Atelier adds user controls for smarter AI image editing
From Semantic Editing Intentions to Persistent Parametric Controls for AI-Assisted Image Editing
Human-Computer Interaction
Summary
Editing images with AI can be confusing because it’s hard to see how the AI changes match what you want. The authors created a tool called Atelier that shows users simple controls based on what the AI thinks you want to do, so you can adjust these controls yourself instead of accepting the AI’s edits right away. People who tested Atelier liked having these controls because it helped them explore creative ideas. However, some found it unclear why certain controls were offered, showing that control alone doesn’t explain AI behavior.
What this means in practice
- •For graphic designers: Use Atelier’s controls to adjust AI-proposed image edits with more precision and creativity during design work.
- •For photo editing software developers: Integrate context-aware parametric controls into photo editing apps to let users refine AI-driven edits interactively.$Commercial implications: Enables development of AI-powered editing features that improve user control and satisfaction in commercial photo editing software.
Authors
Rifat Mehreen Amin, Michelle Dutoit, Anton Rockenstein, Baptiste Caramiaux, Andreas Butz
Abstract
AI-powered image editing increasingly allows users to express high-level editing intentions, yet how these intentions are translated into concrete editing actions often remains difficult to understand. We present Atelier, a semantic image editing tool that uses image semantics and user intent to determine which editing controls to surface and how to initialize them. Guided by a formative study with eight image-editing practitioners, Atelier composes context-specific, directly manipulable controls that users can refine rather than committing AI interpretations directly to finished edits. In a within-subjects study with 12 practitioners, participants preferred AI-composed controls, described them as supporting creative exploration, and actively refined AI proposals. However, experiences of control were mixed, and interviews highlighted limited insight into why particular controls and values were proposed. Our findings suggest that making AI interpretations actionable through manipulable controls can support exploration and refinement, but actionability alone does not make those interpretations transparent to all users.