reComposer: Editable Planning for Controllable Image Generation
Organizations: Tufts University Medford, MA, USA · Apple Cambridge, MA, USA · Apple Seattle, WA, USA · Apple Pittsburgh, PA, USA
Abstract
Prompt-based image generation makes it easy to produce visual assets, but offers little support for engaging with the creative process. Prompting collapses interdependent creative decisions into a monolithic instruction and returns a finished image, leaving control limited to revising the prompt or correcting the model's output. We present reComposer, a process-oriented image-generation interface built around editable planning, which decomposes users' evolving intent into meaningful creative decisions - such as subject, composition, or palette - and collaborates with users to explore underspecified goals and alternatives. reComposer supports conversational and direct manipulation of the plan, moving between concept-level intent and visual-level control. We evaluated reComposer with 8 design and non-design professionals. Compared to prompting, participants felt greater control and ownership while developing visual assets, exploring alternatives, and iterating on details. Our findings suggest how shifting generative AI interaction from specifying outputs to shaping process can support creative thinking alongside asset production.
Figures & tables
| Plan Element | Design Axis | Editor |
|---|---|---|
| subject | The main entity or entities: identity, appearance, pose. | Image editor |
| composition | Arrangement, framing, and spatial layout. | Composition editor: drag/resize/reorder regions; freeform scene or ordered-step graphic layout; sketch/pen tool. |
| background | The setting or backdrop behind the subject. | Image editor. |
| lighting | Light direction, quality, and time of day. Only for photographic or 3D-rendered scenes. | Image editor. |
| palette | The dominant and accent colors. | Palette editor: extract color palette from reference image; generate palette options from natural language prompts; |
| style | Medium and rendering aesthetic, applied globally to the whole image. | Image editor. |
| ID | Title | Task |
| P1 | Graphic Designer | Branding image of company products |
| P2 | UI/UX Designer | Video game stickers for a messaging app |
| P3 | Graphic Designer | Style frames for animation storyboarding |
| P4 | Visual Designer | Technical diagram for a presentation |
| P5 | Account Manager | Image for an employee training presentation |
| P6 | Art Director | Set of sales illustrations |
| Benchmark | Raw | Plan | [95% CI] |
|---|---|---|---|
| GenAI-Bench | .883 | .884 | |
| Qwen-Image-Bench | 60.2 | 56.9 |
| Facet | raw | plan | ||
|---|---|---|---|---|
| Cross-lingual Generation | 6/24/22 | 53/18/13 | ||
| Text Accuracy | 81/54/220 | 138/47/112 | ||
| Information Visualization | 21/22/58 | 43/17/34 | ||
| Font | 6/45/32 | 13/41/12 | ||
| Animals | 58/9/43 | 74/8/26 | ||
| Graphic Design | 3/48/80 | 14/60/52 |
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Human-AI Co-Creation
Human-AI co-creation studies how people and AI jointly develop ideas, artifacts, and decisions while preserving meaningful human agency.
See all 164 papers