Research map / Vision-language and multimodal / Multimodal LLMs
FakeWorld 1.0: An Omni-modal Benchmark for Fake Media and Content
Yifeng Gao, Yifan Ding, Li Wang, Feida Huang, Ye Sun, Yixu Wang, Xin Wang, Yutao Wu, Hanxun Huang, Yunhao Feng, Yingshui Tan, Xingjun Ma and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces FakeWorld 1.0, an omni-modal benchmark that unifies media authenticity and content veracity to evaluate detection of fake media and content.
Problem
Existing research treats media authenticity (real vs. machine-generated) and content veracity (semantic consistency and factual correctness) as separate problems, overlooking how real-world deception combines both.
Approach
A benchmark that jointly evaluates media authenticity and content veracity across multiple modalities, rather than along two separate axes.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add explicit thinking (ablation) | Qwen3-VL-235B-A22B · FakeWorld media authenticity (overall) | accuracy (%): 57.08 -> 71.52 | Improved |
| add explicit thinking (ablation) | Qwen3-VL-8B · FakeWorld media authenticity (overall) | accuracy (%): 65.13 -> 55.60 | Hurt |
| add explicit thinking (ablation) | Qwen3-VL-2B · FakeWorld media authenticity (overall) | accuracy (%): 50.08 -> 37.79 | Hurt |
| add explicit thinking (ablation) | Qwen3-VL-32B · FakeWorld media authenticity (overall) | accuracy (%): 63.90 -> 66.86 | Improved |
| add explicit thinking (ablation) | Qwen3-Omni-30B-A3b · FakeWorld media authenticity (overall) | accuracy (%): 61.17 -> 50.15 | Hurt |
| add explicit thinking (ablation) | Qwen3-VL-32B · FakeWorld fact verification | accuracy (%): 65.35 -> 72.78 | Improved |
| add explicit thinking (ablation) | Qwen3-VL-235B-A22B · FakeWorld cross-modal consistency (overall) | accuracy (%): 71.71 -> 75.01 | Improved |
| increase model size from Qwen3-VL-2B to 235B | Qwen3-VL · FakeWorld media authenticity (overall) | accuracy (%): 50.08 -> 57.08 | Improved |
| relax web search budget (ablation) | Gemini 3 Pro · FakeWorld fact verification | accuracy (%): 0.75 -> 1.27 | Improved |
| use decomposed OmniCheck workflow instead of single pass (ablation) | GPT-5.2 and Gemini 3 Pro · FakeWorld | accuracy (%) | Improved |
| use decomposed OmniCheck workflow instead of single pass (ablation) | Qwen3-VL-235B-Thinking · FakeWorld | accuracy (%) | Hurt |
| train detector jointly on all modalities instead of per modality (ablation) | LoRA Detector · FakeWorld image and video detection | accuracy (%) | Improved |
Experimental setup
Benchmarks: FakeWorld 1.0
Closest papers by meaning
- Seeing Through Deception: Uncovering Misleading Creator Intent in Multimodal News with Vision-Language Models (ICLR 2026 · similarity 0.88)
- Unleashing the Potential of Consistency Learning for Detecting and Grounding Multi-Modal Media Manipulation (CVPR 2025 · similarity 0.88)
- The Coherence Trap: When MLLM-Crafted Narratives Exploit Manipulated Visual Contexts (CVPR 2026 · similarity 0.87)
- Debate with Images: Detecting Deceptive Behaviors in Multimodal Large Language Models (ICML 2026 · similarity 0.87)
- LOKI: A Comprehensive Synthetic Data Detection Benchmark using Large Multimodal Models (ICLR 2025 · similarity 0.87)
- Unveiling Implicit Deceptive Patterns in Multi-Modal Fake News via Neuro-Symbolic Reasoning (AAAI 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
