Advancing the application of intelligent design systems in adaptive co-creation models using AIGC and reinforcement learning

In current intelligent design systems, text prompt word optimisation is a key challenge to improve the quality of image generation.Aiming at the problems of uncertain direction of prompt words and difficult quality evaluation, this paper develops an adaptive co-creation based on AIGC and reinforcement learning.The model adopts a three-stage training framework: first, supervised fine-tuning of the mapping relationship of prompt word pairs is performed, then multi-modal visual feedback of PickScore and aesthetic value models is fused through reward modelling, and finally reinforcement learning is used to fine-tune and optimise the generation strategy.The model performs well in many indexes.In the performance test, the FID of the model decreases to 15.2 ± 1. 5 and the IS increases to 8.9 ± 0. 6.In the robustness test, the FID of the model is 17.5 ± 1. 7 under the noise prompt.In addition, in the practical test, the overall user satisfaction reaches 4.4 ± 0. 3.These results show that the model realises the collaborative optimisation of prompt words and image generation through adaptive mechanism, and provides an efficient end-to-end optimisation scheme.However, the model still needs to be further refined.Therefore, in future research directions, it is necessary to focus on reducing complexity and enhancing real-time feedback integration.