# Model Generation

***

### Model Generation: start from an image

<figure><img src="/files/r2gF7gtLJJRy1hfo16dP" alt=""><figcaption></figcaption></figure>

### Model Generation: start from text

<figure><img src="/files/Rmew2L6nj6rc8QnQrpvd" alt=""><figcaption></figcaption></figure>

***

## Create a New Scene

Every time you create a new scene, you will be asked how would you like to start your generation. To move forward with your generation, select one of the following two methods:<br>

* <i class="fa-image">:image:</i> **Start from image:** best for editing or refining and existing image. (Learn how to [here](#start-from-an-image))
* <i class="fa-message-text">:message-text:</i> **Start from text:** best for generating images from scratch. (Learn how to [here](#start-from-text))

<div data-with-frame="true"><figure><img src="/files/ilklEkJwE7Sw8xqBOWBJ" alt=""><figcaption></figcaption></figure></div>

**How to create a new scene**

* **Scene Panel on the left side of the screen:** \
  On the left upper side of the screen on the "Scene Panel", select the <i class="fa-plus-large" style="color:$primary;">:plus-large:</i> icon to a create a new scene.
* **From anywhere on the playground**\
  Right-click anywhere on your playground and choose <kbd><mark style="color:$primary;">**+ create new scene**<mark style="color:$primary;"></kbd> or use the keyboard shortcut <i class="fa-command">:command:</i>+Shift+S on Mac and Control+Shift+S on Windows.
* **Every time you create a new playground**\
  Upon launching your new playground, you will be prompted to choose how you would like to begin your scene.

***

## Start from an image

Select "start from image" to edit or refine an existing visual. This feature lets you use your image as a base and implement changes or add details through text prompts. The task is highly flexible, but the outcome relies on the clarity of your prompt and the image generation model used.

{% hint style="info" icon="language" %}
**How to Change Your Tutorial Language**\
Tutorials are shown in English by default. To view this tutorial in a different language, hover over the top right corner and select <i class="fa-language">:language:</i> **Language**.
{% endhint %}

{% @arcade/embed flowId="x1ALaO14jk50rwMGo4bh" url="<https://app.arcade.software/share/x1ALaO14jk50rwMGo4bh>" %}

{% stepper %}
{% step %}

#### Start a New Generation

{% endstep %}

{% step %}

#### Select "Start from image"

{% endstep %}

{% step %}

#### Upload Your Image

Select and upload the image you would like to edit.

You may upload an image by:

* Dragging and dropping an image from the Library panel
* Selecting <kbd><mark style="color:$primary;">**Upload an image**<mark style="color:$primary;"></kbd> inside the scene
* Dragging and dropping an image from a folder on your computer

Once uploaded, the product will appear in the scene.
{% endstep %}

{% step %}

#### Select your Image Model

This is the first option on the left-side Task Menu. The best model for consistency <kbd><mark style="color:$primary;">**Nano Banana Pro**<mark style="color:$primary;"></kbd> is already automatically selected. You may choose other models to generate your image.

<details>

<summary><strong>Available AI Generation Models for Image Generation</strong></summary>

* **Nano Banana Pro:** Seamless, consistent edits with accurate details and reflections
* **Nano Banana 2:** Refine and transform visuals with faithful structure and natural detail
* **Nano Banana:** Seamless, consistent edits with accurate details and reflections
* **GPT5:** Photorealistic details with precise semantic understanding.
* **Seedream 4.5:** Ultra-consistent edits with superior detail retention and natural lighting
* **Seedream 4:** High-fidelity edits with natural lighting and detail retention
* **Seedream 3:** Fresh, consistent variations from original images
* **Flux Kontext:** Precise, seamless edits with natural results
* **Qwen:** Versatile visual creation with accurate text integration

</details>
{% endstep %}

{% step %}

#### Type your Text Prompt

On the right-side Generation Settings panel type your prompt.&#x20;

<details>

<summary><strong>Text Prompt Tips</strong></summary>

* **Be Direct, Not Polite:** AI doesn't need formalities. Use clear, simple commands to get exactly what you want. Especially for languages such as Japanese, it is very important to keep the language direct and avoid formal writing.
* **Start with an action verb:** Tell the AI exactly what action to take right at the beginning of the sentence. Use clear commands like "Replace the background with...", "Remove the...", or "Change the lighting to..." instead of just describing a scene.
* **Work from macro to micro:** Building on your point about working in steps, always change the biggest elements first. Fix the background or the overall lighting in step one. Once that is locked in, use a second prompt to add or refine smaller details.
* **Literal over poetic:** Since you want to avoid abstract language, remember that the AI cannot interpret feelings. Don't ask for a "luxurious vibe." Instead, ask for the physical things that create that vibe: "gold accents, soft studio lighting, and smooth marble."
* **Prompt the Edits, Not the Whole Image:** When editing, you do not need to describe the entire original image all over again. Only describe the specific change you want to make to the targeted area.
* **Define the anchor points:** Expanding on your point about being specific—if you are changing the floor but want to keep the wall, explicitly state the boundary: "Keep the back wall exactly as it is, change only the floor to white concrete."

</details>

<details>

<summary><strong>Prompt Library (Save and Re-Use)</strong></summary>

You may save your prompts to your library and use them later. Any prompts saved on the Library through this steps will be available for re-use later. You will be able to save prompts anytime, even after generation has been completed. With the Generation Settings Panel opened:

1. Select <kbd><mark style="color:$primary;">**Library**<mark style="color:$primary;"></kbd> under the prompt text area.
2. On the right-side of the search bar, select the <i class="fa-plus-large" style="color:$primary;">:plus-large:</i> icon to save your current prompt.

</details>

<details>

<summary><strong>Automatic Prompt from Image</strong></summary>

You may create prompts by uploading an image and Beachside will automatically create a description. With the Generation Settings Panel opened:

{% hint style="danger" %}
This step will replace any prompt that is currently on the text area.
{% endhint %}

1. Select <kbd><mark style="color:$primary;">**Library**<mark style="color:$primary;"></kbd> to re-use a previously saved prompt.
2. From the <i class="fa-cloud-arrow-up" style="color:$primary;">:cloud-arrow-up:</i> icon, upload an image and Beachside will create a description and place the text in the prompt field.

</details>
{% endstep %}

{% step %}

#### Optional: Use the Brush Tool

Use Brush to provide localized visual guidance to the AI model. Draw shapes, separate instructions by color, and clarify intent directly on the canvas:

1. Under the scene, select the <i class="fa-paintbrush" style="color:$primary;">:paintbrush:</i> <kbd><mark style="color:$primary;">**Brush**<mark style="color:$primary;"></kbd> from the tools menu.
2. Draw shapes directly on the canvas to indicate areas of change.
3. Reference Brush Layers in Your Prompt

Read more about the Brush tool [here](/tools/brush.md)
{% endstep %}

{% step %}

#### Optional: Add a reference image

You may add up to 3 reference images to the basic text input to give more guidance to the prompt generator. On the Generation Settings Panel:

1. Select the <i class="fa-image" style="color:$primary;">:image:</i> <kbd><mark style="color:$primary;">**Reference Image**<mark style="color:$primary;"></kbd> under the prompt text area.
2. Select up to 3 images from the Playground or Upload tab.
3. If uploading:
   * Select the <kbd><mark style="color:$primary;">**Upload**<mark style="color:$primary;"></kbd> tab.
   * Select <mark style="color:$primary;">**Upload**</mark> or choose a previously uploaded image.
4. To remove a reference image:
   * Hover over the image number.
   * Select the <i class="fa-xmark" style="color:$primary;">:xmark:</i> button.
     {% endstep %}

{% step %}

#### Select the Number of Outputs

In the right-side Light settings panel:

* Select how many variations to generate.
* You may generate up to 4 scenes at once.
  {% endstep %}

{% step %}

#### Generate

After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

The generated scenes will appear in your Playground.
{% endstep %}
{% endstepper %}

***

## Start from text

Select "start from text" to generate an image from scratch. The outcome relies on the clarity of your prompt and the image generation model used.

{% hint style="info" icon="language" %}
**How to Change Your Tutorial Language**\
Tutorials are shown in English by default. To view this tutorial in a different language, hover over the top right corner and select <i class="fa-language">:language:</i> **Language**.
{% endhint %}

{% @arcade/embed flowId="lPigmoOPOsheEpvNrNzB" url="<https://app.arcade.software/share/lPigmoOPOsheEpvNrNzB>" %}

{% stepper %}
{% step %}

#### Start a New Generation

{% endstep %}

{% step %}

#### Select "Start from text"

{% endstep %}

{% step %}

#### Select your Image Model

This is the first option on the left-side Task Menu. <kbd><mark style="color:$primary;">**Seedream 4**<mark style="color:$primary;"></kbd> is already automatically selected. You may choose other models to generate your image.

<details>

<summary><strong>Available AI Generation Models for Text Generation</strong></summary>

* **Seedream 4:** High-fidelity edits with natural lighting and detail retention
* **Nano Banana 2:** Refine and transform visuals with faithful structure and natural detail
* **Seedream 4o:** Ultra-consistent edits with superior detail retention and natural lighting
* **GPT5:** Photorealistic details with precise semantic understanding.
* **Seedream 3:** Fresh, consistent variations from original images
* **Qwen:** Versatile visual creation with accurate text integration
* **Imagen 4:** Photorealistic, detailed imagery with fast typography precision.
* **Recraft:** Consistent visual generation for creative workflows.

</details>
{% endstep %}

{% step %}

#### Type your Text Prompt

On the Text Prompt area type your prompt.&#x20;

<details>

<summary><strong>Text Prompt Tips</strong></summary>

* **Be Direct, Not Polite:** AI doesn't need formalities. Use clear, simple commands to get exactly what you want. Especially for languages such as Japanese, it is very important to keep the language direct and avoid formal writing.
* **Start with an action verb:** Tell the AI exactly what action to take right at the beginning of the sentence. Use clear commands like "Replace the background with...", "Remove the...", or "Change the lighting to..." instead of just describing a scene.
* **Work from macro to micro:** Building on your point about working in steps, always change the biggest elements first. Fix the background or the overall lighting in step one. Once that is locked in, use a second prompt to add or refine smaller details.
* **Literal over poetic:** Since you want to avoid abstract language, remember that the AI cannot interpret feelings. Don't ask for a "luxurious vibe." Instead, ask for the physical things that create that vibe: "gold accents, soft studio lighting, and smooth marble."
* **Prompt the Edits, Not the Whole Image:** When editing, you do not need to describe the entire original image all over again. Only describe the specific change you want to make to the targeted area.
* **Define the anchor points:** Expanding on your point about being specific—if you are changing the floor but want to keep the wall, explicitly state the boundary: "Keep the back wall exactly as it is, change only the floor to white concrete."

</details>

<details>

<summary><strong>Prompt Library (Save and Re-Use)</strong></summary>

You may save your prompts to your library and use them later. Any prompts saved on the Library through this steps will be available for re-use later. You will be able to save prompts anytime, even after generation has been completed:

1. Select <kbd><mark style="color:$primary;">**Library**<mark style="color:$primary;"></kbd> under the prompt text area.
2. On the right-side of the search bar, select the <i class="fa-plus-large" style="color:$primary;">:plus-large:</i> icon to save your current prompt.

</details>

<details>

<summary><strong>Automatic Prompt from Image</strong></summary>

You may create prompts by uploading an image and Beachside will automatically create a description:

{% hint style="danger" %}
This step will replace any prompt that is currently on the text area.
{% endhint %}

1. Select <kbd><mark style="color:$primary;">**Library**<mark style="color:$primary;"></kbd> to re-use a previously saved prompt.
2. From the <i class="fa-cloud-arrow-up" style="color:$primary;">:cloud-arrow-up:</i> icon, upload an image and Beachside will create a description and place the text in the prompt field.

</details>
{% endstep %}

{% step %}

#### Select an Aspect Ratio

Under:

1. Open the Aspect Ratio dropdown menu.
2. Select the format appropriate for your output.

<details>

<summary><strong>Available aspect ratios</strong></summary>

* Portrait (9:16, 768×1344px)
* Widescreen (21:9, 1536×672px)
* Vertical Widescreen (9:21, 672×1536px)
* Flexible 5:4 (5:4, 1152×896px)
* Flexible 4:5 (4:5, 896×1152px)
* Square (1:1, 1024×1024px)
* Photo (3:2, 1248×832px)
* Vertical Photo (2:3, 832×1248px)
* Classic (4:3, 1184×864px)
* Vertical Classic (3:4, 864×1184px)
* Landscape (16:9, 1344×768px)

</details>
{% endstep %}

{% step %}

#### Select the Number of Outputs

In the right-side Light settings panel:

* Select how many variations to generate.
* You may generate up to 4 scenes at once.
  {% endstep %}

{% step %}

#### Generate

After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

The generated scenes will appear in your Playground.
{% endstep %}
{% endstepper %}


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.theseaai.com/tasks/model.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.