Product photography with AI model customisation

Creating product photography that showcases your items at their best is expensive and time-consuming. You need a photographer, a studio, styling expertise, and then editing on top of that. But what if you could generate product images automatically, then customise them with AI-driven adjustments tailored to your brand's aesthetic?...... For more on this, see Product photography studio with AI model training and mar....

This is where combining AI-Boost, Nsketch-AI, and Pickorcraft becomes genuinely useful. Instead of hiring a photographer for each product variant or spending hours in Photoshop, you can build an automated workflow that takes a product description, generates multiple photography styles, applies custom model training to match your brand look, and delivers final images ready for your e-commerce platform. All without touching a single tool manually....... For more on this, see E-commerce product description and image enhancement from....

The workflow works because each tool has a specific job: AI-Boost generates the initial product images from text descriptions; Nsketch-AI applies style customisation based on reference images you provide; Pickorcraft handles the final composition and background work. By wiring these together through an orchestration platform, you can feed in product data once and receive a complete, customised image set at the other end.

The Automated Workflow

Which Orchestration Tool to Use

For this workflow, I'd recommend n8n if you want to self-host and have full control over your data, or Zapier if you prefer a managed solution with minimal setup. Make (Integromat) also works well here. I'll walk through the n8n approach since it offers the best balance of flexibility and cost for image generation pipelines. The workflow uses webhooks to trigger on new product data, then chains the three AI tools together.

The Complete Workflow Architecture

Here's how data flows:

1. Product data arrives via webhook (a new item in your CSV, a Shopify product, or a form submission).

2. AI-Boost receives the product description and generates 3-4 base product photographs in different styles.

3. The generated image URLs are passed to Nsketch-AI, which applies your brand's visual model (trained on your existing product photos).

4. Pickorcraft receives the styled images and applies final composition, background removal, or custom backgrounds.

5. The final images are saved to your cloud storage and the URLs are sent back to your database or e-commerce platform.

Setting Up the Webhook Trigger in n8n

Create a webhook node in n8n that listens for incoming product data. This is your entry point.

Copy
POST https://your-n8n-instance.com/webhook/product-photography

Example payload:
{
  "product_id": "SKU-12345",
  "product_name": "Ceramic Coffee Mug",
  "description": "Handmade ceramic mug, white glaze, ergonomic handle, 350ml capacity",
  "category": "homeware",
  "reference_image_url": "https://your-cdn.com/reference-mugs.jpg"
}

In n8n, your webhook node will parse this incoming JSON automatically. You'll then map the fields to use in subsequent nodes.

Step 1: Generate Initial Images with AI-Boost

The AI-Boost API takes your product description and generates multiple image variations. You'll need an API key from your AI-Boost account.

Copy
POST https://api.ai-boost.com/v1/generate

Headers:
Authorization: Bearer YOUR_AI_BOOST_API_KEY
Content-Type: application/json

Body:
{
  "prompt": "Professional product photo of a white ceramic coffee mug, 350ml, ergonomic handle, studio lighting, white background, high detail, commercial photography style",
  "num_images": 4,
  "style": "product_photography",
  "resolution": "1024x1024",
  "quality": "high"
}

Response:
{
  "request_id": "req_abc123xyz",
  "images": [
    {
      "url": "https://cdn.ai-boost.com/images/req_abc123xyz_1.jpg",
      "style_applied": "studio_clean"
    },
    {
      "url": "https://cdn.ai-boost.com/images/req_abc123xyz_2.jpg",
      "style_applied": "lifestyle"
    },
    {
      "url": "https://cdn.ai-boost.com/images/req_abc123xyz_3.jpg",
      "style_applied": "flat_lay"
    },
    {
      "url": "https://cdn.ai-boost.com/images/req_abc123xyz_4.jpg",
      "style_applied": "detail_shot"
    }
  ]
}

In n8n, use an HTTP Request node set to POST. Map the product description field from your webhook into the prompt. The response gives you four image URLs to feed into the next step.

Step 2: Apply Brand Customisation with Nsketch-AI

Nsketch-AI trains on reference images you've provided and applies that style to new images. This is where your brand's unique aesthetic gets baked in.

Before running the workflow, you'll have created a "style model" in Nsketch-AI by uploading 5-10 examples of your products photographed in your preferred style. The API returns a model_id you'll use here.

Copy
POST https://api.nsketch-ai.com/v1/stylise

Headers:
Authorization: Bearer YOUR_NSKETCH_API_KEY
Content-Type: application/json

Body:
{
  "input_image_url": "https://cdn.ai-boost.com/images/req_abc123xyz_1.jpg",
  "model_id": "model_your_brand_style_v2",
  "strength": 0.85,
  "preserve_subject": true
}

Response:
{
  "stylised_image_url": "https://cdn.nsketch-ai.com/output/st_def456uvw_1.jpg",
  "processing_time_ms": 2100,
  "model_applied": "model_your_brand_style_v2"
}

In n8n, you'll need a Loop node here. Loop through each of the four images returned by AI-Boost, and run this HTTP Request for each one. The loop will generate four stylised images, all matching your brand's visual identity.

The strength parameter controls how much the brand style influences the output. Set it to 0.85 for most cases; go lower (0.6) if you want to preserve more of the original product detail.

Step 3: Final Composition with Pickorcraft

Pickorcraft handles the final details: background removal, custom background insertion, or composition adjustments. This step ensures all images meet your platform's specifications.

Copy
POST https://api.pickorcraft.com/v1/compose

Headers:
Authorization: Bearer YOUR_PICKORCRAFT_API_KEY
Content-Type: application/json

Body:
{
  "image_url": "https://cdn.nsketch-ai.com/output/st_def456uvw_1.jpg",
  "background_action": "remove_and_replace",
  "background_colour": "#f5f5f5",
  "output_format": "png",
  "output_width": 1200,
  "output_height": 1200,
  "auto_crop": true
}

Response:
{
  "composed_image_url": "https://cdn.pickorcraft.com/final/pc_ghi789jkl_1.png",
  "background_removed": true,
  "dimensions": {
    "width": 1200,
    "height": 1200
  },
  "file_size_kb": 245
}

Again, use a Loop node to process all four stylised images. After this step, you have four finished product photographs, all customised to your brand and ready to upload.

Step 4: Save and Send Back Data

Use a node to save the final image URLs to your database or cloud storage (Google Drive, AWS S3, Airtable, etc.). Then send a response back to your e-commerce platform or trigger a notification.

Copy
POST https://your-api.com/products/SKU-12345/images

Body:
{
  "product_id": "SKU-12345",
  "final_images": [
    "https://cdn.pickorcraft.com/final/pc_ghi789jkl_1.png",
    "https://cdn.pickorcraft.com/final/pc_ghi789jkl_2.png",
    "https://cdn.pickorcraft.com/final/pc_ghi789jkl_3.png",
    "https://cdn.pickorcraft.com/final/pc_ghi789jkl_4.png"
  ],
  "generated_at": "2025-01-15T14:23:45Z",
  "workflow_id": "wf_product_photography_v1"
}

Complete n8n Workflow Structure

The n8n canvas would look like this:

1. Webhook trigger (listens for incoming product data)

2. HTTP Request to AI-Boost (generates initial images)

3. Loop through AI-Boost response

4. HTTP Request to Nsketch-AI (applies brand style) - inside the loop

5. Loop through Nsketch-AI response

6. HTTP Request to Pickorcraft (final composition) - inside the nested loop

7. Collect all final image URLs

8. HTTP Request to save results back to your system

9. Send success notification (email or Slack)

This architecture means every product you send in gets four unique, customised images without any manual intervention between steps.

The Manual Alternative

If you prefer more control over individual steps, you can run these tools separately. Upload your product CSV to AI-Boost directly through their web interface and download the generated images. Then manually upload those images to Nsketch-AI, select your brand style model, and process each one. Finally, open the stylised images in Pickorcraft and adjust backgrounds as needed.

This takes roughly 10-15 minutes per product instead of 30 seconds. It's useful if you're doing a one-off project or want to cherry-pick specific variations, but it doesn't scale. Once you're processing more than 5-10 products per week, the workflow automation pays for itself in time alone.

Pro Tips

Rate Limiting and Batch Processing

AI-Boost and Nsketch-AI have rate limits; typically 10-30 requests per minute depending on your plan. If you're processing 100 products at once, n8n's built-in Rate Limit node will prevent you from hitting those limits and getting blocked.

Copy
Use n8n's Rate Limit node set to 10 requests per 60 seconds, placed before the AI-Boost HTTP Request node.

This way, your workflow processes products sequentially rather than all at once, respecting API limits.

Error Handling and Retries

Image generation sometimes fails: the API times out, a reference image is too low quality, or a background removal doesn't work cleanly. Build in retry logic.

Copy
In n8n, use the Error Handling tab on each HTTP Request node.
Set: On Error = Retry with Backoff
Max retries = 3
Backoff multiplier = 2 (waits 2s, then 4s, then 8s)

If a single product image fails, the workflow moves that product to a "review" queue rather than blocking the entire batch.

Cost Optimisation

Don't generate four images per product if you only need two. Adjust the num_images parameter in AI-Boost based on your needs. If you're doing bulk runs during off-peak hours, some providers offer cheaper batch processing rates.

Testing on Small Batches First

Before running this workflow on 500 products, test it on 5. You'll catch issues with your prompt wording, your brand style model strength, or background colour mismatches before spending money on the full batch.

Monitoring and Logging

Set n8n to log every API response. If an image comes out with artefacts or wrong colours, you'll want to see exactly what parameters were sent and what came back. Use n8n's built-in Webhook Response node to return detailed logs alongside your final images.

Cost Breakdown

If you're processing 1,000 product images per month (250 products × 4 images each), your cost per product is about 10-13p. A professional product photographer typically charges £50-£150 per product. The break-even point is around product 20-30.

For small businesses running this workflow, the ROI usually appears within the first month.