A custom AI prompt tells the AI how to transform your guests' photos. You write it once, test it against your own photos, and then use it in as many events as you like.

Find it at **Custom AI prompts** in the side navigation.

### **The page**

| **Control**            | **What it's for**                                                    |
| ---------------------- | -------------------------------------------------------------------- |
| **Create new prompt**  | Opens the builder on a blank prompt                                  |
| **Request a prompt**   | Ask us to write one for you                                          |
| **Saved prompts**      | The prompts you have created                                         |
| **Generated previews** | Every test render you have made, across all your prompts — see below |

Each prompt in the list has an **Actions** menu holding **Duplicate** and **Delete**. To change a prompt, open it — the builder's title changes from *Custom AI prompts* to **Edit custom prompt**, so you can always tell which one you are in.

![Your saved prompts, with search and the two tabs.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCQy9Hd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--95171a5fd8f263744ba4f6f618c12e87dbf06ea6/32-ai-prompts-list.png?cw_image_width=auto)

*Your saved prompts, with search and the two tabs.*

### **How the builder is laid out**

The builder is two columns.

**On the left** you build the prompt: **Prompt** (name, thumbnail, prompt text) and **Generate** (test images and output settings).

**On the right** is **Generated previews** — every render you have made from this prompt, newest first with a **Latest** badge, filtered by **All**, **Photos** or **Videos**, with a thumbnail strip underneath.

**Save** sits in the top right. Generating does not save the prompt; saving does.

### **Name and thumbnail**

**Prompt name** — what you and your colleagues will search for.

**Thumbnail image** — what guests see in the picker. **Select thumbnail** picks one from your Asset Library, or you can promote any render to be the thumbnail later.

### **Write the prompt**

Type your description into **Prompt text**. Underneath, the field tells you how to add tokens:

> Type **@** for reference images or **{{** for survey answers. Tokens are highlighted and delete as one piece with backspace.

Two rows beneath the field show what the prompt is currently using:

* **References in prompt** — each reference token you have inserted

* **Survey answers · click to insert** — each survey question, which you can click to drop into the

text at the cursor

**Important**A reference or a survey question has no effect until its token is in the prompt text. Creating it is not enough.

**Video prompt** is a separate, optional field for camera movement — *"Slow camera movement"* is the placeholder. It only applies when you generate video.

![Type @ for a reference or {{ for a survey answer.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCQzNHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--00213eed28f93cde056a9e2a3bcab661c90915bc/33-ai-prompt-editor.png?cw_image_width=auto)

*Type @ for a reference or {{ for a survey answer.*

### **References**

A reference is an image the AI should reproduce faithfully — a product, a prop, a logo, a face. You make it once and mention it in as many prompts as you like.

Click **References** to open **Reference images**.

The dialog lists every reference you have, with a search box, a count, and an **In prompt** badge on the ones this prompt is using. Each carries a short id, like **`@5723813`**.

**To add one to the prompt:** type **@** in the prompt text and pick it from the list, the way you would tag someone in a message.

#### **Creating a reference**

Click **Add reference** and fill in:

| **Field**            | **Notes**                        |
| -------------------- | -------------------------------- |
| **Title**            | What you will type after **@**   |
| **Description**      | Optional, for your own reference |
| **Reference images** | **1 to 4 images**                |

**Important - Images cannot be changed after creation.** Check them before you save. If you get them wrong, make a new reference and delete the old one.

**Tip -** Describe the reference in the prompt text as well as inserting the token. In the example above, the prompt also names the telephone's *"rotary dial, handset, cord, shape, and red color"* — the token tells the AI which object, the words tell it what matters about it.

![Every reference you have made, and whether it is in the prompt.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCQzdHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--fa0f3efc0bacdce6c00b91a995d54616595a3f17/56-ai-prompt-references.png?cw_image_width=auto)

*Every reference you have made, and whether it is in the prompt.*

![Title, description, and one to four images you cannot change later.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCRFhHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--9df0dd3120f2bdb6edd7c3517a49813806c24001/57-ai-prompt-create-reference.png?cw_image_width=auto)

*Title, description, and one to four images you cannot change later.*

### **Survey questions**

Survey questions are asked at the booth, and the guest's answer is substituted into the prompt. One prompt, a different result per guest.

Click **Survey questions** and switch on **Enable survey questions**.

| **Setting**          | **What it does**                                     |
| -------------------- | ---------------------------------------------------- |
| **Survey Title**     | The heading guests see at the booth                  |
| **Collect answers?** | Whether the answers are stored with the event's data |
| **Questions**        | One block per question                               |

Each question has:

* The **question text** guests read

* **Required** — whether they can skip it

* Its token, **`{{answer1}}`**, **`{{answer2}}`** and so on, numbered in order, with an **In prompt** badge

and **Insert into prompt**

* A list of **Option** / **Token** pairs — the option is what the guest sees, the token is what the

AI receives

* A **default token** — used when the guest gives no answer, and when you generate previews

**Important**Always set a default token. Previews have no guest to answer the question, so without one you are not testing what guests will actually get.

**Note -** The **Option** and the **Token** do not have to match. "Warm and cozy" can send **`1`**, or it can send a whole phrase — whatever reads best inside your prompt sentence.

**Tip -** Turn **Collect answers?** off if you do not need the responses afterwards. It is the simplest way to avoid holding guest data you have no use for.

![Each option maps to a token; the default is used when nobody answers.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCRGZHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--77210de599f4a03d458eb0944fa1c54c22976648/58-ai-prompt-survey.png?cw_image_width=auto)

*Each option maps to a token; the default is used when nobody answers.*

### **Generate previews**

You generate a **batch**, not a single image. Pictor renders every combination of what you pick.

1. **Add test images** — a clear, well-lit photo of a person works best. You can add several.

2. Choose the **Media type** — **Still Image** or **Video**.

3. Choose the **aspect ratios**. For a still image: **1:1 Square**, **4:5 Portrait**, **9:16 Story**,

**16:9 Landscape**, **3:4 Photo**. Multi-select.

1. Choose the **image quality**: **Low Quality** costs 1 token, **High Quality** costs 2. Also

multi-select, so you can compare the two side by side.

1. Read the summary. It spells out the arithmetic and the cost before you commit.

2. Click **Generate N previews**. On a prompt you have not saved yet the button reads

**Save & generate N previews** — it does both.

**Important -** Every test image is rendered against **each** ratio and **each** quality. Three images × two ratios × two qualities is twelve renders, and you are charged for all twelve. The summary box gives you the total before you press anything — read it.

**Note -** Test images are not saved with the prompt.

**Tip -** While you are still finding the wording, use one image, one ratio and **Low Quality**. Add ratios and raise the quality once the look is close.

![The summary shows how many previews and how many tokens before you commit.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCRGJHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--d1e2ec150dede036c07049957c34fcd9cc55d2be/34-ai-prompt-generate.png?cw_image_width=auto)

*The summary shows how many previews and how many tokens before you commit.*

### **Work with your previews**

Every render carries five actions.

| **Action**            | **What it does**                           |
| --------------------- | ------------------------------------------ |
| **Set thumbnail**     | Make this the prompt's thumbnail           |
| **Use as test image** | Feed the render back in as the next input  |
| **Use settings**      | Reload the exact settings that produced it |
| **Download**          | Save the file                              |
| **Delete**            | Remove the render                          |

**Tip - Use settings** is how you work backwards from a good result — reload it, change one thing, generate again.

**Tip - Use as test image** is for building on a look rather than a face. The render is already AI-generated, so the guest's identity drifts a little further with each pass.

![Every render carries these five actions.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCRHJHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--b26a6b80d447627f008b7970e3f95c22a1d7d59e/59-ai-prompt-previews.png?cw_image_width=auto)

*Every render carries these five actions.*

### **The Generated previews tab**

**Work with your previews** above covers the renders from the prompt you have open. The **Generated previews** tab on the Custom AI prompts page gathers *every* render you have ever made, across all your prompts.

> Every test render produced from your prompts — photos and videos — with the prompt, date and download for each.

| **Control**                       | **What it does**                                                                |
| --------------------------------- | ------------------------------------------------------------------------------- |
| Prompt dropdown                   | Narrows the grid to one prompt                                                  |
| **All** / **Photos** / **Videos** | Narrows it by media type. **Videos** is greyed out until you have generated one |

Every card carries a **Photo** or **Video** badge, the prompt that made it, and the date and time it was rendered. Hover a card for two actions: **open the prompt that made it**, and **download the file**.

**Tip -** This is the fastest way to answer *"which version of the prompt produced that one?"* — the date on the card tells you when, and opening the prompt from here takes you straight to it.

**Note -** Renders are test output. They are not what guests receive, and they do not appear in an event's gallery.

![Every render from every prompt, filtered by prompt and media type.](https://app.chatwoot.com/rails/active_storage/blobs/redirect/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBCRHpHd1FZPSIsImV4cCI6bnVsbCwicHVyIjoiYmxvYl9pZCJ9fQ==--84712816ff8991e105c696cf9edd7ac462a46656/60-ai-prompt-previews-tab.png?cw_image_width=auto)

*Every render from every prompt, filtered by prompt and media type.*

### **Save the prompt**

Click **Save**. The prompt appears in **Saved prompts** and can be edited at any time.

### **Use a prompt in an event**

There are two places, depending on how your event is built.

**Simple Mode**

1. Open your event and click **Flow**.

2. Open the **Choose Experience** stage and switch the mode to **Simple**.

3. Under **AI Configuration**, click **Choose Prompts**.

4. Pick your prompt, then set **AI Image Quality** or **AI Video Quality**.

**Advanced (a template)**

1. Open your event and click **Flow**.

2. Open **Choose Experience** and open a template.

3. Select the **Capture** element, open its **Settings**, then **Effect** → **Add effect**.

4. Choose **AI Filter**, **AI Effect** or **AI Video**, then pick your prompt.

**Note -** Using a custom prompt in an event raises that event to the **Pro** tier.

**Important -** Pictor takes a snapshot of the prompt at the moment you add it to an event. Editing the prompt afterwards in **Custom AI prompts** does **not** change events it is already assigned to. To apply an update, come back here and re-select the prompt.

This is deliberate — it stops a prompt edit quietly changing the output of an event a client has already signed off.

**Tip -** If you have changed a prompt and the event is still producing the old look, this is why. Re-select the prompt on the stage.

### **Writing a good prompt**

Work through these in order. The example prompt in the screenshots above uses every one of them.

| **Category**         | **Examples**                                                                   |
| -------------------- | ------------------------------------------------------------------------------ |
| **Prompt name**      | "Retro Red Telephone Portrait", "Game Day Glory", "Cyber Hero"                 |
| **Test image**       | Clear, well-lit photo of a person with a visible face                          |
| **What to keep**     | "Preserve the person's identity, facial features, skin tone, hairstyle"        |
| **Subject**          | Football player, superhero, astronaut, fashion model, warrior, chef            |
| **Action**           | Posing for a portrait, holding a trophy, walking through smoke                 |
| **Setting**          | Football stadium, city skyline at night, 1970s living room                     |
| **Lighting**         | Magic hour, cinematic glow, moody shadows, glowing neon                        |
| **Style**            | Hyper-real, cartoon, comic book, vintage photo, watercolour                    |
| **Specific details** | Holding a football, gold aviators, confetti falling                            |
| **What to avoid**    | "Do not add extra people, distorted hands, extra fingers, text, or watermarks" |

**Tip -** The two people leave out are **what to keep** and **what to avoid**. A line telling the AI to preserve the guest's face is what stops it inventing a different person, and a line listing what you do not want is the fastest fix for extra fingers and stray text.

### **Asking us for a prompt**

If you want a look you cannot get to yourself, click **Request a prompt** at the top of the Custom AI prompts page and describe what you are after. We write it and it appears in your saved prompts.

**Tip**Send a reference photo of the look you want along with the description. It settles more than a paragraph of adjectives will.

### **See also**

* **Using AI Effects** — the fixed looks you apply to a capture element

* **Creating an AI video** — generating video rather than stills

* **Plans and features** — what tokens cost and how to top up