Skip to content
Video

MiniMax H3 INT8/INT6/INT4

квантованные веса для видео в ComfyUI

2 min read

Это набор квантованных чекпоинтов MiniMax H3 от tsolful с Civitai: веса видео-модели в вариантах FL2VA и REF2VA, упакованные для ComfyUI. Доступны четыре уровня точности, чтобы выбрать баланс между скоростью, качеством и VRAM/RAM.

Ключевые особенности

  • Четыре точности: BF16 (~40 ГБ), INT8 W8A8 (~21 ГБ), INT6 W6A8 (~16 ГБ), INT4 W4A8 mixed (~12,5 ГБ).
  • INT8 — рекомендуемый старт: почти без потери качества, работает на INT8 tensor cores.
  • INT6 — компромисс: заметно меньше INT8 при небольшой потере качества.
  • INT4 — максимум экономии: ConvRot codebook декодирует веса в INT8 на вычислениях, поэтому выигрыш именно в памяти, а не в скорости.
  • Чувствительные слои в BF16: I/O, modulation, embeddings и output head не квантуются.
  • Два варианта: FL2VA — видео/аудио из первого и последнего кадра; REF2VA — до 9 изображений, 3 видео со своими дорожками и 3 отдельных аудиоклипа. Оба поддерживают t2v, i2v, v2v, a2v и смешанные референсы.

Пример

В карточке нет готового кода запуска. Автор делится настройками сэмплера: er_sde / res_multistep simple. Для первого запуска советуют взять INT8; если памяти впритык — INT6; если упираетесь в VRAM — INT4.

ПодробнееРазмеры: BF16 ~40 ГБ, INT8 ~21 ГБ, INT6 ~16 ГБ, INT4 ~12,5 ГБ — все варианты pruned.
Нужен ComfyUI с нативной поддержкой MiniMax H3; для INT4 — v0.31.0 или новее.
Рекомендуется NVIDIA GPU с INT8: RTX 30-й серии и новее.
У INT4 потеря качества больше, чем у INT6; INT8 — ближе всего к BF16.
Видео-примеры
Discuss 2.4K 53.5K

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # civitai/2830065-minimax-h3-int8-int6-int4-convrot
    # ===== Tokens: paste yours between the quotes =====
    export CIVITAI_TOKEN=""   # civitai.com/user/account → API Keys, for Civitai LoRAs
    # ==========================================
    mkdir -p ~/models
    vid=$(curl -fsSL "https://civitai.com/api/v1/models/2830065" | python3 -c 'import json,sys;print(json.load(sys.stdin)["modelVersions"][0]["id"])')
    cd ~/models && curl -fL -OJ -H "Authorization: Bearer $CIVITAI_TOKEN" "https://civitai.com/api/download/models/$vid"
    # 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.
  • LightX2V step distill0.6.1from November 9, 2026new release

    5–10× fewer steps

    Distilled LoRAs and an engine: video in 4–8 steps instead of 30–50.

    How to enable: Load their step-distill LoRA, CFG 1.0, 4–8 steps.

  • 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

No known relatives of this model yet.

About the model

civitai/2830065-minimax-h3-int8-int6-int4-convrot
Author
tsolful
Category
Video
Released
Posted

On NeyroLearn

Civitai

53.5K2.4K

Ask AI about this model

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

Comments