Skip to content
Other

MiniMax-H3 PTD LoRAs

облегчённые ускоряющие LoRA для Ref2VA

2 min read

Это derivative-версии few-step LoRA для MiniMax-H3 PTD Ref2VA, подготовленные Winnougan на основе релиза taobao-content-ai. Оригинальные адаптеры rank 128 урезаны через per-hop SVD до rank 64 или 32, чтобы файлы было легче скачивать и загружать в ComfyUI. Задача та же: reference image → video + audio за несколько шагов.

Ключевые особенности

  • PTD обучает отдельную LoRA на каждый шаг денойзинга: 4-step файл — 4 hops, 8-step — 8 hops.
  • Rank снижен со 128 до 64 или 32; каждый hop и модуль урезаны отдельно, lora_alpha равен новому rank, alpha/rank = 1.0.
  • Формат BF16: это rank pruning, а не квантование. proj_out и audio_proj_out не тронуты.
  • Размеры: 4-step r64 ~2.6 GB, 4-step r32 ~1.3 GB, 8-step r64 5.2 GB, 8-step r32 ~2.6 GB. Оригинальный 8-step rank 128 — 10.35 GB.
  • Для ComfyUI нужен оригинальный необрезанный Ref2VA transformer, ноды из оригинального репозитория и ComfyUI-PTD.
  • Файлы community-made и не одобрены оригинальными авторами; лицензия и условия — как у оригинала.

Пример

Сценарий из карточки: положите .safetensors в ComfyUI/models/loras/, установите ноды из оригинального репозитория и ComfyUI-PTD, загрузите LoRA PTD-лоадером с weight 1.0 и выставьте 4 или 8 шагов — ровно под число hops. База — только оригинальный необрезанный Ref2VA transformer; pruned, Turbo, PDD и quantized базы не подходят.

ПодробнееEnergy retained (mean/min): 4-step r64 — 0.989/0.921, 4-step r32 — 0.975/0.814, 8-step r64 — 0.985/0.865, 8-step r32 — 0.965/0.724.
8-step r32 имеет самый слабый worst-case модуль (min 0.72) — тестируйте первым.
Pruning lossy: при drift в motion, lip sync или identity переходите на r64 или оригинал.
Ноды читают rank и alpha из метаданных; если loader жёстко задаёт rank 128, сила будет неверной.
Discuss Telegram 6 1.4K

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 5.2 GB · ready files6.2 GB
  • FP8 / INT8recommendedweights 2.6 GB3.4 GB
  • 4-bitweights 1.5 GB2.1 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 2060 · BF16 — from RTX 5060 Ti 8GB

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-PTD-LoRAs-Quantized-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-PTD-LoRAs-Quantized-ComfyUI
    hf download Winnougan/MiniMax-H3-PTD-LoRAs-Quantized-ComfyUI MiniMax-H3-Ref2VA-PTD-8Step_r64.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
  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.

  • 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 (2)
    • 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.
  • 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-PTD-LoRAs
  3. MiniMax-H3-PTD-LoRAs-Quantized-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

Winnougan/MiniMax-H3-PTD-LoRAs-Quantized-ComfyUI
Author
Winnougan
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1.4K6

Ask AI about this model

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

Comments