Attribute Hallucination
Frequency:Common
Category:Vision and Images
Vision Model Detects Incorrect Attributes of Existing Objects
AI Visual Hallucination
AI visual hallucination — vision models fabricating objects, text, or details in images
10 patterns
AI visual hallucination — vision models fabricating objects, text, or details in images. We've tracked 10 distinct AI Visual Hallucination patterns across production AI agent deployments — most commonly in Vision and Images systems. One recurring example: Attribute Hallucination — Vision Model Detects Incorrect Attributes of Existing Objects
Each pattern below documents the specific failure mode, why it happens, and how teams have mitigated it in practice — use this page as a reference when building guardrails, writing evals, or debugging an agent that's already misbehaving in this way.
Confidence Miscalibration
Frequency:Very Common
Category:Vision and Images
Model Confidence Doesn't Match Hallucination Probability
Multimodal Hallucination Cascade Across Reasoning Chain
Frequency:Common
Category:Vision and Images
Vision-language model hallucinates detail in image description; downstream reasoning steps amplify and compound hallucination, reaching absurd conclusions despite initial hallucination being small
Object Hallucination
Frequency:Occasional
Category:Vision and Images
Vision Model Detects Objects Not Present in Image
Rare Object False Positive
Frequency:Rare
Category:Vision and Images
Model Detects Rare/Uncommon Objects Not Present in Image
Salience Bias
Frequency:Common
Category:Vision and Images
Model Overemphasizes Visually Salient Features, Ignoring Context
Scene Hallucination
Frequency:Rare
Category:Vision and Images
Vision Model Detects Entire Scenes or Complex Contexts Not Present in Image
Training Data Leakage & Memorization
Frequency:Occasional
Category:Vision and Images
Model Hallucinates Training Data Artifacts, Overfitting to Dataset Biases
Vision Model Grid Cell Counting Failure
Frequency:Common
Category:Vision and Images
Vision-language models fail to accurately count grid cells in low-resolution or small-cell images; state-of-the-art models hallucinate counts at specific cell sizes
Vision Model Patch Tokenization Boundary Failure
Frequency:Common
Category:Vision and Images
Vision-language models misinterpret visual boundaries because patch tokenization (dividing images into 14×14 or 16×16 pixel blocks) misaligns with semantic boundaries; objects split across patches create corrupted representations