AI UX PlaygroundNewsletterJoin 2K+ AI designers and PMs on Substack. New teardowns, patterns, and prompts as they drop.

Design Tools

In-painting

Mask a region of an image and edit only that area with a prompt. The rest of the frame stays put. Local edits without regenerating everything.

Interactive demo

Overview

How might we design in-painting so people can trust and act on AI output?

When to use

  • Perfect for AI image editing tools, design software, and creative applications where users need precise control over specific image regions.

When to skip

  • The whole image needs a new concept, not a local fix.
  • Mask tools would confuse casual users on mobile.
  • Output must be pixel-identical outside the edit for compliance.

Rules

  • No visible mask preview before generation.

  • Seams or color shifts at mask edges with no refine pass.

  • Regenerating the entire image when only a patch changed.

  • Hiding which pixels were model-generated.

Evidence

ProductImplementation
Adobe FireflyGenerative fill inside selections with non-destructive layers.
DALL-E 3Edit region flows in ChatGPT image tools.
Stable DiffusionInpaint nodes and mask brushes in WebUI and ComfyUI.
MidjourneyVary region and inpaint-style refinement on upscaled images.

FAQ

What is inpainting in AI image UX?

Users select an area, describe the change, and the model fills only that region while blending with surrounding pixels.

How does inpainting differ from full regeneration?

Full regeneration replaces the image. Inpainting preserves everything outside the mask for iterative local edits.

What mask UX matters most?

Brush size, feather, undo, and a clear overlay so users know exactly what will change.

When pair with object removal?

Object removal is inpainting with an empty prompt or “remove” intent. Same mask UX, different default action.