: MiniMaxH3ImageConditioning (output 0)
negative: ConditioningZeroOut of the positive
Read moreDetails
RTX 5090 benchmarks (6 steps): 2.9–3.4 s for 4 MP, 7.2 s for 3840×2176, 15.7–16.6 s for 16.8 MP; cold start takes 13 s.
Requirements: NVIDIA Blackwell GPU (RTX 5090, peak VRAM 27.6–31.1 GB), ComfyUI 0.45+, and PyTorch 2.10 with CUDA 13 (cu130).
Separate files needed: a text encoder (such as the 15.7 GB qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors) and VAE (minimax_h3_image_vae_fp16.safetensors).
Limitations: using 3+ reference images causes noticeable composition drift compared to the bf16 original; the 2-step turbo LoRA was deemed too raw by the author; the license is not valid in the US, EU, UK, and South Korea.
Examples
Discuss Telegram 0 14

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 12 GB · ready files14 GB
  • FP8 / INT8recommendedweights 5.8 GB7.9 GB
  • 4-bitweights 3.3 GB5.1 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 5060 Ti 8GB · BF16 — from RTX 5080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent RTX A6000

    $0.43 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    #!/usr/bin/env bash
    # MiniMax-H3-Image-NVFP4
    set -e
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    # CUDA 13+ → int8_convrot (Comfy-Org's advice), else fp8_scaled
    CU=$(nvidia-smi | grep -oP 'CUDA Version: \K[0-9]+' || echo 12)
    [ "$CU" -ge 13 ] && TORCH=cu130 || TORCH=cu128
    [ -d ~/ComfyUI ] || git clone https://github.com/comfyanonymous/ComfyUI ~/ComfyUI
    cd ~/ComfyUI
    uv venv --python 3.12 --allow-existing && source .venv/bin/activate
    uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/$TORCH
    uv pip install -r requirements.txt "huggingface_hub[cli]"
    # Base files: MiniMax H3, 8-bit
    if [ "$CU" -ge 13 ]; then
      hf download Comfy-Org/MiniMax-H3 text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensors vae/minimax_h3_video_vae_fp16.safetensors vae/minimax_h3_audio_vae_fp32.safetensors --local-dir models
    else
      hf download Comfy-Org/MiniMax-H3 text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors vae/minimax_h3_video_vae_fp16.safetensors vae/minimax_h3_audio_vae_fp32.safetensors --local-dir models
    fi
    # Main model: MiniMax-H3-Image-NVFP4
    hf download AIModelMaestro/MiniMax-H3-Image-NVFP4 minimax_h3_image_pruned_nvfp4.safetensors --local-dir models/diffusion_models
    # Start ComfyUI. From your computer: ssh -L 8188:localhost:8188 user@IP, then http://localhost:8188
    python main.py --listen 127.0.0.1 --port 8188
    Add more LoRAs to MiniMax-H3-Image-NVFP4
  3. 3. Use it

    From your computer: ssh -L 8188:localhost:8188 user@machine-IP, then open http://localhost:8188. Or open that address in the browser on the machine's desktop. The model's workflow is usually in its Hugging Face card.

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

MiniMax H3 prompts

All prompts

Ready prompts with the resulting video: copy and try.

Model family

  1. MiniMax-H3family root
  2. MiniMax-H3-Image
  3. MiniMax-H3-Image-NVFP4

New posts about versions of MiniMax-H3 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 3All on Hugging Face

About the model

AIModelMaestro/MiniMax-H3-Image-NVFP4
Author
AIModelMaestro
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

140

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments