GPT Image 2
即將推出GPT Image 2 is OpenAI's upcoming next-generation image model. Preview the announcement, browse features, and generate with a comparable model today while GPT Image 2 is getting ready.
本頁面為產品預告。該模型尚未開放——目前在本平台生成由一個能力相近的模型承擔，本站與原模型提供方無官方關聯。
Real-World Knowledge
Early demos suggest GPT Image 2 renders real brands, interfaces, and locations with striking accuracy, pulling on GPT's world knowledge.
Natural Text Rendering
Readable, scene-integrated text — handwritten notes, speech bubbles, signage, posters — in a way previous diffusion models struggled with.
Neutral Color Fidelity
Early reports say GPT Image 2 has dropped the warm yellow cast that became a signature flaw of GPT-Image-1.
Deep Prompt Understanding
GPT Image 2 is expected to compose complex scenes from long prompts without dropping details — the kind of multi-clause descriptions that used to require manual touch-ups. Until release, these previews are produced with Nano Banana Pro, which is closest in capability today.
提示詞
Editorial photo: a chef plating a seven-course tasting menu at a stainless steel pass, low-key lighting, dramatic steam, legible handwritten tickets pinned behind
Preview (Nano Banana Pro)
Preview (Nano Banana Pro)
Integrated Typography
Text that belongs in the scene — signage, speech bubbles, handwritten notes — rendered with correct letterforms and proper placement. This has historically been the hardest single capability for diffusion image models.
提示詞
Subway poster for a fictional film titled 'Night Shift', bold grotesque sans-serif, tagline 'The city never sleeps. Neither did we.', clean minimal layout
Preview (Nano Banana Pro)
Preview (Nano Banana Pro)
Physical Lighting
Reflections, shadows, and material response that read as physically plausible. Golden hour, neon environments, glass refraction — the edge cases that expose older models.
提示詞
Vintage motorcycle parked in an empty parking garage lit only by a single sodium vapor lamp, long orange shadows, wet concrete reflections
Preview (Nano Banana Pro)
使用方法
- 1
Describe your scene
Write a concrete, detail-rich prompt. Coming-soon features aside, this is the biggest lever for image quality.
- 2
Tap Try in Workspace
Your prompt opens in Workspace. While GPT Image 2 is unreleased, generation routes to Nano Banana Pro — you'll see a banner confirming the substitution.
- 3
Pick resolution
Choose 1K for thumbnails, 2K for hero use, 4K for print or zoomable detail.
- 4
Generate and iterate
Refine the prompt, swap references, tweak aspect ratio. When GPT Image 2 launches, your flow stays the same — we just flip the backend.
Be first to try GPT Image 2
GPT Image 2 is not yet available. Tap below to open Workspace and generate with Nano Banana Pro in the meantime — we'll flip the backend the moment OpenAI releases the model.
Try in Workspace
