方法对比

	微调扩散模型	微调变分自编码器
领域	深度学习	深度学习
方法族	Machine learning	Machine learning
起源年份	2020–2023	2014 (VAE); fine-tuning practice from 2015 onward
提出者	Ho, J., Jain, A., Abbeel, P. (base DDPM); Ruiz et al. (DreamBooth fine-tuning paradigm)	Kingma, D. P. & Welling, M. (VAE); fine-tuning strategy from transfer learning literature
类型	Generative model (fine-tuned via subject-specific or domain-specific data)	Generative model with fine-tuning
开创性文献	Ruiz, N., Li, Y., Jampani, V., Pritch, Y., Rubinstein, M., & Aberman, K. (2023). DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 22500–22510.	Kingma, D. P., & Welling, M. (2014). Auto-Encoding Variational Bayes. In Proceedings of the 2nd International Conference on Learning Representations (ICLR 2014).
别名	DDPM fine-tuning, diffusion model adaptation, personalized diffusion model, subject-driven diffusion fine-tuning	fine-tuned VAE, domain-adapted VAE, transfer-learned VAE, adapted variational autoencoder
相关	5	6
摘要	A fine-tuned diffusion model adapts a large pretrained denoising diffusion model — such as Stable Diffusion or DALL-E — to a specific subject, style, or domain by continuing training on a small curated dataset. Techniques such as DreamBooth, textual inversion, and LoRA make this adaptation feasible on consumer hardware while preserving general generative capability.	A Fine-Tuned Variational Autoencoder begins with a VAE pre-trained on a large source dataset and then continues training on a smaller target-domain dataset. This approach adapts the learned latent representation and generative capacity to new data, preserving general structure while specializing to the target distribution — yielding better results than training from scratch when labeled or large target data is scarce.