ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization

A new video anonymization pipeline called ReGenHuman synthesizes full-body human appearances from identity-free structural cues, achieving what its creators describe as the first system to be simultaneously realistic, temporally consistent, and anonymous by construction [1][2]. The work, posted to the arXiv preprint server on June 12, 2026, addresses a gap in human-centric video anonymization. Prior methods either blur or redact pixels, sacrificing realism and downstream utility, or generate frames independently, breaking temporal coherence [2]. The authors propose a “regenerate, don’t edit” paradigm that composites 2D pose, segmentation, and monocular depth into two conditioning streams called StructAll and StructHuman [2]. These streams fine-tune a video-to-video diffusion backbone on in-the-wild human videos, synthesizing human regions entirely from structural information that carries no identity signal [2]. The researchers evaluated ReGenHuman on privacy, quality, and utility, reporting that it achieves the best tradeoff across all three axes against current baselines [2]. They also demonstrated that the anonymized videos remain effective for downstream tasks, including video question answering [2]. The paper appears on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives roughly 24,000 submissions per month and has surpassed two million articles [6]. The repository is not peer-reviewed; papers are approved after moderation [6]. The ReGenHuman abstract page includes the arXivLabs framework, a set of experimental tools developed by community collaborators that appears as tabs below the article record [4]. arXivLabs was formalized in 2020 to enable individuals and organizations to build features that enhance the reading experience while adhering to arXiv’s values of openness, community, excellence, and user data privacy [4]. Current Labs integrations include the Bibliographic Explorer for navigating citation trees, the CORE Recommender for discovering related open-access papers, and links to code and data via Papers with Code [5]. The ReGenHuman page also surfaces Connected Papers and Litmaps, tools that generate interactive citation maps to help researchers explore the literature surrounding a preprint [5]. These integrations are part of arXiv’s broader effort to let the community contribute functionality without altering the core mission of rapid, no-cost dissemination of research findings [5].