Gemini image generation complaints point to safety checks wasting tokens
The current experience is described as badly broken, especially for . The main complaint is that the does not understand meaning well enough, so normal creative requests can be blocked as unsafe.
Even a simple request such as making an image of a white cat on a bed may be rejected if the filter reacts too strongly. Another issue is the : a tiny is checked together with the full .
That can spend the user’s on background checks instead of the actual task. Customer support is also described as unhelpful, with automated replies and no clear troubleshooting or refund help.
Key points
- The may block ordinary image requests because it is too sensitive.
- A small may be checked with the full attached.
- That design can waste on work the user did not ask for directly.
- Agent builders should avoid sending full context to every minor step unless it is truly needed.