Skip to content
Images

1⃣ Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF

тяжёлая генерация картинок в 16 ГБ памяти

2 min readupdated

Автор kabcnd выпустил полный комплект квантованных (сжатых) весов модели генерации изображений MiniMax-H3-Image, адаптированный для локального запуска на устройствах с 16 ГБ объединённой памяти Apple Silicon. В сборку вошли 20-миллиардный диффузионный трансформер (DiT), текстовый энкодер Qwen3-VL-32B (50 слоёв) и оригинальный VAE в формате FP16. С помощью комбинированной схемы сжатия Pin-Grid объём модели удалось уменьшить более чем в четыре раза без потери работоспособности на потребительском железе. Решение пригодится владельцам Mac с 16 ГБ RAM, которые хотят запускать крупные генеративные модели локально без ошибок нехватки памяти (OOM).

Ключевые особенности

  • Сжатие Pin-Grid: основной блок 20B DiT весит 8,40 GiB вместо исходных 37,47 GiB в BF16 (уменьшение в 4,46 раза).
  • Компактный энкодер: 50-слойный Qwen3-VL-32B распакован из int8_convrot и сжат до 10,46 GiB с сохранением визуальных слоёв visual.* в Q8_0 для редактирования картинок.
  • Поддержка популярных движков: работает как через нативный stable-diffusion.cpp (sd-cli и sd-server с Metal), так и в ComfyUI через плагин ComfyUI-GGUF.
  • Открытые рецепты: в репозитории лежат Python-скрипты convert_dit_to_gguf.py и convert_te_to_gguf.py для самостоятельной потоковой конвертации.

Что можно сделать

  • Генерировать изображения 1024×1024 локально через терминал с помощью sd-cli на Mac mini M6 с 16 ГБ памяти.
  • Поднять фоновый локальный сервер sd-server на порту 8189, чтобы не тратить время на повторную загрузку весов при каждом запросе.
  • Подключить веса в ComfyUI через ComfyUI-GGUF для сборки пайплайнов генерации и редактирования изображений.

Пример

./sd-cli --diffusion-model Kenzan-MiniMax-H3-Image-20B-PinGrid-Q3_K-M6-16G.gguf \
  --llm Kenzan-Qwen3VL-32B-L50-MiniMaxH3-PinGrid-Q2_K-M6-16G.gguf \
  --vae minimax_h3_image_vae_fp16.safetensors \
  --offload-to-cpu --flash-attn on \
  -W 1024 -H 1024 --steps 6 --cfg-scale 1.0 \
  --sampling-method euler --scheduler simple -p "Your prompt here" -o output.png
ПодробнееТесты на Apple Mac mini M6 (16 GB Unified Memory, 1024×1024, 6 шагов euler/simple):
— stable-diffusion.cpp (Metal 4): 7,20–8,24 с/шаг, декодирование VAE 2,43–2,59 с (с флагом --offload-to-cpu).
— ComfyUI + ComfyUI-GGUF (PyTorch MPS): 13,27 с/шаг, декодирование VAE 5,36 с, пиковое потребление памяти 11,88 GiB.
Состав файлов: DiT 8,40 GiB (смесь Q2_K/Q3_K/Q4_0/Q5_K и F32), энкодер 10,46 GiB (Q2_K/Q3_K/Q4_0/Q5_K/Q8_0), VAE 1,09 GiB (FP16).
Discuss Telegram 0 173

Will it run?

Runs in GGUF Q2_K

The most precise of the ready quantised files fits.

  • GGUF Q2_Kweights 10 GB · ready files13 GB
  • GGUF Q3_Kweights 8.4 GB · ready files11 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q2_K — 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
    # Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF
    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: Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF
    [ -d custom_nodes/ComfyUI-GGUF ] || git clone https://github.com/city96/ComfyUI-GGUF custom_nodes/ComfyUI-GGUF
    uv pip install -r custom_nodes/ComfyUI-GGUF/requirements.txt
    hf download kabcnd/Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF --include '*Q2_K*.gguf' --local-dir models/unet
    # 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
  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
  4. Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF

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

kabcnd/Kenzan-MiniMax-H3-Image-20B-PinGrid-GGUF
Author
kabcnd
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

1730

Ask AI about this model

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

Comments