CREA: A Collaborative Multi-Agent Framework for Creative Image Editing and Generation

TL;DR

CREA presents a collaborative multi-agent framework for creative image editing and generation that mirrors human creative workflows. It decomposes the process into specialized roles—Creative Director, Prompt Architect, Generative Executor, Art Critic, and Refinement Strategist—guided by six creativity principles and evaluated via a Creativity Index. Through iterative planning, generation/editing, evaluation, and self-enhancement, CREA achieves greater diversity, semantic alignment, and artistic transformation than state-of-the-art baselines. The approach enables user-guided co-creation and has potential for extending to video and personalized workflows, highlighting the viability of autonomous, creative AI systems.

Abstract

Creativity in AI imagery remains a fundamental challenge, requiring not only the generation of visually compelling content but also the capacity to add novel, expressive, and artistically rich transformations to images. Unlike conventional editing tasks that rely on direct prompt-based modifications, creative image editing requires an autonomous, iterative approach that balances originality, coherence, and artistic intent. To address this, we introduce CREA, a novel multi-agent collaborative framework that mimics the human creative process. Our framework leverages a team of specialized AI agents who dynamically collaborate to conceptualize, generate, critique, and enhance images. Through extensive qualitative and quantitative evaluations, we demonstrate that CREA significantly outperforms state-of-the-art methods in diversity, semantic alignment, and creative transformation. To the best of our knowledge, this is the first work to introduce the task of creative editing.

Figures (31)

• Figure 1: We introduce CREA, an agentic framework inspired by the human creative process for image editing and generation. Our approach can be extended to video domain for creative video generation or can be integrated with personalization methods to further enrich creative workflows.
• Figure 2: CREA Framework. We introduce a collaborative multi-agent framework for creative image editing and generation. Our framework consists of four stages, 1.a Pre-Generation Planning, 1.b Creative Image Generation/Editing, 2. Post-Generation Evaluation and 3. Self-Enhancement. Here, K is the number of maximum iterations.
• Figure 3: Qualitative Results for Creative Image Editing and Generation Tasks. (a) Creative Image Editing: CREA takes either a real-world or AI generated input image to produce a conceptually enriched and stylistically novel edit while preserving key structural elements. (b) Creative Image Generation: CREA receives only a minimal concept description (e.g., “a couch” or “a car”) and generates diverse, imaginative outputs without any visual input, extrapolating rich visual metaphors and materials. These results demonstrate CREA’s ability to disentangle and control creativity across editing and generation workflows. For additional results, see the appendix section.
• Figure 4: Qualitative Comparison of Creative Image Editing Task. We compare CREA with state-of-the-art editing methods. As shown, CREA successfully reimagines objects into creative variants in a disentangled manner, whereas other approaches either fail to produce distinctly creative edits or introduce unintended alterations.
• Figure 5: Qualitative Comparison of Creative Image Generation Task. We compare CREA with ConceptLab, SDXL and Flux. CREA consistently produces diverse and creative generations across multiple domains.