# ChatGPT AI influencer photos: real looks, same face

ChatGPT AI influencer photos that look real and keep one face: which reference images to upload, prompts that work, the cost per image and what OpenAI forbids.

- Canonical: https://aiofmcity.com/blog/chatgpt-ai-influencer-images
- Topic: Character consistency (https://aiofmcity.com/topics/character-consistency)
- Author: Gus
- Published: 2026-10-06
- Updated: 2026-10-28

---
## Key points

- ChatGPT Images 2.5 is available on every ChatGPT plan. In the API, OpenAI's current image models are GPT Image 2.5 Sunburst and GPT Image 2.5 Flare.
- The OpenAI API takes up to 16 reference images per edit, but OpenAI publishes no number of images that keeps a face most stable.
- The Images API has no seed setting, so the same prompt does not give the same face twice. Your reference images keep the face, not the prompt.
- On WaveSpeed, a GPT Image 2.5 edit at 1K medium quality costs $0.039 with one reference image and $0.015 for each extra one.
- OpenAI's terms forbid presenting its images as made by a human, so a ChatGPT persona has to be labelled as AI.

---
To make ChatGPT AI influencer photos with the same face, upload four to six reference images of your own generated persona in one message. Say which image shows what, list the features that must not change, and then describe one scene with camera, light and skin detail. Change details later with "change only" edits, and check every image against your references.

ChatGPT's image model is very good at photos that look real. Below: the model you get, the reference set, prompts to copy, the cost per image and the limits.

## Which GPT image model ChatGPT uses in October 2026

ChatGPT runs ChatGPT Images 2.5 on every plan. In the API, OpenAI's current image models are GPT Image 2.5 Sunburst and GPT Image 2.5 Flare.

> "ChatGPT Images 2.5 brings sharper details, more precise editing, and faster image generation." "ChatGPT Images is available on all tiers. Images with thinking is available on Plus, Pro, and Business."
>
> — Source: OpenAI Help — Images in ChatGPT, read 2026-10-04 — https://help.openai.com/en/articles/11084440-images-in-chatgpt

> "GPT Image 2.5 Flare is the small model, optimized for speed, with image quality comparable to GPT Image 2. GPT Image 2.5 Sunburst is the base model, optimized for quality, with higher image quality than GPT Image 2. Both models offer improvements in precise editing and subject preservation."
>
> — Source: OpenAI API docs — Image prompting guide, read 2026-10-04 — https://developers.openai.com/api/docs/guides/image-prompting

For a persona, "subject preservation" is the line that matters. It means the model keeps the person from your reference images closer to the original.

Use the ChatGPT app for a few images a day by hand. Use the API, or a host that resells it, for batches. To run those batches from a coding agent with your persona's references and a price check, see [Claude Code image generation for an AI persona](https://aiofmcity.com/blog/claude-code-image-generation).

## Which reference images to upload for a consistent character

Upload four to six images of your persona that each show something the others do not. The API accepts up to 16 images per request, but every extra image costs money.

> Create image edit, `images`: "Input image references to edit. For GPT image models, you can provide up to 16 images."
>
> — Source: OpenAI API reference — Images, read 2026-10-04 — https://developers.openai.com/api/reference/resources/images

Your persona must already exist as generated images. If you are starting from zero, generate her first and pick your best images as a [reference sheet](https://aiofmcity.com/glossary#reference-sheet). Never use photos of a real person. The rules section below explains why.

A good set for ChatGPT:

1. **Front view**, head and shoulders, soft even daylight.
2. **Three-quarter view from the left.**
3. **Three-quarter view from the right.**
4. **Profile or a tilted head**, so the model sees the nose and jaw from the side.
5. **A full-body image**, so her height and proportions stay the same.
6. **One image in different light**, such as warm indoor light, so the face is not tied to one lighting setup.

Some rules for the set:

- **Change the outfit at least once.** If she wears the same top in every image, the model treats the top as part of her.
- **One person per image.** Crop out anyone in the background or in a mirror.
- **Keep the set fixed.** Save the files in one folder and upload the same ones every time.

The same set works for video: [MiniMax H3](https://aiofmcity.com/blog/minimax-h3) takes up to nine reference images of your persona in one clip.

OpenAI's own advice is to number your inputs and give each one a job:

> "Assign roles to references. Identify each input by number and purpose: subject, style, clothing, or background. Explain how the inputs should combine and which elements should move where."
>
> — Source: OpenAI API docs — Image prompting guide, read 2026-10-04 — https://developers.openai.com/api/docs/guides/image-prompting

So say "Images 1 to 5 show the same woman. Use them only for her identity." If you add a clothing image, say "Image 6 is the jacket she wears."

## Prompts for realistic ChatGPT AI influencer photos

Describe the photo the way a photographer would: who, what she does, where the camera is, what the light does, and how the skin looks. Ask for "photorealistic" or "real photograph" out loud, and say what you do not want.

> "Request 'photorealistic' or 'real photograph' explicitly when that is the goal, and describe framing and texture. Treat camera specifications as cues for appearance, not a guarantee of exact physical simulation." For edits: "say 'change only X' and list the details to preserve, such as identity, geometry, layout, lighting, or labels."
>
> — Source: OpenAI API docs — Image prompting guide, read 2026-10-04 — https://developers.openai.com/api/docs/guides/image-prompting

Start every request with the same character block. Write it once and paste it in front of every scene:

```text
Images 1 to 5 are reference photos of the same woman, Mara, in her late twenties.
Use them only for her identity.
Keep exactly: face shape, eye colour and spacing, nose, lips, jawline, skin tone,
light freckles across the nose, shoulder-length dark brown hair with a middle parting.
Do not beautify, slim, smooth or age her. Do not change her proportions.
```

Then add the scene:

```text
Photorealistic candid smartphone photo of the woman from images 1 to 5.
She sits at a small café table by a window, holds a coffee cup with both hands
and laughs, looking slightly past the camera.
Phone main camera at eye level, about 1.5 metres away, vertical 4:5 framing.
Soft overcast daylight from the window on the left, natural white balance.
Real skin texture with visible pores and a few small blemishes, a few flyaway hairs,
light sensor noise. No retouching, no studio lighting, no text, no watermarks, no logos.
```

These details make the biggest difference to realism:

| Write this | Instead of | Why it helps |
| --- | --- | --- |
| "smartphone photo", "35mm film photo, 50mm lens" | "high quality", "8k" | A camera gives the model a look to copy. Quality words do not. |
| "soft overcast daylight from the left" | "beautiful lighting" | The light direction decides shadows on the face. |
| "visible pores, small blemishes, flyaway hairs" | "flawless skin" | Perfectly smooth skin makes a photo look generated. |
| "no retouching, no heavy colour grading" | nothing | OpenAI's own photo example asks for this to keep the image natural. |
| "a woman in her late twenties" | no age | It keeps her clearly adult in every image. |

Keep every prompt strictly SFW. OpenAI filters all prompts and images, and a blocked request costs you time.

## How to keep the same face across a ChatGPT image series

Give the model your fixed reference set for every new image, and repeat the character block word for word. Do not build image 20 from image 19.

OpenAI is open about the weak point:

> Limitations: "Consistency: While capable of producing consistent imagery, the model may occasionally struggle to maintain visual consistency for recurring characters or brand elements across multiple generations."
>
> — Source: OpenAI API docs — Image generation guide, read 2026-10-21 — https://developers.openai.com/api/docs/guides/image-generation

> "Repeated edits can still change details you intended to preserve. Restate those constraints and inspect each result. If a region must remain pixel-identical, composite the approved edit into the original image instead of relying on prompting alone."
>
> — Source: OpenAI API docs — Image prompting guide, read 2026-10-04 — https://developers.openai.com/api/docs/guides/image-prompting

If you always start from the last output, small changes add up over a series. If you always start from the same reference files, every image is measured against the same face.

A routine that works:

1. **Open a new chat for each batch** and upload the full reference set first. In a long chat, older images stay in the conversation and the model can draw on them.
2. **Paste the character block** and one scene. One scene per request.
3. **Compare the result with image 1** before you keep it. Look at the eye spacing, the nose and the jaw. If they moved, throw the image away. Do not try to fix a wrong face.
4. **Keep a monthly grid** of the same six scenes, as the [character consistency guide](https://aiofmcity.com/guides/character-consistency#the-ten-minute-check-that-catches-drift-early) explains. Drift shows up across months.

### What the sources do not say: Finding a published way to make the same GPT Image prompt return the same face

We looked for two things: how many reference images keep a face most stable, and how to make the same prompt return the same face. OpenAI publishes neither.

The Images API reference lists every setting for creating and editing images, and none of them is a seed. So you cannot fix the randomness as you can in ComfyUI. The image generation guide gives the maximum of 16 input images, but it does not say whether 3 or 10 images of one face work better. Its Limitations section says the model "may occasionally struggle" with recurring characters, without a number.

You can check this yourself. Open OpenAI's Images API reference and search the page for "seed". Then open the image generation guide and the image prompting guide and search for "reference" and "consistent". Every number you find is a maximum or a price.

## How to change one detail without changing the face

Select the area in the ChatGPT editor, or write "change only" in the chat, and list what must stay.

> "The editor lets you add, remove, and update parts of your image. Highlights are not always precise, and edits may extend beyond the area you selected."
>
> — Source: OpenAI Help — Images in ChatGPT, read 2026-10-04 — https://help.openai.com/en/articles/11084440-images-in-chatgpt

An edit prompt that keeps the face safe:

```text
Change only her jumper to a navy cable-knit jumper.
Keep her face, hair, expression, pose, the café, the light and the framing exactly as they are.
Do not add text, logos or watermarks.
```

Use edits for small things: an outfit, a background object, a hand, a colour. For a new pose or place, generate a new image from your reference set. After every edit, compare the face with image 1 again.

## What ChatGPT and the GPT Image API cost per image

In the ChatGPT app you pay for the plan, not per image. Through the API, a medium-quality portrait image costs a few cents, and each reference image adds to it.

> "ChatGPT Plus is a subscription plan that provides enhanced access to the ChatGPT web app for $20/month." Plus subscriptions "may include usage limits such as message caps, especially during high demand." API usage is billed separately.
>
> — Source: OpenAI Help — What is ChatGPT Plus?, read 2026-10-04 — https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

> Free users can create images in ChatGPT. Image creation "can have stricter limits than paid plans."
>
> — Source: OpenAI Help — ChatGPT Free Tier FAQ, read 2026-10-04 — https://help.openai.com/en/articles/9275245-chatgpt-free-tier-faq

OpenAI publishes no image count per plan. ChatGPT tells you when you hit the limit.

> GPT Image 2 per image, 1024 x 1536: low $0.005, medium $0.041, high $0.165. "Because `gpt-image-2` always processes image inputs at high fidelity, edit requests that include reference images can use more input tokens." Both GPT Image 2.5 models use the same token rates as GPT Image 2: $8 per million image input tokens and $30 per million image output tokens.
>
> — Source: OpenAI API docs — Image generation guide, read 2026-10-21 — https://developers.openai.com/api/docs/guides/image-generation

OpenAI gives no per-image table for GPT Image 2.5, only token rates. A host shows the price per image:

> $0.039 per run. "Prices below include one input image. Each additional input image adds $0.015." At 1K: low $0.025, medium $0.039, high $0.105, xhigh $0.175, max $0.375. At 2K, medium is $0.055. Up to 16 input images.
>
> — Source: WaveSpeedAI — GPT Image 2.5 Sunburst Edit, read 2026-10-21 — https://wavespeed.ai/models/openai/gpt-image-2.5-sunburst/edit

> The same price table as Sunburst Edit: $0.039 per run at 1K medium, and $0.015 for each additional input image.
>
> — Source: WaveSpeedAI — GPT Image 2.5 Flare Edit, read 2026-10-21 — https://wavespeed.ai/models/openai/gpt-image-2.5-flare/edit

Here is what one image of your persona costs with a reference set, worked out from WaveSpeed's published prices at 1K medium quality:

| Reference images | GPT Image 2.5 Edit (Sunburst or Flare) |
| --- | --- |
| 1 | $0.039 |
| 4 | $0.084 |
| 6 | $0.114 |
| 16 (the maximum) | $0.264 |

With four or more references, most of the price is the references, not the output. So test whether four references do the same job as six for your face.

Compared with the other image-edit models on [what generation costs](https://aiofmcity.com/model-prices), as checked there on 29 September 2026: Seedream V5.0 Pro Edit costs $0.0405 per generation, GPT Image 2 Edit $0.0665, Nano Banana 2 $0.07 and Nano Banana Pro $0.14 per image. GPT Image 2.5 with one reference sits at the cheap end. With six references it comes close to Nano Banana Pro's $0.14. [What an AI persona costs before it earns](https://aiofmcity.com/blog/what-an-ai-persona-costs-to-start) puts these image prices into a full monthly budget.

## What OpenAI does not allow: real people, fake humans, missing labels

OpenAI allows you to build an invented persona. Its rules forbid three things that matter here: using a real person's face without consent, deceiving people, and claiming the images were made by a human.

> Under "Respect privacy", you may never use OpenAI's services for "use of someone's likeness, including their photorealistic image or voice, without their consent in ways that could confuse authenticity". Under "Empower people", the services may not be used for "deceit, fraud, scams, spam, or impersonation". Effective 29 October 2025.
>
> — Source: OpenAI — Usage policies, read 2026-10-04 — https://openai.com/policies/usage-policies/

> "You may not use Visual Capabilities to reproduce the likeness of any person without express consent and all necessary rights."
>
> — Source: OpenAI — Service terms, read 2026-10-04 — https://openai.com/policies/service-terms/

> Among the things you are prohibited from: "Representing that Output was human-generated when it was not." The same terms say output "may not be unique and other users may receive similar output from our Services."
>
> — Source: OpenAI — Europe Terms of Use, read 2026-10-04 — https://openai.com/policies/eu-terms-of-use/

For your persona this means:

- **Only your own generated persona goes in as a reference.** Not a model from Instagram, not a celebrity, not a friend.
- **Say it is AI.** A bio that presents her as a real woman breaks OpenAI's terms, as well as the rules of most platforms. The [label matrix](https://aiofmcity.com/ai-label-matrix) shows what each platform wants.
- **Do not count on a unique face.** OpenAI says another user can get similar output.

Every image also carries markers that say where it came from:

> "Supported images generated with ChatGPT, Codex, and the OpenAI API include both signals": C2PA metadata and SynthID watermarks. C2PA metadata "can sometimes be removed by platforms, editing tools, or file conversions." Anyone can check a file at openai.com/verify.
>
> — Source: OpenAI Help — Provenance signals in OpenAI-generated content, read 2026-10-04 — https://help.openai.com/en/articles/8912793-c2pa-and-synthid-in-openai-generated-images

> Article 50(2): providers of AI systems generating synthetic image content "shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated." Article 50(4): "Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated."
>
> — Source: EU AI Act Art. 50(1), read 2026-10-04 — https://artificialintelligenceact.eu/article/50/

The marking duty is OpenAI's. Your duty in the EU is the visible label. The AI Act does not forbid you to remove the C2PA data, but keep it anyway. Instagram and other platforms read it to label AI images. [Don't strip your watermarks](https://aiofmcity.com/blog/dont-strip-your-watermarks-provenance) shows how to check a file before you post. None of this is legal advice.

## Where GPT image generation falls short

GPT Image is strong at real-looking light and skin. It is weaker at an exact repeat of the face, large files, and control you own.

> Custom sizes: "neither edge may exceed 3840 pixels" and "Resolutions above `2560x1440` are experimental." Limitations: "Complex prompts may take up to 2 minutes to process." "All prompts and generated images are filtered in accordance with our content policy."
>
> — Source: OpenAI API docs — Image generation guide, read 2026-10-21 — https://developers.openai.com/api/docs/guides/image-generation

- **No seed and no LoRA.** If the face drifts, your only tools are the reference set and the prompt. A trained LoRA, covered in the [character consistency guide](https://aiofmcity.com/guides/character-consistency), keeps the face in a file you own.
- **Resolution.** Plan for 2560 x 1440 or less. OpenAI does not publish the output size of the ChatGPT app.
- **Refusals.** If a harmless prompt is blocked, describe the scene in plainer words.
- **New versions.** ChatGPT switches when OpenAI releases a new version. In the API you can pin a dated snapshot, such as gpt-image-2.5-sunburst-2026-09-08.

Start today with the reference set. Pick your five best images of your persona, write the character block, and render the café scene from this guide ten times. If eight of the ten look like her next to image 1, ChatGPT is a good main tool for your account. If fewer do, use it for edits and outfits, and keep your main face in a LoRA.

## Frequently asked

### Can ChatGPT make consistent characters?

Yes, if you give it reference images of the character every time. Upload the same fixed set of images, say what must stay the same, and change only the scene. OpenAI itself says the model can still struggle to keep a recurring character the same across many images, so check every result against your references.

### Is it allowed to make an AI influencer with ChatGPT?

Yes, for a person you invented, if you say it is AI. OpenAI's terms forbid presenting its output as made by a human, and its usage policies forbid using someone's likeness without consent in ways that could confuse people about what is real.

### How many reference images can I give GPT Image?

Up to 16 per edit request through the OpenAI API. OpenAI does not publish a limit for the ChatGPT app, and it does not say how many images of one face give the most stable result. Four to six clear images of your persona is a sensible start.

### How much does ChatGPT image generation cost?

In the ChatGPT app it is part of your plan: the Free plan includes it with stricter limits, and Plus costs $20 a month. Through the API, OpenAI lists GPT Image 2 at $0.041 per medium-quality portrait image, plus the cost of your reference images.

### Do ChatGPT images have a watermark?

Not a visible one, unless you ask for it. OpenAI says images from ChatGPT and its API carry C2PA metadata and an invisible SynthID watermark. Keep both in the file, because platforms read them.