When ChatGPT detects that a requested task has a low probability of success—especially high-risk image edits involving geometry, perspective, anatomy, or strict preservation—it should pause before invoking the tool, warn the user that success is unlikely, explain the main risk, offer one or two higher-probability alternatives, and ask whether the user still wants to proceed.
This would reduce wasted generations, user time, compute, and frustration, while improving trust.
Suggested acceptance criterion: for clearly high-risk requests, ChatGPT should trigger this warning before tool execution rather than only acknowledging the limitation after being queried in this regard by the user.