Reddit user bypasses Ideogram 4 safety filters using local LLM

User modified Ideogram 4 workflow to use local Gemma-4-31B via llama.cpp, getting 0 rejections on 50 prompts. The workaround avoids Ideogram's built-in LLM-based safety checks.