> Index of Phersa pages for AI agents: https://phersa.com/llms.txt

# AI Image Prompt Generator

Build an AI image prompt for a photo of your AI persona: who they are, where they are, what they wear and do, the light, the camera and the framing. It is for anyone who runs an AI influencer page and wants photos that look taken on a phone. Each prompt is written the way Nano Banana 2, GPT Image or Grok Imagine reads best.

## How to use the AI Image Prompt Generator

1. Describe your persona or pick a type. Choose a setting, outfit, pose, light, camera, framing and aspect ratio, or leave any on Auto.

2. Pick your image model. Press Generate prompt for new details with the same picks, or Randomize for a new photo idea.

3. Copy the prompt as text or JSON and paste it into your image tool. If the reference line is on, attach the persona’s photo.

## How to write an image prompt for an AI persona

A good persona prompt describes one photo, block by block. Each block answers one question, so you can change one thing and keep the rest.

- **Subject.** Age, skin, hair, eyes, face and build in plain words, plus who they are: a home cook, a gardener, a mechanic (more in these [AI persona ideas](https://phersa.com/personas)).
- **Action and expression.** What the hands do and what the face does, at one moment: mid-sentence, mid-stir, glancing up.
- **Setting.** The place, what is behind them and two or three real objects. Lived-in rooms read as real life.
- **Wardrobe.** Named garments and colors, “a cream cable-knit sweater”, never just “a sweater”.
- **Light.** Where it comes from, its color and its shadows.
- **Camera and framing.** A phone, how it is held, the angle and what the frame cuts at. The background stays in focus, the way a phone photographs a room.
- **Realism.** Pores and fine lines, single hair strands with a few flyaways, creased fabric, color straight from the phone.

Describe only what is in the frame, in positive words. Google’s guide for its image models puts it this way: ask for “an empty street”, not “no cars”. Naming a thing you don’t want can put it in the picture.

When you attach the persona’s photo, the prompt opens with the identity line _Use reference image. Maintain EXACT persona._ It then names the traits to keep from the photo instead of describing a new face.

## Nano Banana 2, GPT Image and Grok Imagine: what changes

| Model | How the prompt is written | With the persona’s photo |
| --- | --- | --- |
| Nano Banana 2 | A narrative description, not a keyword list. Google’s formula: subject, action, location, composition, style. It starts with a verb: “Create…”. | Reference, then how to use it, then the new photo. Google lists up to 14 reference images for this model, with up to 4 of them of people. |
| GPT Image | Short labeled lines in a fixed order: scene, subject, details, then what to keep. The word “photorealistic” and words like “candid” and “unposed” set the look. | A keep list for the face and the rest. OpenAI suggests repeating that list on every edit so the person drifts less. |
| Grok Imagine | One detailed description of the photo. | xAI’s docs describe edits as a source image plus the change you want, in plain words. |

The JSON view holds the same picks as labeled blocks. Paste whichever your tool takes, and set the aspect ratio in the tool as well as in the prompt.

Turn the image into a clip of your AI persona: make the image from the persona’s photo, then the video from that image, on your own API keys. [How making videos works →](https://phersa.com/docs/create)

## AI Image Prompt Generator questions

### Which AI image generator should I use these prompts with?

Any model that takes a text prompt. The text view is written for the model you pick: Nano Banana 2, GPT Image or Grok Imagine. For an AI persona, use a model that also takes a reference image, so the face stays the same.

### What aspect ratio should an AI persona photo be?

9:16 when the image will be the first frame of a Reel or a Story. For a feed photo, 3:4, 4:5 or 1:1 fit Instagram, and Meta recommends 4:5 for Facebook Feed images.

### Should I paste the text or the JSON?

Either. Both describe the same photo. The text is shaped for the model you picked. The JSON puts each block on its own line, so one field is easy to change and save.

### Can I use negative prompts with Nano Banana?

Better not. Google’s prompting guide for its image models says to describe what you want, such as “an empty street”, instead of what to leave out, such as “no cars”. Every line here is written that way.

## More free tools

- [Consistent Character Prompt Generator — One identity block, many shots: image prompts that repeat the same face, hair and look word for word.](https://phersa.com/tools/consistent-character-prompt-generator)
- [Realistic iPhone Photo Prompt Generator — Phone-photo looks for AI images: selfie, 0.5x, mirror, flash at night, golden hour, car, café and more.](https://phersa.com/tools/iphone-photo-prompt-generator)
- [AI Persona Generator — A full AI persona: name, niche angle, content pillars, bio, voice and an image prompt for the photo.](https://phersa.com/tools/ai-persona-generator)

[See all free tools for AI persona creators →](https://phersa.com/tools)

Source: https://phersa.com/tools/ai-image-prompt-generator