Skip to content
Video

MiniMax-H3 Fused Turbo

4-step video generation with audio

3 min read

A single 21 GB file for ComfyUI by MATLOWAI that bundles the MiniMax-H3 video model with two baked-in LoRAs (lightweight add-ons on top of the base model): lightx2v Turbo for speed and Mystic v2.0 for smoother motion. Under the hood is a merged fl2va transformer with the ref2va weight delta fused in, letting one file handle text-to-video, first/last-frame interpolation, and reference-guided video with synchronized audio—without swapping checkpoints. Baking the adapters in before INT8 (ConvRot) quantization keeps ComfyUI from storing backup copies of patched layers, saving roughly 21 GB of VRAM compared to loading separate files. Ideal for generating 10-second 1152x640 clips quickly, even on 16 GB GPUs.

Key features

  • 4-step generation: despite using a baked-in 8-step Turbo LoRA, the model hits its sweet spot at 4 steps (or 4+4 in the two-pass de-rope pipeline).
  • Saves ~21 GB of VRAM: merging the LoRAs prior to INT8 quantization eliminates the need to hold duplicates of 200 modified layers in memory at runtime in ComfyUI.
  • Crisp audio with SLA: pairing H3 SLA sparse attention with the res_multistep sampler preserves high-frequency audio detail while speeding up rendering by about 40%.
  • Ready-to-run workflows for different setups: the repo includes 5 ComfyUI workflows, covering AMD GPUs (ROCm) and a Low VRAM mode with FFN chunking for 16 GB cards.

What you can do

  • Generate 10-second 1152x640 videos with audio from a text prompt in a single pass (01_reference_4step_sla workflow).
  • Animate still images and create smooth transitions between a start and end frame using the 04_i2v_fl2v_4step_sla graph.
  • Keep a character's or object's appearance consistent in motion from a reference image via the MiniMaxH3ReferenceToVideo node (05_ref2va_4step_sla workflow).
  • Smooth out abrupt motion jumps with the two-pass 02_derope_4plus4_sla pipeline, or boost dynamics using the external warm375 adapter.

Example

Drop the main model file into ComfyUI/models/diffusion_models/, load it with the standard UNETLoader node (weight_dtype: default), and drag a workflow from the workflows/ folder onto the canvas. Recommended sampler settings for a 10-second clip (243 frames, 1152x640): res_multistep sampler, simple scheduler, 4 steps, BasicGuider (no CFG), and MiniMaxH3SigmaShift (video 12 / audio 3) paired with the H3 SLA sparse attention node (sparsity 0.90, block 64).

Read moreDetails
Speed (RTX PRO 6000, 1152x640, 243 frames, SLA): 4 steps — 76 s, 6 steps — 80 s, 8 steps — 103 s, 25 steps — 292 s.
Hardware requirements: in Low VRAM mode on a 16 GB GPU (offloading to 32 GB system RAM), 4-step generation takes 59 s with peak usage of 10.9 GiB VRAM (the standard workflow uses 11.0 GiB and takes 65 s).
Required companion files: you will need the qwen3vl_32b_minimax_h3_nvfp4_awq encoder, minimax_h3_video_vae_int8_convrot video VAE, minimax_h3_audio_vae_fp32 audio VAE, plus the ComfyUI-MAINodes and SLA node packs.
Limitations: the baked-in LoRAs cannot be toggled off or reweighted; under heavy motion, the warm375 adapter can push 2D animation toward photorealism and cause object duplication.
License: MiniMax H3 Community License Agreement (with territorial restrictions); the baked-in Turbo LoRA is licensed under Apache-2.0.
Video examples
Discuss 311 383.2K

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.

  • INT8recommendedweights 20 GB · ready files25 GB
  • 4-bitweights 11 GB16 GB

Tip: go 8-bit.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: INT8 — from RTX 5090

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-fused-turbo-int8-convrot
    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-fused-turbo-int8-convrot
    hf download MATLOWAI/minimax-h3-fused-turbo-int8-convrot diffusion_models/minimax_h3_fused_refdelta_r1024_turbo8_mystic07_int8_convrot.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.

  • 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

  1. MiniMax-H3family root
  2. MiniMax-H3-Pruned-Ref-Delta-Fused-r1024
  3. MiniMax-H3-Pruned-Ref-Delta-Fused-r1024-ComfyUI
  4. minimax-h3-fused-turbo-int8-convrot

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

MATLOWAI/minimax-h3-fused-turbo-int8-convrot
Author
MATLOWAI
Category
Video
Released
Posted

On NeyroLearn

Hugging Face

383.2K311

Ask AI about this model

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

Comments