CRAFT: Continuous Reasoning and Agentic Feedback Tuning for Multimodal Text-to-Image Generation
V. Kovalev, A. Kuvshinov, A. Buzovkin, D. Pokidov, D. Timonin
TL;DR
CRAFT introduces a training-free, modular framework that brings explicit, constraint-driven reasoning to multimodal text-to-image generation. It decomposes prompts into dependency-structured visual questions, verifies outputs with a vision-language model, and applies targeted prompt edits via an LLM until all constraints are satisfied, enabling an interpretable inference-time refinement loop. Across five backbone families and benchmarks like DSG1K and Parti-Prompt, CRAFT consistently improves compositional accuracy, text rendering, and artifact handling, with pronounced gains for lighter generators and negligible overhead. The results suggest that structured, constraint-based reasoning at inference time offers a practical, scalable path to more reliable multimodal generation without retraining.
Abstract
Recent work has shown that inference-time reasoning and reflection can improve text-to-image generation without retraining. However, existing approaches often rely on implicit, holistic critiques or unconstrained prompt rewrites, making their behavior difficult to interpret, control, or stop reliably. In contrast, large language models have benefited from explicit, structured forms of **thinking** based on verification, targeted correction, and early stopping. We introduce CRAFT (Continuous Reasoning and Agentic Feedback Tuning), a training-free, model-agnostic framework that brings this structured reasoning paradigm to multimodal image generation. CRAFT decomposes a prompt into dependency-structured visual questions, veries generated images using a vision-language model, and applies targeted prompt edits through an LLM agent only where constraints fail. The process iterates with an explicit stopping criterion once all constraints are satised, yielding an interpretable and controllable inference-time renement loop. Across multiple model families and challenging benchmarks, CRAFT consistently improves compositional accuracy, text rendering, and preference-based evaluations, with particularly strong gains for lightweight generators. Importantly, these improvements incur only a negligible inference-time overhead, allowing smaller or cheaper models to approach the quality of substantially more expensive systems. Our results suggest that explicitly structured, constraint-driven inference-time reasoning is a key ingredient for improving the reliability of multimodal generative models.