BusterX++: Towards Unified Cross-Modal AI-Generated Content Detection and Explanation with MLLM
Organizations: University of Liverpool, UK
Abstract
The rapid advancement of generative AI has substantially improved image and video synthesis, amplifying the risk of multimodal visual misinformation. Recent MLLMs have shown promise for transparent AI-generated content detection through reasoning and explanation, yet existing approaches largely treat image and video forensics as isolated tasks, leaving cross-modal synergies underexplored. To address this, we present BusterX++, a unified MLLM for joint image and video detection with interpretable reasoning. We also introduce GenBuster-Bench++, a curated benchmark containing balanced image and video samples spanning recent generation models and diverse real-world scenarios, with synthetic images and videos screened using the same human difficulty criterion. Using this controlled setting, we revisit the widely adopted SFT+RL post-training paradigm. Notably, in the unified image-video setting, our results demonstrate that a single-stage pure RL strategy driven primarily by sparse outcome rewards outperforms a strong SFT+RL baseline. We observe that single-stage pure RL maintains higher policy entropy than SFT-initialized RL throughout training. This observation suggests a possible role for exploration in the performance gains observed under unified image-video training. Extensive experiments demonstrate that BusterX++ achieves state-of-the-art performance, highlighting the powerful potential of RL for unified cross-modal visual reasoning.
Figures & tables
| MLLMs | Image | Video | Overall | ||
|---|---|---|---|---|---|
| Real | Fake | Real | Fake | ||
| GLM-4.6V-Flash | 96.8 | 02.3 | 91.0 | 32.4 | 55.6 |
| Qwen3.5-4B | 25.6 | 11.5 | 85.6 | 27.7 | 37.6 |
| Qwen3.5-122B-A10B | 82.1 | 23.9 | 83.4 | 32.8 | 55.6 |
| Qwen3.6-27B | 80.0 | 27.7 | 87.3 | 34.8 | 57.5 |
| BusterX | 82.1 | 68.3 | 88.7 | 73.4 | 78.1 |
| Training | Data | Image | Video | Overall | ||||
| SFT | RL | Image | Video | Real | Fake | Real | Fake | |
| ✓ | - | ✓ | - | 53.6 | 60.5 | 79.8 | 58.5 | 63.1 |
| ✓ | - | - | ✓ | 48.0 | 56.1 | 91.0 | 66.7 | 65.5 |
| ✓ | - | ✓ | ✓ | 64.4 | 59.7 | 90.8 | 63.5 | 69.6 |
| ✓ | ✓ | ✓ | ✓ | 84.2 | 63.6 | 93.4 | 70.1 | 77.8 |
| - | ✓ | ✓ | - | 85.9 | 75.4 | 82.7 | 65.2 | 77.3 |
| Methods | ID (2024) | OOD (2025) | Wild (2026) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Real | Fake | ACC | Sora | Pika | Gen3 | Luma | WanX | Kling | Jimeng | Vidu | Fake | ACC | Fake | ACC | |
| Classical Detectors | |||||||||||||||
| 3D ResNeXt Hara et al. (2018) | 84.4 | 60.8 | 72.6 | 31.0 | 75.0 | 39.0 | 42.0 | 48.7 | 22.0 | 65.0 | 46.0 | 44.7 | 64.6 | 39.3 | 61.9 |
| Vivit Arnab et al. (2021) | 92.9 | 64.1 | 78.5 | 74.0 | 52.0 | 69.0 | 53.0 | 62.0 | 55.0 | 33.0 | 68.7 | 60.6 | 76.8 | 46.7 | 69.8 |
| VideoMAE Tong et al. (2022) | 93.3 | 64.9 | 79.1 | 48.5 | 79.0 | 65.0 | 52.0 | 76.7 | 24.0 | 55.0 | 72.0 | 59.5 | 76.4 | 50.7 | 72.0 |
| DeMamba Chen et al. (2024) | 93.8 | 70.2 | 82.0 | 65.0 | 75.0 | 73.0 | 56.0 | 46.0 | 68.0 | 71.0 | 84.7 | 66.9 | 80.4 | 48.0 | 70.9 |
| Perturbation | Overall | ||
| JPEG | Noise | Blur | |
| - | - | - | 81.8 |
| ✓ | - | - | 78.9 |
| - | ✓ | - | 76.1 |
| - | - | ✓ | 80.4 |
| ✓ | ✓ | ✓ | 75.3 |
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.