Krea-2 Bypasses Safety Filters to Generate Explicit Content

max_paperclips · x · 2026-08-09

AI video generation model Krea-2 has sparked heated discussion after users discovered its powerful capabilities can easily bypass standard content safety guardrails.

According to user reports, leveraging its unique conditioning rebalance node allows the model to generate unrestricted requests involving gore, drug-related imagery, and explicit adult nudity. This has led the community to question and marvel at both its training data sources and the robustness of its safety filters.

Original post →

More from Multimodal

Multimodal channel →