Skip to content
Video

Agnes-Video-H3-MoE

sparse MoE add-on for video generation based on MiniMax-H3

2 min readupdated

This is an experimental checkpoint that turns dense network layers in blocks 15–34 of the MiniMax-H3 FL2VA model into a sparse MoE (Mixture of Experts) architecture at the token level. For each token, the model runs one shared expert and selects one of eight private experts, with different tokens in the same video able to activate different paths. The result is video and audio generation that supports conditional generation using the first, last, or both frames. The repository ships as a composite package with base model weights, a routing map, and trained LoRA weights (a small add-on to the base model).

Key features

  • Sparse MoE architecture: conversion of 20 out of 50 diffusion blocks into a token-level Top-1 structure with shared and private experts.
  • Multimodal output: simultaneous generation of high-quality video footage with audio accompaniment.
  • Flexible control: support for working with the first and last frames as conditions.
  • Reference loader: includes ready-to-use DiffSynth scripts for inference and package verification.

What you can do

  • Generate a high-quality 10-second video clip using the reference Python inference script.
  • Run text-to-video generation by adding starting or ending frames as reference images.
  • Check the integrity of the weight package and routing configuration using an offline validation script.

Example

hf download Agnes-AI/Agnes-Video-H3-MoE --local-dir E8T1-shared1
cd E8T1-shared1
python deploy/check.py
python deploy/infer.py \
  --package . \
  --prompt "A cat walks across a piano." \
  --out sample.mp4 \
  --height 480 --width 832 --frames 124 --steps 50 --seed 42
Read moreDetails
Based on the MiniMax-H3 FL2VA model in BF16 format.
The total repository size is around 135 GB.
Requires a powerful CUDA-enabled graphics card, PyTorch, and significant system memory with CPU offloading.
Training took 60 steps on 128 examples.
Video examples
Discuss Telegram 2

Will it run?

Won't run

Not enough memory even for 4-bit. Below are GPUs you can rent.

  • BF16weights 195 GB · ready files228 GB
  • FP8 / INT8recommendedweights 164 GB192 GB
  • 4-bitweights 151 GB177 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

No single instance holds the whole model: it needs RAM offload or a cluster.

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.

This is a LoRA: counted together with its base model MiniMax-H3.

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 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.

    #!/usr/bin/env bash
    # Agnes-Video-H3-MoE
    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/$TORCH
    uv pip install -r requirements.txt "huggingface_hub[cli]"
    # Base files: MiniMax H3, W6A8 pruned
    if [ "$CU" -ge 13 ]; then
      hf download Comfy-Org/MiniMax-H3 diffusion_models/minimax_h3_fl2va_pruned_w6a8.safetensors 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
    else
      hf download Comfy-Org/MiniMax-H3 diffusion_models/minimax_h3_fl2va_pruned_w6a8.safetensors 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
    # LoRAs (1)
    hf download Agnes-AI/Agnes-Video-H3-MoE FL2VA/video_vae/source/model.safetensors FL2VA/text_encoder/model-00001-of-00014.safetensors FL2VA/text_encoder/model-00002-of-00014.safetensors FL2VA/text_encoder/model-00003-of-00014.safetensors FL2VA/text_encoder/model-00004-of-00014.safetensors FL2VA/text_encoder/model-00005-of-00014.safetensors FL2VA/text_encoder/model-00006-of-00014.safetensors FL2VA/text_encoder/model-00007-of-00014.safetensors --local-dir models/loras/Agnes-AI__Agnes-Video-H3-MoE
    # 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 MiniMax-H3
  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

Requires the base model

MiniMaxAI/MiniMax-H3

This model is a LoRA / adapter. Download the base model first, then load this one on top.

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

Agnes-AI/Agnes-Video-H3-MoE
Author
Agnes-AI
Category
Video
Released
Posted

On NeyroLearn

Hugging Face

02

Ask AI about this model

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

Comments

Video

Wan2.1

Wan2.1