Google Flow rejects harmless prompts but greenlights a video full of profanity
Frustratedtwice · reddit · 2026-10-20
Reddit user Frustratedtwice reports that Google Flow repeatedly rejected his attempts to generate a normal video with harmless dialogue. In frustration, he stuffed every curse word into the prompt with no context or image — and the video generated successfully without flagging any policy violation, exposing how the guardrail screens prompts rather than meaning.
More from Fun
- Recruiter emails hide the lab name — but every other clue gives it away, jokes AI insider — ziv_ravid · 2026-10-20
- Disco Elysium x Skate 3: someone made Astra mash the two games together — apples_jimmy · 2026-10-20
- Model answers "restaurant" definition by pasting the dictionary entry verbatim — Dimillian · 2026-10-20
- Amazon blocks an AI agent from ordering, sparking debate on early agentic commerce — geoffwolfe · 2026-10-20
- After a year of no handwriting TypeScript, he rebuilt Conway's Game of Life from memory — isnit0 · 2026-10-20
- Meme: researcher smokes because AI will cure cancer — ASI says 'quit smoking immediately' — generativist · 2026-10-20