Copyright © 2025 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{178361,
        author = {Vishal Kumar Sinha and Samriddhi Negi and Shubhra Komal and Arup Gope and Dr. Savitha Chaudhary},
        title = {Generative AI with Diffusion Models: Toward Efficient, Fair and Scalable Content Synthesis},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {11},
        number = {12},
        pages = {2988-2990},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=178361},
        abstract = {Diffusion-based generative models have eclipsed GAN and VAE architectures in fidelity, robustness and mode coverage, but their iterative denoising chains impose steep com- putational and energy budgets. We present a complete stack that (i) fuses Transformer self-attention and convolution into a lightweight denoiser, (ii) compresses a 1 000-step teacher into a four-step student via progressive distillation, (iii) applies loss- aware pruning, mixed-precision kernels and adaptive timestep scheduling, and (iv) embeds a real-time bias-detection guardrail. Trained on a 550 k image–text corpus filtered for legal and ethical compliance, the system delivers an FID of 6.9 on MS- COCO while running 4.6× faster and 4.5× greener than a 50- step baseline, and it surpasses Stable Diffusion 1.5 by 1.2 FID at 38 % lower energy. Experiments on desktop GPUs, laptop GPUs and edge NPUs confirm viability for interactive design, AR filters and mobile creativity apps, moving diffusion models closer to trustworthy, resource-aware deployment.},
        keywords = {Generative AI, diffusion models, model com-pression, image synthesis, AI fairness, energy efficiency},
        month = {May},
        }