Skip to content
Video

HyperFlow H3 Flash

быстрая генерация видео со звуком за 6 шагов

2 min readupdated

Команда Video Rebirth выпустила сжатую версию модели MiniMax-H3, которая одновременно генерирует видео (от 5,2 до 15 секунд, 1344×768 при 24 fps) и синхронизированную стереодорожку 32 кГц. Вместо ~31B параметров оригинального трансформера здесь используется 13,7B или 13,9B, а генерация занимает всего 4–8 шагов без CFG (classifier-free guidance — метод усиления следования промпту). Благодаря квантизации (сжатию весов) в форматы GGUF и INT8 модель запускается на потребительских видеокартах с 24 ГБ VRAM и работает в ComfyUI, diffusers и stable-diffusion.cpp. Подойдёт тем, кто хочет локально создавать ролики со звуком из текста, ключевых кадров или референсов без серверных ферм.

Ключевые особенности

  • Три режима работы: генерация по тексту (t2va), по первому или первому и последнему кадру (fl2va), а также по референсам (ref2va — до 12 изображений, а в Python также видео и аудио).
  • Быстрый сэмплинг за 4, 6 или 8 шагов: ролик на 5,2 секунды за 6 шагов считается за 92 секунды на RTX 5090, 36 секунд на H200 и 8,7 секунды на 4× B200.
  • Доступны форматы GGUF (от Q3_K_M весом 7,2 ГБ до Q8_0 на 15,2 ГБ), INT8 convrot (~16 ГБ) для ComfyUI и Python, а также неквантизированные веса BF16 (~27,4–27,9 ГБ).
  • Готовые шаблоны воркфлоу для ComfyUI (hyperflow_t2v, hyperflow_i2v, hyperflow_flf2v, hyperflow_ref2v) и поддержка Unsloth Studio.

Что можно сделать

  • Генерировать короткие сцены с готовым эмбиентом и звуковыми эффектами по текстовому описанию через шаблоны hyperflow_t2v в ComfyUI или CLI.
  • Анимировать статичные изображения и делать плавные переходы между начальным и конечным кадрами с озвучкой в режиме fl2va.
  • Собирать ролики с сохранением персонажей, объектов и стиля окружения, подавая до 12 референсов в режиме ref2va (в Python-пакете можно подмешивать и видео/аудио-референсы).

Пример

import torch
from diffusers.utils.export_utils import encode_video
from hyperflow_h3_flash import load_pipeline

pipe = load_pipeline("t2va")
out = pipe(prompt="A chef slides a steaming bowl of ramen across a wooden counter; rain taps on the window.", num_frames=124, num_inference_steps=6, output=["videos", "audio", "sampling_rate"])
encode_video(out["videos"][0], fps=24, output_path="ramen.mp4", audio=out["audio"][0], audio_sample_rate=out["sampling_rate"])
ПодробнееРазмер модели: 13,7B (fl2va) и 13,9B (ref2va) против ~31B у базовой MiniMax-H3.
Требования к железу: протестировано на GPU с 24 ГБ и 32 ГБ VRAM при 16 ГБ+ ОЗУ (карты на 16 ГБ ещё не замерялись); без оффлоада в Python нужно ≥ 90 GiB VRAM.
Дополнительно качаются текстовый энкодер Qwen3-VL-32B и VAE из репозиториев MiniMax-H3 / Comfy-Org.
Лицензия MiniMax H3 Community License ограничивает использование на территории ЕС, Великобритании, Южной Кореи и США без отдельного разрешения MiniMax.
Discuss 349 848.6K

Will it run?

Runs in GGUF Q4_K_M

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

  • GGUF Q8_0recommendedweights 28 GB · ready files35 GB
  • GGUF Q6_Kweights 22 GB · ready files28 GB
  • GGUF Q5_K_Mweights 19 GB · ready files25 GB
  • GGUF Q4_K_Mweights 17 GB · ready files23 GB
  • GGUF Q4_K_Sweights 15 GB · ready files20 GB
  • GGUF Q3_K_Mweights 14 GB · ready files19 GB

Tip: go 8-bit.GGUF Q8_0 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: Won't fit any home GPU — renting is cheaper.

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
    # hyperflow-h3-flash
    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: hyperflow-h3-flash
    [ -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 VideoRebirthAI/hyperflow-h3-flash --include '*Q8_0*.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
    Add more LoRAs to hyperflow-h3-flash
  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.
  • 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

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

VideoRebirthAI/hyperflow-h3-flash
Author
VideoRebirthAI
Category
Video
Released
Posted

On NeyroLearn

Hugging Face

848.6K349

Ask AI about this model

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

Comments

Video

HyperFlow

HyperFlow