ChatGPT Images 2.0 tests IP limits

The real story is less “fan art works now” and more that frontier image models are pushing into legally messy production territory faster than policy messaging can keep up.

• –Reddit commenters already report uneven guardrails, with some anime-style characters accepted while franchises like Pokemon, Batman, Spider-Man, and Sesame Street trigger refusals.
• –For developers building creative tools on top of image models, inconsistent refusal behavior is a product risk: the same prompt class may pass or fail depending on character, brand, or likeness.
• –OpenAI’s official examples emphasize complex layouts, readable text, multilingual design, and market-ready assets, which makes IP and likeness handling more consequential than it was for toy image demos.
• –Treat this as anecdotal behavior, not a documented policy shift; teams still need their own copyright, trademark, and publicity-rights controls before shipping user-facing workflows.