ComfyUI model file · VAE (video)

minimax_h3_video_vae_fp16.safetensors

MiniMax H3 video VAE (fp16) · Video VAE for MiniMax H3

This is MiniMax H3’s video VAE in fp16. It encodes input frames into H3’s 24-channel latent space and decodes the sampler’s output back to video. Put it in ComfyUI/models/vae and load it with Load VAE.

  • MiniMax H3
  • Video decoder
  • 5.21 GB
  • fp16

Short on VRAM or disk? minimax_h3_video_vae_int8_convrot.safetensors is the same VAE at about half the size.

minimax_h3_video_vae_fp16.safetensors

Type
VAE (video)
Format
Safetensors · fp16
Size
5.21 GB (5,207,808,496 bytes)
SHA256
7c1f1314…cdce5e522
ComfyUI folder
models/vae/
Loader node
Load VAE
SHA256 and size from Hugging Face file metadata (Comfy-Org/MiniMax-H3), checked by RunComfy on .

The short version

What is minimax_h3_video_vae_fp16.safetensors?

MiniMax H3 samples video in a compressed latent space. The video VAE is the bridge in and out of it: it encodes start frames, last frames and reference images for FL2VA and Ref2VA, and decodes the finished latent into frames.

The VAE itself is large, about 5 billion parameters, which is why the fp16 file is 5.21 GB. Comfy-Org also publishes an int8_convrot version that is about half the size. Searches for minimax/minimax_h3_video_vae_fp16.safetensors come from workflows that keep it in a minimax subfolder; the file is the same.

Install

Download and install minimax_h3_video_vae_fp16.safetensors

Where it goes
ComfyUI/
└─ models/
   ├─ …
   └─ vae/
      └─ minimax_h3_video_vae_fp16.safetensors
terminal · Linux / macOS
$ cd ComfyUI/models/vae$ wget -O minimax_h3_video_vae_fp16.safetensors \$   https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/vae/minimax_h3_video_vae_fp16.safetensors$ sha256sum minimax_h3_video_vae_fp16.safetensors# 7c1f131492e7eddacaac9069a61b81bdd39de5cc96561e677c5eab1cdce5e522

Safetensors files hold tensors only and cannot run code. Still compare the SHA256: a failed download often leaves an HTML error page saved under the model’s filename.

Use

How to load minimax_h3_video_vae_fp16.safetensors in ComfyUI

Load VAE reads the file. Connect it to the H3 image-to-video encode step (for start frames and references) and to the final decode.

Minimal graph. Click a node to open its reference page.
  • Out of memory at decode

    Use a tiled VAE decode, or switch to the int8_convrot video VAE.

  • Add the audio VAE

    H3 outputs sound too. Load minimax_h3_audio_vae_fp32 in a second Load VAE.

Works with

Files a MiniMax H3 video workflow needs

minimax_h3_video_vae_fp16.safetensors is one part of the stack. A workflow that reports a missing model usually needs these too. Use the Ref2VA diffusion model instead of FL2VA for reference-to-video workflows.

FileRoleFolderSizeDownload
minimax_h3_fl2va_pruned_int8_convrot.safetensorsDiffusion model (FL2VA)models/diffusion_models/20.97 GB Get
qwen3vl_32b_minimax_h3_nvfp4_awq.safetensorsText encoder (Qwen3-VL 32B)models/text_encoders/15.69 GB Get
minimax_h3_audio_vae_fp32.safetensorsAudio VAEmodels/vae/605 MB Get
minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensorsTurbo LoRA, 4 steps(optional)models/loras/1.96 GB Get

Run it

ComfyUI workflows that can use minimax_h3_video_vae_fp16.safetensors

We match workflows by the model files they load: this exact file first, then other versions of it, then workflows built for the same use, then other models from the same family.

  1. Uses this exact file35
  2. Uses another version of this file1

Uses this exact file

35 workflows

These workflows already load minimax_h3_video_vae_fp16.safetensors. Open one and run it as is.

Uses another version of this file

1 workflow

These workflows load a different precision or format of the same model. Select minimax_h3_video_vae_fp16.safetensors in their Load VAE node to use this version.

Compare

Same file, other names, and alternatives

FileRelationSizeBest for
minimax_h3_video_vae_fp16.safetensorsThis page · fp165.21 GBBest decode quality
minimax_h3_video_vae_int8_convrot.safetensorsSame VAE · int8 convrot2.81 GBLower VRAM; needs PyTorch with CUDA 13
minimax_h3_audio_vae_fp32.safetensorsCompanion · audio VAE605 MBDecoding the soundtrack

Troubleshoot

Common errors and fixes

Value not in list: vae_name: 'minimax_h3_video_vae_fp16.safetensors' not in [...]

Why: The workflow asks for this exact filename, but ComfyUI cannot find it in models/vae. Usually the file is missing, sits in a different folder, or was saved under another name such as minimax/minimax_h3_video_vae_fp16.safetensors.

Fix: Move the file into ComfyUI/models/vae/, rename it to match the workflow (or pick it again in the dropdown), then refresh the node list or restart ComfyUI.

SafetensorError: Error while deserializing header

Why: The download is incomplete, or the browser saved an HTML sign-in or error page under the model’s filename.

Fix: Check that the file is exactly 5,207,808,496 bytes and that its SHA256 matches the value above. Download it again from Hugging Face · Comfy-Org/MiniMax-H3 if not.

Rather skip the setup?

Run MiniMax H3 Multi-Character Swap in your browser on a cloud GPU. Nothing to download.

Run MiniMax H3 online

FAQ

minimax_h3_video_vae_fp16.safetensors FAQ

fp16 or int8_convrot?

Use fp16 if it fits. The int8_convrot file halves the size, and Comfy-Org recommends int8_convrot formats when your PyTorch build has CUDA 13 (cu130).

Does the path minimax/minimax_h3_video_vae_fp16.safetensors mean a different file?

No. Some workflows keep H3 files in a models/vae/minimax/ subfolder, so the dropdown shows the folder name. Put the file there or pick it again from the dropdown.

Where do I put minimax_h3_video_vae_fp16.safetensors in ComfyUI?

In ComfyUI/models/vae/. Then refresh the node list or restart ComfyUI so Load VAE lists the file.

Which ComfyUI node loads minimax_h3_video_vae_fp16.safetensors?

Load VAE. It is built into ComfyUI, so no custom nodes are needed.

Can I use minimax_h3_video_vae_fp16.safetensors commercially?

Yes, within the MiniMax H3 Community License. Commercial products must show “MiniMax H3” in their interface, and companies with more than US$20 million in yearly revenue need written permission from MiniMax first. The license also limits use to its listed territories, so read it before shipping.

Sources: Comfy-Org/MiniMax-H3 on Hugging Face · MiniMaxAI/MiniMax-H3 (original weights) · ComfyUI MiniMax H3 tutorial

RunComfy
Copyright 2026 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.