# HumanForge: A Human-Centric Deepfake Video Benchmark with Multi-Agent Forgery Rationales

arXiv: https://arxiv.org/abs/2607.08705 · HTML: https://mlsanity.com/paper/2607.08705/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models.
- Builds on 24 of its 40 references that are in the map.
- First posted on arXiv 9 Jul 2026.

## Placed in

- Diffusion models / [Image generation, text-to-image and editing](https://mlsanity.com/diffusion/image-generation-text-to-image-and-editing/)

## Builds on (most cited first, 24 of 24)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Wan: Open and Advanced Large-Scale Video Generative Models](https://mlsanity.com/paper/2503.20314/) (2025)
- [CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer](https://mlsanity.com/paper/2408.06072/) (2024)
- [HunyuanVideo: A Systematic Framework For Large Video Generative Models](https://mlsanity.com/paper/2412.03603/) (2024)
- [A Closer Look at Spatiotemporal Convolutions for Action Recognition](https://mlsanity.com/paper/1711.11248/) (2017)
- [FaceForensics++: Learning to Detect Manipulated Facial Images](https://mlsanity.com/paper/1901.08971/) (2019)
- [LTX-Video: Realtime Video Latent Diffusion](https://mlsanity.com/paper/2501.00103/) (2024)
- [StyleGAN-Human: A Data-Centric Odyssey of Human Generation](https://mlsanity.com/paper/2204.11823/) (2022)
- [Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding](https://mlsanity.com/paper/2601.10611/) (2026)
- [UniAnimate: Taming Unified Video Diffusion Models for Consistent Human Image Animation](https://mlsanity.com/paper/2406.01188/) (2024)
- [DeMamba: AI-Generated Video Detection on Million-Scale GenVideo Benchmark](https://mlsanity.com/paper/2405.19707/) (2024)
- [Animate-X: Universal Character Image Animation with Enhanced Motion Representation](https://mlsanity.com/paper/2410.10306/) (2024)
- [Face Forensics in the Wild](https://mlsanity.com/paper/2103.16076/) (2021)
- [HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning](https://mlsanity.com/paper/2509.08519/) (2025)
- [InfinityStar: Unified Spacetime AutoRegressive Modeling for Visual Generation](https://mlsanity.com/paper/2511.04675/) (2025)
- [Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence](https://mlsanity.com/paper/2604.24954/) (2026)
- [OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning](https://mlsanity.com/paper/2603.24458/) (2026)
- [SkyReels-V3 Technique Report](https://mlsanity.com/paper/2601.17323/) (2026)
- [OmniAID: Decoupling Semantics and Artifacts for Universal AI-Generated Image Detection in the Wild](https://mlsanity.com/paper/2511.08423/) (2025)
- [One-to-All Animation: Alignment-Free Character Animation and Image Pose Transfer](https://mlsanity.com/paper/2511.22940/) (2025)
- [A Multimodal Deviation Perceiving Framework for Weakly-Supervised Temporal Forgery Localization](https://mlsanity.com/paper/2507.16596/) (2025)
- [Open-World Deepfake Attribution via Confidence-Aware Asymmetric Learning](https://mlsanity.com/paper/2512.12667/) (2025)
- [HumanSAM: Classifying Human-centric Forgery Videos in Human Spatial, Appearance, and Motion Anomaly](https://mlsanity.com/paper/2507.19924/) (2025)
- [ActivityForensics: A Comprehensive Benchmark for Localizing Manipulated Activity in Videos](https://mlsanity.com/paper/2604.03819/) (2026)

## Abstract (arXiv)

Rapid advancements in video diffusion models and temporal editing tools have enabled the generation of highly realistic human-centric videos, presenting unprecedented challenges to digital content forensics. Existing benchmarks primarily focus on face-swapping or global text-to-video synthesis, overlooking the crucial dimensions of multimodal alignment and complex human-object or human-human interactions. To address these limitations, we introduce HumanForge, a unified, large-scale, and multi-paradigm human-centric video forgery benchmark containing over 18,000 synthesized videos across four distinct scenarios: audio-driven, pose-driven, semantic-driven, and interaction. To construct and annotate this dataset without labor-intensive manual labeling or blind monolithic prompting, we propose Gen2Anno (Generation-to-Annotation), a cooperative multi-agent pipeline. Gen2Anno orchestrates six specialized agents-ranging from driving asset profiling to MoE-based reference analysis and closed-loop verification-to dynamically execute video synthesis and produce structured annotations containing binary authenticity labels, generative model attribution, and natural-language contrastive forgery rationales. By systematically contrasting expected states derived from generation provenance with actual visual observations, the framework generates logically grounded forensic reasoning chains. Extensive benchmarks using state-of-the-art traditional detectors and Vision-Language Models demonstrate the significant challenges of cross-generator generalization, perturbation robustness, and explainable reasoning on HumanForge. The code and dataset will be publicly released.