> For the complete documentation index, see [llms.txt](https://docs.theseaai.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.theseaai.com/learn/tutorials.md).

# Tutorials

## Prompts

* [Create a prompt from a reference image](#reference-image)
* [Create an Ideation prompt](#ideation-prompt)

## Repair image quality

* [Repair a damaged product image quality](#repair-a-damaged-product-image-quality)
* [Repair a damaged portrait image quality](#repair-a-damaged-portrait-image-quality)
* [Realistic blending of packshots](#realistic-blending-of-packshots)

## Creative effects

* [3D Typography](#id-3d-typography)

***

## Create a prompt from a reference image <a href="#reference-image" id="reference-image"></a>

Use an AI chat tool, such as Google's Gemini, to turn a reference image into a text prompt. Refine the prompt, then generate a new visual.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F0JMSGzK0dOURoPLxntst%2F1.png?alt=media&amp;token=d74b8665-2d70-4461-a49d-af6a86644fba" alt=""><figcaption></figcaption></figure>

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FAMgjtDJORCTJzLB2znRl%2F2.png?alt=media&amp;token=52de924d-30cc-4b9e-be10-bcb266aac54e" alt=""><figcaption></figcaption></figure>

### Interactive Tutorial

{% 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 url="<https://app.arcade.software/share/87SjVktgKuB1mGU59DGc>" flowId="87SjVktgKuB1mGU59DGc" %}

### How to

{% stepper %}
{% step %}

#### Select your reference image

Select a reference image.

* A clear composition that includes an unmistakable subject.
* Avoid images with lots of text or people, or no clear subject, make it harder for the AI to understand what to use.

<div data-with-frame="true"><figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FRs8ghMgTbfqvlbREin6k%2Freference-image-1.png?alt=media&amp;token=ea5f631f-1cc5-4ffd-8200-20d64c5a4a74" alt=""><figcaption></figcaption></figure></div>
{% endstep %}

{% step %}

#### Upload your image to an AI chat tool

Beachside recommends Google Gemini for better results.
{% endstep %}

{% step %}

#### Copy and paste the following template

***

Create a text prompt.\
Do not generate any images.

Based on the attached image, create text descriptions for *<mark style="color:green;background-color:green;">**3 different scenes.**</mark>* Reproduce the *<mark style="color:green;background-color:green;">**same angle, light, coloring and finish**</mark>*<mark style="color:green;background-color:green;">**.**</mark>\
Each scene keeps the background of the attached image unchanged, with the product placed in it (always refer to it as product in the text description), to create a flat advertising visual.

Write the text descriptions following these guidelines:

* Write each text description as one continuous paragraph of text;
* Never mention the file name of the attached reference image;
* Only refer to the product shown on the image using the word: product;
* Never describe the shape, design or other qualities of the product;
* *<mark style="color:red;background-color:$danger;">**Add any additional requests here**</mark>*

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."

***

{% prompt description="Copy your prompt here" icon="message-lines" openInAIProviders="false" %}

```markdown
Create a text prompt.
Do not generate any images.

Based on the attached image, create text descriptions for 3 different scenes. Reproduce the same angle, light, coloring and finish.Each scene keeps the background of the attached image unchanged, with the product placed in it (always refer to it as product in the text description), to create a flat advertising visual.

Write the text descriptions following these guidelines:
- Write each text description as one continuous paragraph of text;
- Never mention the file name of the attached reference image;
- Only refer to the product shown on the image using the word: product;
- Never describe the shape, design or other qualities of the product;
- Add any additional requests here

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."
```

{% endprompt %}
{% endstep %}

{% step %}

#### Edit the template

You must edit the area marked in *<mark style="color:red;background-color:$danger;">**red**</mark>*, the area marked in *<mark style="color:green;background-color:green;">**green**</mark>* should be edited if necessary. Do not forget to remove the "*Add any additional requests here"* if not necessary.

{% hint style="info" %}

* **Be concrete rather than abstract:** write "matte stone surface, soft studio lighting, gold accents" instead of "a premium feel".
* **Keep each run to 5 prompts or fewer:** beyond that the results start to repeat one another.
  {% endhint %}

This is the prompt after editing:

***

Create a text prompt.\
Do not generate any images.

Based on the attached image, create text descriptions for *<mark style="color:green;">`3 different scenes.`</mark>* Reproduce the *<mark style="color:green;">`same angle, light, coloring and finish.`</mark>*\
Each scene keeps the background of the attached image unchanged, with the product placed in it (always refer to it as product in the text description), to create a flat advertising visual.

Write the text descriptions following these guidelines:

* Write each text description as one continuous paragraph of text;
* Never mention the file name of the attached reference image;
* Only refer to the product shown on the image using the word: product;
* Never describe the shape, design or other qualities of the product;
* *<mark style="color:red;">`Replace the orange for a Japanese Yuzu peel and remove all floral elements; the wooden stump surface carries fine water droplets. Always include these elements in the scene.`</mark>*

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."

***

{% endstep %}

{% step %}

#### Check and edit the result if necessary

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FkCag8UD4noLRopeWvmOF%2FGroup%2084%20(1).png?alt=media&amp;token=2b7c29bf-c009-4ee9-9880-86ac8005a914" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

#### Use your prompt on Beachside

Follow these steps:

1. Create a new scene.
2. Place your product (or products) on the new scene.
3. Paste one of the created prompts.
4. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FPEh9gpCq9Jy8kKajqtRS%2FGroup%2086.png?alt=media&amp;token=5b14afda-3a29-4288-b80c-52c216ae6cd9" alt=""><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

### FAQ

<details>

<summary><strong>Why does the template start with "Do not generate any images"?</strong></summary>

The output of this step is text. Without that instruction, a conversational AI that receives an image will usually reply with an image and skip writing the prompt.

</details>

<details>

<summary><strong>Can the English paragraph at the end be translated?</strong></summary>

No. That paragraph carries the image-quality keywords, and the image model recognises the original English most reliably. Translating or rewriting it lowers output quality.

</details>

<details>

<summary><strong>Why not describe the product's shape and design?</strong></summary>

The product's appearance comes from the photo you upload. If the prompt says something like "amber glass pump bottle", the image model will draw a new bottle from that text instead of using your product, and the result will not match.

</details>

<details>

<summary><strong>The multi-product image has inconsistent scale and shadows going different ways.</strong></summary>

Check whether the two constraints were deleted from the multi-product template: "always state the arrangement, and the front-to-back and height hierarchy" and "all products share the same lighting direction and shadow logic".<br>

</details>

***

## Create an Ideation prompt <a href="#ideation-prompt" id="ideation-prompt"></a>

Use an AI chat tool, such as Google's Gemini, to create an ideation prompt based on elements such as: brand style, color palette, key elements, materials and camera angle.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FLn8vNaw56imHGHHEkwKG%2FGroup%2093.png?alt=media&amp;token=4595c547-f5b7-4e61-8162-6e1f624137d4" alt=""><figcaption></figcaption></figure>

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FybQkO2eC2OfK41VjJHHH%2FGroup%2096.png?alt=media&amp;token=9ade2fcc-85ed-49fa-b294-0fba4cfdd098" alt=""><figcaption></figcaption></figure>

### Interactive Tutorials

{% 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 url="<https://app.arcade.software/share/HKi3t58cPsThjBcmvcfR>" flowId="HKi3t58cPsThjBcmvcfR" %}

{% @arcade/embed url="<https://app.arcade.software/share/00ew3OElfRMKiPGC5dHc>" flowId="00ew3OElfRMKiPGC5dHc" %}

{% @arcade/embed url="<https://app.arcade.software/share/4sMaFIH6rKnQarfuG03F>" flowId="4sMaFIH6rKnQarfuG03F" %}

### How to use

{% stepper %}
{% step %}

#### Decide your visual guideline

You need to decide the following contents:

1. Brand style
2. Color palette
3. Key elements
4. Materials
5. Camera angle

\
You may extract these elements from your brand guidelines, or use AI to create them for you, follow the steps according to your needs:
{% endstep %}

{% step %}

#### If you don't have a visual guideline

You may extract fromexisting brand documents, visuals or create your own from scratch:

<details>

<summary><i class="fa-file-lines">:file-lines:</i> <strong>Existing Brand Guidelines document</strong></summary>

Based on the brand guidelines, the best way to keep a set of images consistent is to write all descriptions yourself. However you may ask an AI chat to extract all information from a Brand Guidelines File:

1. Upload your Brand Guildeines file to an AI chat (Beachside recommends Google Gemini for better results).
2. Use the following prompt to extract the visual guidelines:

{% code title="Text prompt" overflow="wrap" %}

```
Based on the attached document, help me extract its visual elements.
Do not generate any images.

Extract each of the following:
1. Brand style
2. Color palette
3. Key elements
4. Materials
5. Camera angle

The result should be keywords only, organized under each one of the 5 categories above.
```

{% endcode %}

</details>

<details>

<summary><i class="fa-image">:image:</i> <strong>Use existing reference image</strong></summary>

1. Upload a single or multiple image references to an AI chat (Beachside recommends Google Gemini for better results).
2. Use the following prompt to extract the visual guidelines:

{% code title="Text prompt" overflow="wrap" %}

```
Based on the attached image, help me extract its visual elements.
Do not generate any images.

Extract each of the following:
1. Brand style
2. Color palette
3. Key elements
4. Materials
5. Camera angle

The result should be keywords only, organized under each one of the 5 categories above.
```

{% endcode %}

</details>

<details>

<summary><i class="fa-square-dashed-circle-plus">:square-dashed-circle-plus:</i> <strong>Create from scratch using AI chat</strong></summary>

Use AI to generate keyword batches from a broad concept (for instance a premium red gift set, a gold-packaged pressed powder, or holiday atmosphere). Select promising options for further refinement.

{% code title="Text prompt" overflow="wrap" %}

```
Help me brainstorm ideas based on the "〇〇〇〇" element.
Do not generate any images.

Based on 〇〇〇〇 (brand/product/style...), list 5 ideas for each of:
- Color palette
- Key scene elements (props, materials...)
- Camera angles
- Scene lighting

The result should include idea number, one line of explanation and keywords. The goal is to organize it in a way that it is easy to choose from, not full copy.
```

{% endcode %}

</details>
{% endstep %}

{% step %}

#### Copy and paste the following template

***

Create a text prompt.\
Do not generate any images.

Using the following elements as inspiration, write text descriptions for *<mark style="color:green;background-color:green;">**3 different scenarios.**</mark>*

* Brand/style impression: *<mark style="color:red;background-color:$danger;">**Add your description here**</mark>*
* Color palette: *<mark style="color:red;background-color:$danger;">**Add your description here**</mark>*
* Key elements: *<mark style="color:red;background-color:$danger;">**Add your description here**</mark>*
* Materials/textures: *<mark style="color:red;background-color:$danger;">**Add your description here**</mark>*
* Camera angle: *<mark style="color:red;background-color:$danger;">**Add your description here**</mark>*

For each scenario, place the product within it (always refer to it as "product" in the text description). This process is intended to create visuals for print advertisements.

Follow these guidelines when writing the descriptions: ​

* Write each description as a single, continuous paragraph;
* Use only the word "product" to refer to the item shown in the image; Do not describe the product's shape, design, or other specific features;
* Clearly state the product's placement within the scene and the surrounding negative space;
* Ensure distinct differences between the 3 scenes—not limited to color variations alone;
* Fill in details not explicitly specified (such as background, surfaces, lighting specifics, etc.) while maintaining a style consistent with the elements above;
* Do not include text, logos, or people.

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."

***

{% prompt description="Copy your prompt here" icon="message-lines" openInAIProviders="false" %}

```markdown
Create a text prompt.
Do not generate any images.
Using the following elements as inspiration, write text descriptions for 3 different scenarios.

- Brand/style impression: Add your description here
- Color palette: Add your description here
- Key elements: Add your description here
- Materials/textures: Add your description here
- Camera angle: Add your description here

For each scenario, place the product within it (always refer to it as "product" in the text description). This process is intended to create visuals for print advertisements.

Follow these guidelines when writing the descriptions: ​
- Write each description as a single, continuous paragraph;
- Use only the word "product" to refer to the item shown in the image; Do not describe the product's shape, design, or other specific features;
- Clearly state the product's placement within the scene and the surrounding negative space;
- Ensure distinct differences between the 3 scenes—not limited to color variations alone;
- Fill in details not explicitly specified (such as background, surfaces, lighting specifics, etc.) while maintaining a style consistent with the elements above;
- Do not include text, logos, or people.

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."
```

{% endprompt %}
{% endstep %}

{% step %}

### Edit the template

You must edit the area marked in *<mark style="color:red;background-color:$danger;">**red**</mark>*, the area marked in *<mark style="color:green;background-color:green;">**green**</mark>* should be edited if necessary.

{% hint style="info" %}

* Keep each run to 5 prompts or fewer: beyond that the results start to repeat one another.
* Use words that carry visual information, not judgement: "wabi-sabi, understated luxury" or "warm earth tones, matte travertine"
* At minimum, keep the color palette and key elements so the AI still has enough direction to stay on-brand.
* Only Key elements, Materials/textures and Camera angle need to be concrete. Brand/style impression and Color palette are meant to stay directional. Over-specifying them leaves the AI no room to differentiate, and the three scenes come back nearly identical.
  {% endhint %}

This is the prompt after editing:

***

Create a text prompt.\
Do not generate any images.

Using the following elements as inspiration, create text descriptions for *<mark style="color:green;">`3 brand-new scenes:`</mark>*

* Brand / style impression: *<mark style="color:red;">`ReeeeWabi-sabi understated lixury`</mark>*
* Colour palette: *<mark style="color:red;">`Warm earth tones;beige travertine base`</mark>*
* Key elements: *<mark style="color:red;">`Stacked towels`</mark>*
* Material / texture: *<mark style="color:red;">`Natural travertine,soft terry clath`</mark>*
* Camera angle: *<mark style="color:red;">`Slightly high-angle eye-level view;medium close-up shot`</mark>*

Write the text descriptions following these guidelines:

* write each text description as one standalone document
* always state where the product sits in the scene, and how much space is left around it
* the product and everything around it share one light source; all shadows fall on the same side
* the 3 scenes must differ clearly from one another, not only in colour
* fill in anything the elements do not specify yourself, keeping the same character as the elements above
* never describe the shape, design or other qualities of the products

To improve image quality, always include this text exactly as written in english at the end of the prompt: "Flat advertising image. Photorealism, 8K resolution, incredibly sharp focus, hyper-detailed textures, immaculate studio lighting, crisp highlights, commercial beauty advertising style, ultra-premium quality."
{% endstep %}

{% step %}

### Check and edit the result if necessary

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FD8FsvQFZiVpj2Pi9ZDEA%2Fimage%2069.png?alt=media&amp;token=e88d97ca-6c55-4fe3-9619-a9b4dd70282c" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Use your prompt on Beachside

Follow these steps:

1. Create a new scene.
2. Place your product (or products) on the new scene.
3. Paste one of the created prompts.
4. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FNeRMFdN5kHLdKIwbWAfr%2Fimage.png?alt=media&amp;token=986ab066-b1cc-4292-b7dd-424a90e12310" alt=""><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

### FAQ<br>

<details>

<summary><strong>How do I get a consistent series of images?</strong></summary>

Keep the elements fixed and vary only the composition. Hold the palette, material and lighting constant, and change only the props and viewpoint — the three prompts then read as one world seen from three angles. Change the elements each time and the series falls apart.

</details>

***

<br>

## Repair a damaged product image quality

Repair a distorted product image or restore missing details. Create a balanced image with consistent texture.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FD1ca4zgqyvVdSd3d5pNS%2Fgitbook_task_upscale.png?alt=media&amp;token=05d5f356-323c-4654-9150-75a4838ce1b1" alt=""><figcaption></figcaption></figure>

### Interactive Tutorials

{% 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 url="<https://app.arcade.software/share/OkbMsJbfWernHM4i90p9>" flowId="OkbMsJbfWernHM4i90p9" %}

### How to

{% stepper %}
{% step %}

#### Split image layers

Upload the image requiring restoration to Beachside and process it using "layering."

* Separate the product and the background onto two artboards to facilitate independent processing later on.

Follow these steps:

1. Create a new scene.
2. Place your scene graph into the new scene.
3. Right-click and select <kbd><mark style="color:$primary;">**Split into layers**<mark style="color:$primary;"></kbd>.
4. Once the layering is complete, copy the background and the product separately onto a new scene

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2Fdinreul2ASL0K4uAwXWj%2FSPLIT.png?alt=media&amp;token=90947d45-dd83-4194-a506-a98e996e53a1" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

#### Copy and paste the following template

Select the background canvas, choose the "model generation" task, and enter the following prompt.

***

* Restore the image quality of the background
* *<mark style="color:red;background-color:$danger;">**Add any additional requests here**</mark>*
* Keep other visual elements unchanged.

***

{% prompt description="Copy your prompt here" icon="message-lines" openInAIProviders="false" %}

```markdown
Restore the image quality of the background.
Add any additional requests here
Keep other visual elements unchanged.
```

{% endprompt %}
{% endstep %}

{% step %}

#### Edit the prompt template

You must edit the *<mark style="color:red;background-color:$danger;">**red**</mark>*-marked area. If no additional requirements are needed, don't forget to delete "Add any other requirements here".

{% hint style="info" %}

* **Be concrete rather than abstract:**

  Write "correct the unnatural color gradient of the sky in the background" instead of "add a sense of naturalness".
  {% endhint %}

This is the prompt after editing:

***

* Restore the image quality of the background,
* correcting unnatural sky color gradients
* Keep other visual elements unchanged.

***

{% endstep %}

{% step %}

### Place your products on your background

Paste the product layer onto the repaired background and adjust its position. Use "Combo" task to blend the product and background.

Follow these steps:

1. Use the text tool to select the product.
2. Enter the prompt: "*`Seamlessly integrate the product into the background.`*"
3. Add a reference image and label it "product lighting and shadows."
4. Make sure Nano Banana pro is the selected model.
5. Make sure [Precise Mode](https://docs.theseaai.com/tasks/model#generation-effort-for-nano-banana-pro) is selected.
6. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2Fik7pklsWnYWdgtEVF9rB%2FProduct%20Integration.png?alt=media&amp;token=51c06690-ad69-4bbe-a8c9-be68f3900966" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Repair existing products

In case your products are distorted, follow these steps in order to replace them and match the desired final angle. Use <kbd><mark style="color:$primary;">**Combo**<mark style="color:$primary;"></kbd>to repair product details:

1. Use the <kbd><mark style="color:$primary;">**Combo Product reference**<mark style="color:$primary;"></kbd>feature
2. Add high-definition product images for each product

   <div data-gb-custom-block data-tag="hint" data-style="info" class="hint hint-info"><p>For products at unusual angles (such as lying down/flat angles), it is recommended to use products with similar angles to avoid misaligned labels.</p></div>
3. Enter the prompt *`Optimize product labels and clearly render the details of the product structure`*
4. Make sure Nano Banana pro is the selected model.
5. Make sure [Precise Mode](https://docs.theseaai.com/tasks/model#generation-effort-for-nano-banana-pro) is selected.
6. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FP5HObCaCvSF4w3Pwbvq1%2FFix%20product%20labels.png?alt=media&amp;token=bef7ed6f-26c1-4086-a588-65f88ea34988" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

### Optional step: create a new product angle before blending it into the background

Follow these steps to create a product angle based on a reference image:

{% @arcade/embed url="<https://app.arcade.software/share/7dX9fpGAcg7ba6AQBRLW>" flowId="7dX9fpGAcg7ba6AQBRLW" %}
{% endstep %}
{% endstepper %}

### FAQ

<details>

<summary><strong>It's still soft after a repair — can I run it again?</strong></summary>

You can, but stop at two rounds. Each repair builds on the previous result, so repeated repair accumulates loss the same way repeated generation does — by the third or fourth round you are losing more than you gain. If two rounds haven't got you there, go back to an earlier, sharper version and start from that instead of pushing the current one further.

</details>

<details>

<summary><strong>The background came back as a different scene</strong></summary>

The prompt went beyond the template. This step restores quality; it does not re-create the scene. The closing clause `keep other visual elements unchanged` is what pins everything else down, so never delete or reword it. Go back one step and submit the template plus a single specific repair instruction.

</details>

<details>

<summary><strong>The product looks pasted on rather than photographed</strong></summary>

Always add the reference image and label it *product lighting and shadows* — the model needs to know which direction the light came from in the original frame. Then check three things: the shadow direction matches everything else, a contact shadow sits where the product meets the surface, and the edges carry no white fringe. That floating look is nearly always a missing contact shadow.

</details>

***

## Repair a damaged portrait image quality

Repair a distorted portrait image or restore missing details. Create a balanced image with consistent texture.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2Fh7KPKRqn5fQINpLqjxOP%2Fgitbook_task_upscale.png?alt=media&amp;token=a038f733-f5a3-4a4e-b834-e71a602f399a" alt=""><figcaption></figcaption></figure>

### Interactive Tutorial

{% 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="PtkvKNQjStCohGUts9UZ" url="<https://app.arcade.software/share/PtkvKNQjStCohGUts9UZ>" %}

### How to

{% stepper %}
{% step %}

#### &#x20;Split your image into 2: person and background

Upload the image to be restored, separate into layers. Since the human figure is more complex than the product and is holding it, using the background removal feature is not recommended:

1. Create a new scene
2. Add the image that needs repair
3. Right click and select <kbd><mark style="color:$primary;">**Split into layers**<mark style="color:$primary;"></kbd>.
4. Wait for the layering process to complete, delete the background and leave only the person. Add a white solid behind the person by choosing the square shape on the tools menu under the scene.

Now repeat the process but remove the person:

1. Create a new scene
2. Select <kbd><mark style="color:$primary;">**Model Generation**<mark style="color:$primary;"></kbd>
3. Enter the following prompt to remove the person: *`Remove the model and restore the background`*

You will end up with the following two images:&#x20;

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2Fu8VbTzpHFX7WniCe5AQl%2FSplit%20vision2.png?alt=media&amp;token=b6373806-5560-4d7b-8a00-608be4209fd2" alt=""><figcaption></figcaption></figure>
{% endstep %}

{% step %}

#### Use an AI chat platform to create your prompt

Provide the background and the woman separately to your preferred AI Chat platform (Gemini, ChatGPT, Claude) to generate prompts for the details that need optimization. The following steps need to be performed separately for the background and the person.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F9ZIuVUAz5slbL7RgiWZt%2FSPLIT.png?alt=media&amp;token=c2fda95a-6d70-4e44-8ea0-3e16df3e62dd" alt=""><figcaption></figcaption></figure>

Use the following prompt with one image at a time:

***

Based on the attached image, please extract the complete set of visual prompts for a restored version; do not generate any images. Provide detailed description of the material of each item, and accurately reproduce the colors, layout, and materials. \
\
&#x20;*<mark style="color:red;background-color:$danger;">**Add any other requests here.**</mark>*&#x20;

The prompt must begin with this sentence: "Perform a perfect restoration based on the reference image."

***

{% prompt description="Copy your prompt here" icon="message-lines" openInAIProviders="false" %}

```markdown
Based on the attached image, please extract the complete set of visual prompts for a restored version; do not generate any images. Provide detailed description of the material of each item, and accurately reproduce the colors, layout, and materials. 

Add any other requests here. 

The prompt must begin with this sentence: "Perform a perfect restoration based on the reference image."
```

{% endprompt %}
{% endstep %}

{% step %}

#### Edit your prompt

You must edit the area marked in *<mark style="color:red;background-color:$danger;">**red**</mark>*. If no additional specific material requirements are needed, remember to remove "Add any other requirements here".  The AI chat might not understand the materials correctly, read your prompt and add information as needed.

{% hint style="info" %}
**Be concrete rather than abstract:** write "matte stone surface" instead of "stone surface".
{% endhint %}

This is the prompt after editing:&#x20;

***

Based on the attached image, please extract the complete set of visual prompts for a restored version; do not generate any images. Provide detailed description of the material of each item, and accurately reproduce the colors, layout, and materials.

*<mark style="color:red;">`The hair needs detailed description; it should have a voluminous texture, not clumpy strands. The skin texture should have realistic pores and grain, not be smooth and flawless. The woman vest is made of mesh.`</mark>*

The prompt must begin with this sentence: "Perform a perfect restoration based on the reference image."

***

{% endstep %}

{% step %}

#### Use your prompts on Beachside

Follow these steps:

1. Create a new scene.
2. Place the image you need to repair on the new scene.
3. Paste the prompt text generated in the AI ​​chat.
4. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

Here are the results after restoring each layer:

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FvgKAcOQdP10kzd01B8LD%2FGroup%2093.png?alt=media&amp;token=94aef70f-2ce0-4b76-b700-327f3aeba518" alt=""><figcaption></figcaption></figure>

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FPTN4lHDkej2MENEHfbYE%2FGroup%2094.png?alt=media&amp;token=3eea9fcc-50f0-45df-812c-9861dca8f618" alt=""><figcaption></figcaption></figure>

{% endstep %}

{% step %}

#### Blend the restored images together

Blend the restored background and the model using Combo

Follow these steps:

1. Create a new scene
2. Select <kbd><mark style="color:$primary;">**Combo**<mark style="color:$primary;"></kbd> task
3. Place the background image
4. Add an image sticker (image reference sticker) and upload the image of the woman. You must add hints to integrate the image seamlessly. You must be very concrete and clarify exactly what is the scene, for instance: *<mark style="color:red;background-color:red;">`Seamlessly integrate the person on the image.A woman is sitting on the carpet, gently leaning against the sofa while drying her hair. Keep the rest of the background completely unchanged`</mark>*
5. Add restriction prompts\
   Restrict the scope of modifications and add specifications regarding special materials to ensure the AI ​​preserves the object's material properties as much as possible.\
   Add a prompt to the Text Prompt box: *<mark style="color:red;background-color:red;">`Preserve the texture and high-resolution effect of the objects in the original background image. Preserve the authentic of the sofa and carpet`</mark>*
6. Add a reference image\
   Add the original image (before restoration happened) to guide the AI ​​regarding the positioning or color scheme you require. It is important to tell Beachside what this image is for by adding an explanation text: *<mark style="color:red;background-color:red;">`Refer to the position and size of the woman`</mark>*
7. Select precise mode
8. Generate

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F1J1TNmYDa54K5grIVgbK%2FGroup%2090.png?alt=media&amp;token=280a98d6-eed9-4788-b8f0-1ef66099261e" alt=""><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

### FAQ

<details>

<summary><strong>Is the step of "adding negative prompts" mandatory? What happens if I don't include them?</strong></summary>

Its function is to lock the scope of changes during the "Combo" step—by adding a prompt like "keep everything else unchanged and preserve the original textures and material qualities," you prevent the model from inadvertently regenerating the textures of other background elements. Without this instruction, you might intend to replace only the character, but end up having the textures of the sofa and carpet altered as well.

</details>

<details>

<summary><strong>Are the generated material prompts sometimes very long, with only a portion actually being executed?</strong></summary>

If you overload the prompt with too many minute details, the AI ​​will prioritize the most important elements, often overlooking descriptions placed later in the prompt or those that don't significantly alter the main visual. Once you receive the output from Gemini, streamline it yourself first: retain only the 3–5 material details that have the greatest impact on the image (such as hair volume or skin texture).

</details>

<details>

<summary><strong>The same portrait has been repeatedly repaired several times, and the details of the facial features seem to be slowly "going astray." How do I determine whether I should continue?</strong></summary>

With each round of generation, facial details are regenerated based on the results of the previous round rather than the original source image. After two rounds of refinement, compare the current result with the original face regarding facial feature proportions and the facial contour; if discrepancies are already apparent, do not continue refining this specific version. Instead, revert to the original source image and restart the process, rather than layering further corrections onto the current version.

</details>

***

## 3D Typography

Replicate typography effects using a reference image.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FYhxTFH5tyXMXGLHljCom%2F%E5%AD%97%E4%BD%93%EF%BC%92.png?alt=media&amp;token=11dd41fd-6ef6-49ca-b5ff-2d6ec8046410" alt=""><figcaption></figcaption></figure>

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FN6O3IKaG0EUHcxjahYQx%2F%E5%AD%97%E4%BD%93.png?alt=media&amp;token=a4dadf9f-6980-4d6d-9260-c49da211af0c" alt=""><figcaption></figcaption></figure>

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F4woZ0T166ghXw0gz9Z7I%2F%E5%AD%97%E4%BD%935.png?alt=media&amp;token=8750a6b0-fd24-4e11-89c2-bfafb936df2b" alt=""><figcaption></figcaption></figure>

### Interactive Tutorials

{% 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="GlAhr8vzz5VL3KNzBtjv" url="<https://app.arcade.software/share/GlAhr8vzz5VL3KNzBtjv>" %}

### How to

{% stepper %}
{% step %}

#### Use the "Text" tool to enter text

Follow these steps:

1. Create a new scene in Beachside
2. Add a solid color background
3. Select the <kbd><mark style="color:$primary;">**Text**<mark style="color:$primary;"></kbd> tool
4. Enter text
   {% endstep %}

{% step %}

#### Use the Combo tool to create 3D effects.

Follow these steps:

1. Select the <kbd><mark style="color:$primary;">**Combo**<mark style="color:$primary;"></kbd> tool
2. Upload visual reference
3. Annotation:*<mark style="color:red;background-color:$danger;">**Font and 3D effects**</mark>*
4. Submit

#### Edit the template

You can edit the *<mark style="color:red;background-color:$danger;">**red**</mark>**&#x20;***  prompts; for example, replace them with "typeface design," "3D visuals," "Font replacement effect"or other specific reference content.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FVy1DnbTTVOEn51MA3mRy%2FGroup%202.png?alt=media&amp;token=6c42cf59-1eb6-4587-bdde-f7b18c618269" alt=""><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}

### FAQ

<details>

<summary><strong>When applying a style, should the reference image use the original flat text or the final rendered 3D version?</strong></summary>

You must use the finished 3D render as the reference image. Information regarding materials and lighting exists only in the rendered output—not in the original flat text layer—so if you use the latter as a reference, the composition tool has no basis from which to "copy" those details.

</details>

<details>

<summary><strong>Can it be used for typeface design as well?</strong></summary>

Yes. The annotation text in the Combo step is fully editable — the red-marked part can say more than a material description like `Font effects` or `3D visuals`; you can write `typeface design` instead. When you do, Combo no longer just replicates the reference image's material and lighting — it replicates the letterforms themselves: stroke weight, connecting strokes, decorative details, applied wholesale to your text.

</details>

***

## Realistic blending of packshots

Use an AI chat tool to find anything that looks unrealistic. Then use its fix prompt to make the product look natural in the scene.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F0zsVgCZrs6jTpNETWQnf%2FSplit%20vision.png?alt=media&amp;token=89e7595f-a64f-4752-b40c-14df59c4a989" alt=""><figcaption></figcaption></figure>

### Interactive Tutorial

{% 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="EIgaio37Y4gamLkU2Dxd" url="<https://app.arcade.software/share/EIgaio37Y4gamLkU2Dxd>" %}

### How to

{% stepper %}
{% step %}

#### Use an AI chat platform to create your prompt

Please provide the image to your preferred AI chat platform (such as Gemini, ChatGPT, or Claude) to generate prompts for details that need optimization.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2FtZvraRX3mBi5HJLkZbBS%2FGroup%201.png?alt=media&amp;token=628c4694-1947-4ce4-ac46-d4728b7eb3b6" alt=""><figcaption></figcaption></figure>

Use only one image at a time, along with the following prompt:

***

Please analyze this image. \
Do not generate any images.

Find the elements where the product doesn't look like it was actually photographed. Check whether the product's reflections and refraction, angle and perspective, how the product's material behaves in the environment, and the lighting match a real photo shoot, and give detailed corrections for each point.

Without changing the background's angle, match the product's angle to the background's shooting angle.

Blend the whole scene naturally. When multiple products are placed together, make the lighting realistic and natural: light all products with the same key light as the background, but vary the brightness slightly according to each product's distance from the light source. Products closer to the light source should be slightly brighter, and those farther away slightly darker.

Give the foreground and background realistic focus.

Make glossy product surfaces realistically reflect the surrounding environment and objects.

*<mark style="color:red;background-color:$danger;">**Add any additional requests here**</mark>*

Output only one fix prompt that can be copied and used as is. It must start with "Repair based on the material I've provided" and end with "Keep the product's labels, text, structure, and other visual elements unchanged."

***

{% prompt description="Copy your prompt here" icon="message-lines" openInAIProviders="false" %}

```markdown
Please analyze this image. 
Do not generate any images.

Find the elements where the product doesn't look like it was actually photographed. Check whether the product's reflections and refraction, angle and perspective, how the product's material behaves in the environment, and the lighting match a real photo shoot, and give detailed corrections for each point.
Without changing the background's angle, match the product's angle to the background's shooting angle.
Blend the whole scene naturally. When multiple products are placed together, make the lighting realistic and natural: light all products with the same key light as the background, but vary the brightness slightly according to each product's distance from the light source. Products closer to the light source should be slightly brighter, and those farther away slightly darker.
Give the foreground and background realistic focus.

Make glossy product surfaces realistically reflect the surrounding environment and objects.

Add any additional requests here

Output only one fix prompt that can be copied and used as is. It must start with "Repair based on the material I've provided" and end with "Keep the product's labels, text, structure, and other visual elements unchanged."

```

{% endprompt %}
{% endstep %}

{% step %}

### Edit the prompt

You must edit the area marked in *<mark style="color:red;background-color:$danger;">**red**</mark>*. If no additional specific material requirements are needed, remember to remove "Add any other requirements here". The AI chat might not understand the materials correctly, read your prompt and add information as needed.

{% hint style="danger" %}
**Write your requests properly**\
It is very important to note that if you wish to control the output, you should write it clearly. For instance, if you would like to control the product glossiness or transparency you should write it down. Be concrete rather than abstract: write "do not add shiny edges on the product" instead of "no shine".
{% endhint %}

This is the prompt after editing:

***

Please analyze this image. \
Do not generate any images.

Find the elements where the product doesn't look like it was actually photographed. Check whether the product's reflections and refraction, angle and perspective, how the product's material behaves in the environment, and the lighting match a real photo shoot, and give detailed corrections for each point.

Without changing the background's angle, match the product's angle to the background's shooting angle.

Blend the whole scene naturally. When multiple products are placed together, make the lighting realistic and natural: light all products with the same key light as the background, but vary the brightness slightly according to each product's distance from the light source. Products closer to the light source should be slightly brighter, and those farther away slightly darker.

Give the foreground and background realistic focus.

Make glossy product surfaces realistically reflect the surrounding environment and objects.

<mark style="color:red;">`Do not change the perfume's left-right angle. Keep it facing straight to the front, but the up-down viewing angle may be adjusted.`</mark>

Output only one fix prompt that can be copied and used as is. It must start with "Repair based on the material I've provided" and end with "Keep the product's labels, text, structure, and other visual elements unchanged."

***

{% endstep %}

{% step %}

### Use your prompts on Beachside

Follow these steps:

1. Create a new scene.
2. Place the image you need to repair on the new scene.
3. Paste the prompt text generated in the AI ​​chat.
4. Make sure Nano Banana pro is the selected model.
5. Make sure [Precise Mode](https://docs.theseaai.com/tasks/model#generation-effort-for-nano-banana-pro) is selected.
6. After confirming all settings, select <kbd><mark style="color:$primary;">**Generate**<mark style="color:$primary;"></kbd>.

<figure><img src="https://3377749587-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FOUQ3EAPIPziQuBKQgmTy%2Fuploads%2F5GI8xe2zpj6qQ7ehoApd%2FGroup%2012.png?alt=media&amp;token=53d1b7cf-94b8-47ab-b2c6-6a91fb66ea3d" alt=""><figcaption></figcaption></figure>
{% endstep %}
{% endstepper %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

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

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.theseaai.com/learn/tutorials.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

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.