GPT-IMAGE-2 drops transparent image support

This is a meaningful regression for asset-generation workflows, even if GPT-IMAGE-2 improves resolution, fidelity, and editing elsewhere.

• –Developers building sticker, icon, game-asset, ecommerce, and design-tool pipelines now need post-processing or an older GPT Image model for alpha-channel output
• –The confusing part is UX mismatch: users can ask for transparency, but the new model's supported output controls explicitly exclude it
• –GPT-IMAGE-2 still accepts alpha-channel masks for edits, so OpenAI supports transparency as input guidance while dropping transparent final outputs
• –For API users, this is the kind of model migration detail that should be feature-gated, not discovered through failed generations