GPT Image 2 是什么?
GPT Image 2 is OpenAI's current image generation model — the direct successor to DALL-E 3, which OpenAI is retiring from the API on May 12, 2026 after already moving ChatGPT itself onto the newer model in December 2025. Most people use it the same way they used DALL-E: typed inside a ChatGPT conversation, where you describe an image and the model generates and can iteratively edit it in the same chat thread.
Compared to DALL-E, GPT Image models handle text rendering inside images, multi-turn edits (asking for a specific change to a previous generation rather than starting over), and more literal prompt-following on complex, multi-object scenes. It's built for anyone already inside the ChatGPT ecosystem — marketers mocking up creative, writers illustrating a blog post, developers prototyping UI concepts — rather than as a dedicated standalone image app like Midjourney.
Access is free with usage limits inside ChatGPT's free tier, with higher limits on Plus and Pro subscriptions ($20/mo and up). Developers integrating image generation directly into their own products use the API, billed per image from roughly $0.009 up to $0.20 depending on resolution and quality tier.
GPT Image 2 的核心功能
- In-chat generation: Generate images directly inside a ChatGPT conversation without switching tools.
- Multi-turn editing: Ask for a specific change to a previous image and get an updated version instead of a fresh regeneration.
- Text rendering: Generates legible text within images — signage, labels, UI mockups — more reliably than earlier models.
- API access: Integrate the same model into a third-party product, billed per image.
GPT Image 2 的使用场景
- Illustrate a blog post or social post without leaving a ChatGPT conversation
- Mock up marketing creative or ad concepts quickly
- Prototype a UI or product concept visually
- Generate images with legible on-image text, like signage or labels
- Iteratively refine a generated image through follow-up instructions