False or misleading visual content including photographs, memes, screenshots, videos, and other images used to spread misinformation. This topic covers detection methods, typologies of visual misinformation (out-of-context images, photoshopped content, deepfakes, false captions), and platform-specific dynamics of visual content spread.
Taxonomy
Visual misinformation broadly falls into three categories:
- Out-of-context images ("cheap fakes" or "shallow fakes") — real images presented without proper context or with misleading captions; often old content reshared during new events; low production cost but effective due to realism and plausibility.
- Manipulated images — images altered through photoshop, color adjustment, cropping, or splicing; can range from benign humor to convincing fabrications (e.g., fake news screenshots).
- Synthetic media — completely AI-generated or deepfake video/images that create false depictions of people or events.
Key videos
- nodate-garimella-misinformation-whatsapp — Characterizes image-based misinformation on WhatsApp: 13% of shared images are misinformation across three categories: out-of-context images (40%), false quotes/statistics overlaid on images (30%), and photoshopped content (30%). Shows misinformation images persist far longer than other content (50% last over 100 days). Discusses detection and mitigation via perceptual hashing, machine learning, on-device fact-checking, and community-driven correction.
Related topics
- cheapfakes-synthetic-media — out-of-context and manipulated media
- deepfakes — AI-generated synthetic video
- multimodal-detection — methods for detecting misinformation across text and images
- social-media-analysis — how visual content spreads on platforms
