Omnifake98K: A Benchmark for Human and AI-Generated Multimodal Misinformation

Approximately 98,000 image-text pairs form a benchmark for detecting fake content. The dataset, created by hy1228, integrates both human-crafted misinformation and AI-generated manipulations under realistic social media settings. It was last updated on July 16, 2026.

Use Cases

• Training classifiers to detect AI-synthesized images based on the integrated AI-generated manipulations.
• Benchmarking unified fake content detection models using the multimodal image-text pairs.
• Studying the characteristics of human-crafted misinformation in social media contexts.
• Developing models to distinguish between human and AI-originated fake content.

Strengths

• Approximately 98,000 image-text pairs provide a substantial scale for model training and evaluation.
• Integrates both human-crafted and AI-generated fake content within a single benchmark for unified detection tasks.
• Designed under realistic social media settings, which may increase practical relevance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface user hy1228

Collection Method

Likely compiled for benchmarking purposes, integrating various sources of misinformation.

Freshness

Last updated 2026-07-16 09:06:01; freshness should be verified.