Skip to content
Images

MiniMax-H3-Image pDMD 2-Step

готовый кадр за две оценки денойзинга

2 min readupdated

Это LoRA для официального 33,1B-трансформера MiniMax-H3-Image. Она делает реальным 2-step режим нативного разрешения 4096×4096 (16 МП): без CFG, промпты как есть. LoRA поставляется одним файлом lora_model_0.safetensors в bf16 с нативными ключами MiniMax, поэтому сливается напрямую, без ремаппинга.

Ключевые особенности

  • Две оценки денойзинга вместо фиксированных 6 шагов PDD; flow shift 12, без CFG и negative prompt.
  • Rank 128 по attention и FFN всех 50 блоков, LoRA слита в официальный трансформер.
  • Один файл bf16 с нативными ключами MiniMax — без переименования ключей.
  • Нативное разрешение 4096×4096; на H200 около 17,7 с на изображение, пик 84 ГБ.
  • Для ComfyUI есть конверсия в соседнем репозитории: simple BasicScheduler, 2 шага, flow shift 12.

Пример

scheduler.set_shift(12.0)
scheduler.set_timesteps(3)
# передать sigmas=scheduler.sigmas, без CFG

Сценарий: слить LoRA в официальный трансформер и сделать два вызова денойзера — получается кадр 4096×4096. В ComfyUI используется та же рабочая точка: 2 шага и flow shift 12.

ПодробнееНа H200: пик 84 ГБ, ~17,7 с на кадр 4096×4096, 24 изображения примерно за 13 мин.
Штатный MiniMax-H3-Image на 2 шагах остаётся мягким и недоделанным; с LoRA композиция и текстуры доходят за те же две оценки.
Лицензия LoRA — Apache-2.0, но результаты регулируются MiniMax-H3-Image Community License.
Если 2 шага дрейфуют на вашем промпте, авторы советуют 3–4 шага.
Примеры
Discuss Telegram 66 42K

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

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    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.

    # akhaliq/minimax-h3-image-pdmd-2step
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download akhaliq/minimax-h3-image-pdmd-2step --local-dir ~/models/minimax-h3-image-pdmd-2step
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

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-pdmd-2step

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

akhaliq/minimax-h3-image-pdmd-2step
Author
akhaliq
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

42K66

Ask AI about this model

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

Comments