Can GPT Image 2 Generate People?
Probably yes, but with policy limits. Here is the clearest current read on whether GPT Image 2 will generate human subjects and what restrictions to expect.
Can GPT Image 2 Generate People?
TL;DR
Probably yes. OpenAI's current image-generation system already deals with people-related use cases, and the company's public materials discuss safeguards around depictions of people rather than banning the category outright. That suggests GPT Image 2 will likely generate fictional or generic people, portraits, and human scenes. The limits matter, though: safety rules, identity misuse rules, and policy enforcement will almost certainly shape what kinds of people-related images are allowed.
| Question | Best answer now |
|---|---|
| Can it likely generate people? | Yes |
| Any restrictions? | Definitely |
| Safe assumption for business use? | Fictional people first |
Why “people” is different from “anything”
OpenAI's usage policies already restrict harmful uses of someone's likeness, while the native image-generation system card addendum discusses people and public figures as an active safety area. That means the category exists, but it is governed.
| People-related use case | Likely outlook |
|---|---|
| Generic fictional person | Likely allowed |
| Stylized portrait | Likely allowed |
| Sensitive or deceptive person image | More restricted |
My expectation
GPT Image 2 should be able to generate people because that is a core commercial use case for ads, education, product mockups, and storytelling. The safer assumption is that ordinary fictional people are easier than real identifiable people.
For narrower policy questions, read Can GPT Image 2 Generate Celebrities? and Can GPT Image 2 Generate NSFW?. If your concern is the guardrail layer, start with Will GPT Image 2 Have Safety Filters?.
What this means in practice
For most teams, the useful working assumption is that GPT Image 2 will be fine for fictional models, lifestyle scenes, and generic portrait work, but less permissive around real identifiable people and sensitive scenarios. That is enough for many commercial workflows, and it is a more realistic expectation than assuming either a total ban or total freedom.
In other words, the likely answer is not “yes to everything.” It is “yes to many mainstream use cases, with tighter rules once identity risk rises.”
That is still enough to make people-generation one of the more likely core GPT Image 2 use cases.
If you are briefing clients or stakeholders, frame the capability around safe commercial scenes rather than unrestricted likeness generation. That is the most defensible expectation today.
It is also the expectation least likely to be invalidated by launch-day policy details.
FAQ
Does “generate people” mean photorealistic people too?
Possibly, yes. But realism does not remove policy restrictions.
Are real-person likenesses treated differently?
Yes, very likely. OpenAI's current policy materials already treat likeness misuse as a distinct risk area.
Is this officially confirmed for GPT Image 2?
No. It is the strongest inference from OpenAI's current image-policy and model direction.
Sources
- OpenAI usage policies
- OpenAI introducing 4o image generation
- OpenAI native image generation system card addendum
GPT Image Countdown is not affiliated with OpenAI. All trademarks belong to their respective owners.