Skip to content
Images

MiniMax-H3-Image Full BF16 для ComfyUI

полные веса, но без GPU-проверки

2 min readupdated

AI-KSK выложил community-конвертацию официальных полных весов MiniMax-H3-Image в один safetensors для ComfyUI. Это не новая модель и не файнтюн: веса не квантованы, не обрезаны и не дообучены, сохранены в BF16/F32. Файл предназначен для папки models/diffusion_models/ и рассчитан на структуру ComfyUI 0.45.0. Это именно image-модель, не видеомодель H3. Главное ограничение: автор проверил только CPU-конвертацию, а загрузку в ComfyUI, генерацию на GPU, качество картинок и пиковый VRAM не тестировал.

Ключевые особенности

  • Полные необрезанные веса официальной MiniMax-H3-Image — не pruned-релиз Comfy-Org и не вариант P-H3-Image от PrunaAI.
  • Один файл safetensors: 531 тензор, BF16/F32 без квантования; целевая структура — minimax_h3_image, 50 слоёв, 24 latent channels, полный AdaLN path, PDD 6 steps.
  • QKV de-interleave без потерь: 52 тензора, 56 heads × 128 head_dim; ещё 2 PDD-головы закодированы с FP32-округлением, поэтому файл не побайтовая копия upstream.
  • Не проверены загрузка в ComfyUI, GPU-инференс, text-to-image, image editing, качество и пиковый VRAM; примеров изображений нет.
  • Лицензия MiniMax-H3-Image Community License: территория действия исключает ЕС, Великобританию, Южную Корею и США; коммерция свыше $1M в год или Hosted Service требует отдельного разрешения MiniMax.

Что можно сделать

  • Собрать workflow в ComfyUI: положить файл в models/diffusion_models/ и подключить совместимые text encoder, VAE и prompt-enhancement — в репозитории только diffusion model.
  • Проверить полные веса на своих промптах в text-to-image и image editing, если хватает VRAM или готовы к offloading.
  • Сравнить с pruned Comfy-Org и P-H3-Image (PrunaAI), когда появится GPU-тест.
  • Использовать как основу для экспериментов, если нужен именно full BF16, а не квантованная или обрезанная версия.

Пример

Быстрый сценарий из карточки: разместите файл по пути ниже. Для первых тестов автор упоминает ориентир 6 steps / CFG 1 / Euler / simple — это только стартовая точка, а не подтверждённый результат.

ComfyUI/models/diffusion_models/minimax_h3_image_full_bf16_comfyui.safetensors
ПодробнееРазмер файла: 66 289 412 544 байт (~66.29 GB / 61.74 GiB), 531 тензор. CPU-проверка: 531/531 тензоров без расхождений, максимальная абсолютная ошибка PDD — 1.86264514923e-09.
Веса больше 48 GiB GPU, поэтому ожидайте offloading; пиковый VRAM неизвестен, размер файла не равен потреблению VRAM.
Не тестировались загрузка в ComfyUI, генерация, качество и VRAM. Лицензия ограничивает территорию и коммерческое использование.
Discuss Telegram 3

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 62 GB · ready files69 GB
  • FP8 / INT8recommendedweights 31 GB35 GB
  • 4-bitweights 17 GB21 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.

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-Full-BF16-ComfyUI
    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-Full-BF16-ComfyUI
    hf download AI-KSK/MiniMax-H3-Image-Full-BF16-ComfyUI minimax_h3_image_full_bf16_comfyui.safetensors --local-dir models/diffusion_models
    # 16-bit model: ComfyUI holds it in 8 bits (--fp8_e4m3fn-unet), half the memory
    # 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 --fp8_e4m3fn-unet
  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-Full-BF16-ComfyUI

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

AI-KSK/MiniMax-H3-Image-Full-BF16-ComfyUI
Author
AI-KSK
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

03

Ask AI about this model

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

Comments