AuraSR-v2

AuraSR-v2

Upscales generated images four times with a GAN-based super-resolution model. Made for people who generate pictures and want them larger, from fal.

Made by fal 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: AuraSR-v2 is a super-resolution model that enlarges images made by image generators. It is a GAN, a variation on the GigaGAN paper adapted to upscale from an input image. The Torch code builds on an unofficial GigaGAN implementation.

Who it suits: It suits people who make pictures with AI and want a larger copy, plus developers who add an upscaling step to an image pipeline.

Title: Who runs it
Industry | Who uses it | How they use it | Example
Design | Artists who use image generators | Enlarge a generated picture | A small render turned into a larger copy
Software | Developers | Add an upscale step after image generation | A Python script that upscales each output
Research | Image researchers | Study GAN-based upscaling | Compare the output with other upscalers

If you do not make or edit images, you can skip it.

Where to use it:

• Python on Windows, macOS or Linux, through the aura-sr package
• Image pipelines, as a step after a generator produces a picture
• The transformers library, the library tag on the model page

Title: Jobs it does
Use case | Who | What they get back
Upscaling a generated image | Artists | A larger copy of the picture
Overlapped four-times upscaling | Developers | A four-times enlarged image from one call

How to set it up:

• Python on Windows, macOS or Linux
• The aura-sr package
• A picture to upscale, loaded with PIL

1. Install the aura-sr package with pip.
2. Load the model by name with from_pretrained.
3. Optionally download the model files by hand.

$ pip install aura-sr

from aura_sr import AuraSR

aura_sr = AuraSR.from_pretrained("fal/AuraSR-v2")

pip install -U huggingface_hub
hf download fal/AuraSR-v2

How to use it: Load a picture, shrink it if you like, and call the four-times overlapped upscale on it. The card resizes its sample picture to 256 by 256 first.

upscaled_image = aura_sr.upscale_4x_overlapped(image)

Title: Calls it shows
Tool | What it does
AuraSR.from_pretrained | Loads the model from the name fal/AuraSR-v2
upscale_4x_overlapped | Enlarges an image four times, working in overlapping parts

Example: You take a 256 by 256 picture from an image generator and pass it to the upscale call. The call returns a new, larger image, which you can then save or view with PIL.

For newcomers and experts: If you are new, this is a tool that makes a small picture bigger while trying to add believable detail. The first small thing to try is the install line and the load line, then one small image.

If you know the field, the model is a GAN with 618M parameters, adapted from GigaGAN for image-conditioned upscaling. The overlapped method handles the picture in overlapping parts. The trade-off is that a GAN invents detail, so the result is a plausible enlargement rather than a recovery of the original.

Load the image with PIL
Resize it to 256 by 256
Call the overlapped four-times upscale
Get back a larger image

Good to know:

• Size: 618M parameters, about 2.5 GB in F32.
• Method: GAN-based, adapted from the GigaGAN paper.
• License: apache-2.0.
• Cost to run: no API key, account or paid service is named.
• Data it reaches: it downloads the model from Hugging Face, and the card's sample also fetches a picture from the web.

On Hugging Face, 13 November 2026

• Task: image to image
• Library: transformers
• Parameters: 618M
• Weights: about 2.5 GB in F32; about 309 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,554
• Likes: 349
• License: apache-2.0
• Tags: transformers, safetensors, pytorch, super-resolution, image-to-image

Links

• Model: https://huggingface.co/fal/AuraSR-v2
• Files: https://huggingface.co/fal/AuraSR-v2/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download fal/AuraSR-v2

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