<p>In text-to-image generation, capturing fine-grained details in controllable generation remains a significant challenge, especially in scenarios involving complex control inputs. To address this limitation, we propose FLC-T2I, an innovative Feature Loop Consistency optimization model. The model divides the training process into two stages: in the first stage, a pre-trained extractor is used to reverse-extract control conditions from the images generated by the model. This new control information is used as a dataset for the second round of training, establishing a closed-loop relationship between the input and generated images. The second stage introduces a new cycle consistency loss to enhance fine-grained control over the generated images. Compared to existing methods, the second-stage feature loop strategy refines the consistency within the closed loop, improving the precision of controllable generation. Through comprehensive quantitative and qualitative analysis, we demonstrate the reliability of the FLC-T2I model and its superior fidelity in control tasks, highlighting the effectiveness of the secondary loop strategy in achieving fine-grained and precise control.</p>