pig-vae

pig-vae

A small GGUF VAE in the pig architecture for image and video workflows, with a loader in the gguf-node package. Aimed at faster runs and lower memory use on older machines.

Made by calcuis on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: pig-vae is a VAE, an image encoder and decoder, packed as a GGUF file in the pig architecture. It has about 83.7M parameters, and a 4-bit copy would be around 42 MB. It plugs into node-based workflows through the gguf-node package and its GGUF VAE loader.

Who it suits: It suits people who build image or video generation workflows and want a lighter VAE, especially on an older computer. The author says it is compatible with every model, whether the model is in safetensors or GGUF form.

Title: Who uses it
Industry | Who uses it | How they use it | Example
Creative production | Workflow builders | Swap in a GGUF VAE to save memory | Load the pig VAE through the GGUF VAE loader
Software | Tool developers | Pair it with other GGUF models in one workflow | Combine it with a GGUF model of their choice
Hobby use | Owners of older machines | Run generation workflows with less memory | Fit a workflow onto an aging computer

People who do not build generation workflows can skip it.

Where to use it:

• Node-based workflows that use the gguf-node package
• Windows, macOS or Linux with Python, to download the file
• Older machines that need to save memory

Title: Top jobs
Use case | Who | What they get back
Decoding images | Workflow builders | A VAE that works with safetensors and GGUF models
Saving memory | Owners of older machines | A smaller footprint, which the author puts at 25 to 50 percent
Converting files | Tool developers | GGUF versions of models, encoders and VAEs

How to set it up:

• Python on your computer
• The huggingface_hub package
• The gguf-node package, upgraded to a version that supports the pig VAE

1. Install the Hugging Face hub package.
2. Download the model file to your computer.
3. Upgrade your gguf-node setup so it has the GGUF VAE loader.
4. Point the loader at the downloaded file in your workflow.

pip install -U huggingface_hub
hf download calcuis/pig-vae

How to use it: Use the new GGUF VAE loader in your node workflow and select this file. The author says you can drag the picture or video from the model page into your browser to get an example workflow. A conversion tool called convertor zero is suggested for making GGUF models, encoders and VAEs.

Example: You download the file, upgrade gguf-node and add the GGUF VAE loader to a workflow. You choose pig-vae as the VAE, then run the workflow as usual. The card shows no printed output, so check the generated image yourself.

For newcomers and experts: If you are new, a VAE turns the model's internal image data into the picture you see. A first try is to load the example workflow from the model page and swap in this file.

For experts, the author claims the pig architecture runs 25 to 50 percent faster than the safetensors version and saves 25 to 50 percent of memory. The card also lists a newer FLUX.2 VAE file named pig_flux2_vae_fp32-f16.gguf in its update note.

Download the pig VAE file
Upgrade gguf-node
Add the GGUF VAE loader to a workflow
Run the workflow

Good to know:

• Claims: the speed and memory figures come from the author, and the card shows no measurements.
• Update: a newer VAE file for FLUX.2 is mentioned in the card.
• License: MIT.
• Cost to run: none named.
• Data it reaches: it downloads files from Hugging Face, and the source names nothing else.
• Not for: use without the gguf-node loader, which the card says is needed.

On Hugging Face, 13 November 2026

• Parameters: 83.7M
• Weights: about 42 MB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 8,869
• Likes: 46
• License: mit
• Languages: en
• Tags: gguf, gguf-node

Links

• Model: https://huggingface.co/calcuis/pig-vae
• Files: https://huggingface.co/calcuis/pig-vae/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download calcuis/pig-vae

Then follow the usage in its model card. Nothing in this kit runs by itself.