Summary of Oneactor: Consistent Character Generation Via Cluster-conditioned Guidance, by Jiahao Wang et al.
OneActor: Consistent Character Generation via Cluster-Conditioned Guidance
by Jiahao Wang, Caixia Yan, Haonan Lin, Weizhan Zhang, Mengmeng Wang, Tieliang Gong, Guang Dai, Hao Sun
First submitted to arxiv on: 16 Apr 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach, OneActor, is proposed to address the challenge of generating consistent images of the same subject using text-to-image diffusion models. Existing methods often rely on external data or require expensive tuning of the model. OneActor achieves one-shot tuning solely driven by prompts and learned semantic guidance, bypassing laborious backbone tuning. The method formalizes the objective of consistent subject generation from a clustering perspective and designs a cluster-conditioned model. To mitigate overfitting, auxiliary samples are added to the tuning process, and two inference strategies – semantic interpolation and cluster guidance – are devised. Experimental results show that OneActor outperforms various baselines in terms of subject consistency, prompt conformity, and image quality. The method is compatible with popular diffusion extensions, offers a 4x faster tuning speed than tuning-based baselines, and can be used to pre-train consistent subject generation networks from scratch. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper solves the problem of generating consistent images of the same subject using text-to-image models. Current methods often need external data or are time-consuming. The new approach, called OneActor, makes it possible to generate consistent images with just one try and a prompt. It uses learned guidance to help the model make better choices. To prevent overfitting, the method adds extra examples during training. The results show that this method works better than others in generating consistent images with good quality. It’s also faster and can be used as a starting point for other image generation tasks. |
Keywords
» Artificial intelligence » Clustering » Diffusion » Image generation » Inference » One shot » Overfitting » Prompt
