Research map / Image generation / Text-to-image diffusion

MCCD: Multi-Agent Collaboration-based Compositional Diffusion for Complex Text-to-Image Generation

CVPR 2025 · Image generation › Text-to-image diffusion · approach: Compositional text-to-image guidance · 7 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes MCCD, a training-free framework combining a GPT-4o-mini multi-agent scene parser with hierarchical compositional diffusion (Gaussian masks, regional enhancement, latent smoothing) to improve complex text-to-image generation of SD and SDXL models.

Problem

Diffusion text-to-image models struggle with complex prompts involving multiple objects, attributes, and relations, yielding poor attribute binding and spatial relations.

Approach

A multi-agent system of MLLM agents (object, background, action, spatial, layout, aesthetics) coordinated by a conductor and evaluator with forward CoT and backward feedback parses the prompt into object prompts, background prompt, and bounding boxes; a hierarchical compositional diffusion then fuses per-object cross-attention latents using depth-aware Gaussian masks, regional enhancement, and Gau

Key results

• SDXL-base + MCCD improves Color from 0.5744 to 0.6278 and Texture from 0.1971 to 0.2350 on T2I-CompBench.
• Reported overall improvement up to 9.04% across diffusion models, with Spatial metric increased by up to 41.49%.
• SD1.5 + MCCD raises Color from 0.3134 to 0.3508 on T2I-CompBench.

Measured effects reported in the paper

Limitations

• Relies on proprietary GPT-4o-mini agents for parsing, so reproducibility depends on API access.
• Evaluation limited to T2I-CompBench six metrics; gains on Texture and Complex remain modest in absolute terms.
• Several hyperparameters (lambda_pos, lambda_neg, mu, sigma) are fixed without reported sensitivity analysis.

Experimental setup

Datasets: T2I-CompBench

Benchmarks: T2I-CompBench

Metrics: Color Shape Texture Spatial Non-Spatial Complex

Baselines: Composable Diffusion Structured Diffusion Attn-Exct v2 GORS DALL-E 2 PixArt-α

Builds on: Stable Diffusion RPG LMD GLIGEN

Closest papers by meaning

• Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML 2024 · similarity 0.88)
• LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.87)
• Generating compositional scenes via Text-to-image RGBA Instance Generation (NeurIPS 2024 · similarity 0.87)
• Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.86)
• IterComp: Iterative Composition-Aware Feedback Learning from Model Gallery for Text-to-Image Generation (ICLR 2025 · similarity 0.86)
• Compositional Text-to-Image Synthesis with Attention Map Control of Diffusion Models (AAAI 2024 · similarity 0.86)