Skip to content

InstructIR

Restore images by following written instructions

Visit official site

What InstructIR is

The model is described as taking an image and a human-written instruction as input and performing all-in-one image restoration with a single neural model. The material lists denoising, deraining, deblurring, dehazing and low-light image enhancement, and points to a demo tutorial and a GitHub repository for more information.

What you can do with it

  • Denoise an input image
  • Remove rain, blur or haze from an image
  • Enhance a low-light image
  • Try a general image enhancement prompt

What to watch out for

  • The demo needs an image with blur, noise, rain, low light or haze and a prompt describing the operation
  • The demo is stated to be not a product and to have limits
  • High-resolution inputs such as 2K and 4K may crash the app under GPU memory limits
  • Training used mainly synthetic data, so complex real-world images may not work well

Pros & cons

✓ What we like

  • One neural model handles several restoration tasks
  • General image enhancement prompts can also be tried

! What to watch out for

  • The demo is described as not being a product and has limits
  • High-resolution inputs such as 2K and 4K may crash the app due to GPU memory limits

FAQ

What inputs does the demo need?

An image with a degradation such as blur, noise, rain, low light or haze, plus a prompt describing the operation.

Which restoration tasks are listed?

The feature list names denoising, deraining, deblurring, dehazing and low-light image enhancement.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review