reComposer replaces one-shot prompting with an editable plan that exposes intermediate image-making decisions

The reComposer paper, submitted on November 9, introduces an image-generation interface centered on an editable plan. It decomposes intent into elements such as subject, composition, and palette, lets a user revise each through conversation or direct manipulation, then synthesizes the plan into a final prompt and image. In a first-use study, eight professionals who create images in their day-to-day work reported greater control and ownership than with ordinary prompting.

Why it matters

The work shifts the unit of generative UI from a finished image to visible, editable, and delegable intermediate decisions. For product teams, this shared representation can expose model assumptions, support local revision, and make alternatives easier to explore. It also shows how conversation and direct manipulation can operate on the same generated state.

A closer look

The system creates a plan from the user's goal, decomposes it into elements that can be added, removed, edited, or assigned influence, and supports both conversational updates and node-level manipulation. Examples cover stylized data visualization, infographic design, architectural concept rendering, and animation storyboards. The study included eight roles spanning graphic, UI/UX, and visual design, art direction, marketing, account management, and machine learning; participants used tasks from their own work. The authors also ran a non-interactive pipeline validation indicating that planning generally preserved the underlying model's image quality, but that is not a longitudinal evaluation of production outcomes.

Who this is for

Research, product, design, and frontend teams building creative tools, image generation, generative UI, or human-AI co-creation workflows.

Original sources

• reComposer paper and full text ↗arxiv.org
• reComposer full text, study design, findings and limitations ↗arxiv.org