Skip to content

Differential Diffusion

Set how much each image region changes

Visit official site

What Differential Diffusion is

The framework is described as a platform for image generation and editing that modifies a picture from a text prompt and a map specifying how much each region changes. It offers customised change amounts for every pixel or region, which opens up control over how much an individual object is altered.

What you can do with it

  • Edit an image from a text prompt
  • Assign a change amount to each region
  • Control how much an individual object is modified
  • Introduce gradual spatial changes across the image

Example uses

  • Edit an input image with the prompt tree life under the sea and a change map
  • Edit an input image with the prompt racing game and a discrete change map
  • Edit an input image with the prompt dream world and a continuous change map
  • Integrate the Stable Diffusion XL model and edit with the prompt wool and a change map

Model notes

  • The framework runs only at inference time with no model training or fine-tuning
  • Image completion blends new content while tuning surrounding areas
  • Tools are provided for exploring different change amounts
  • Integration with open-source models plus quantitative, qualitative and user study comparisons are described

Pros & cons

✓ What we like

  • Fine-grained control over each pixel or region
  • No model training or fine-tuning is required

! What to watch out for

  • Edits need both a text prompt and a change amount map
  • Outputs depend on the change amounts set in the map

FAQ

Does it need training?

The material states the framework runs only at inference time and needs no model training or fine-tuning.

What does the map control?

It sets the amount of change for each pixel or image region, including gradual spatial changes.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review