GPT Image 2.5
OpenAI
這裡唯一能在直式構圖下真的輸出 4K 的模型——2881×3840,不是放大來的。 OpenAI's newest image model. Best-in-class legible text, and the only one that renders true 4K in portrait ratios. Cannot take pixel sizes or masks — use GPT Image 2 for those.
Using GPT Image 2.5 in uooki
uooki has no model picker. You name the model in the conversation and the agent routes to it — “draw the packaging mockup with GPT Image 2.5” is the whole interface. Common spellings and older ids resolve to the same model, so you do not have to match our internal naming.
Leave the model out and the agent picks one for you based on what you asked for. Naming it is for when you have a reason to — this page is about those reasons.
What it is good at
- 直式海報和封面——放到滿尺寸看,字也得是清楚的
- 要拿去印刷、或在手機上滿版呈現的圖
- 圖裡的文字要讀得出來——和 GPT Image 2 一樣強,單張還更便宜
Where it is not the right pick
它不吃像素尺寸,只認具名比例;也沒有遮罩欄位,做不了局部重繪。這兩件事仍舊留給 GPT Image 2。它只有一條供應商線路,所以「非出圖不可」的場合別押它。
When to reach for something else
- GPT Image 2
你要做遮罩局部重繪,或傳的是精確像素尺寸而不是比例。
- Nano Banana
同一個主體要貫穿一整系列圖片。
Specifications
| Type | Image generation |
|---|---|
| Billing | Billed per generated image. |
| Plan | Available on every plan |
| Reference images | Up to 5 |
| Text in image | Renders legible text |
| Credits | See pricing |
Specifications are read from uooki’s model registry, not written by hand — if the registry changes, this table changes with it.
Model id
What uooki calls upstream when you pick GPT Image 2.5. You never need this to use it — it is here so the vendor name and the actual model are unambiguous.
Try it in a sentence.
Describe what you want and name GPT Image 2.5 in the same breath. The Free plan is $0 and starts with 100 one-time credits.