Create AI-generated product photography variations with background swaps and model training

Your e-commerce business ships ten products monthly, each requiring photos against five different backgrounds: minimalist white, lifestyle kitchen, flat lay, product detail, and seasonal backdrop. Hiring a photographer for each variant costs £800 per shoot. Stock image subscriptions run £200 a month and still don't match your exact product colours. Most teams accept this as a cost of doing business. But here's the thing: you can generate dozens of product photo variations, retrain them for brand consistency, and swap backgrounds entirely using AI. The entire process can run automatically when you upload a new product image, delivering variations within minutes instead of weeks. No photographer scheduling. No stock image hunting. No manual touchpoints. This workflow combines three purpose-built tools: PickOrCraft for training a custom model on your existing product photography style, AI Boost for background removal and replacement, and Pixelcut AI for final polish and adjustment. Tied together with n8n, the pipeline becomes a single production line where photos enter one end and multiple, branded product variations exit the other.

The Automated Workflow

Choose n8n for this workflow instead of Zapier or Make. Why? N8n handles file uploads and local storage better, supports webhook triggers without hitting rate limits as quickly, and lets you run processes on a schedule. The flow works like this: a product image lands in a shared folder (either cloud storage or webhook trigger), gets processed through PickOrCraft's model training endpoint, then branches into parallel background variations using AI Boost, and finally receives touch-ups via Pixelcut AI before landing in your final output folder.

Step 1: Set up the webhook trigger in n8n

When a user uploads a product photo to your system (perhaps via a form or a shared cloud folder), that triggers the workflow. Use n8n's webhook node to catch the upload event.

CopyPOST https://your-n8n-instance.com/webhook/product-photo-upload { "product_id": "SKU-12345", "image_url": "https://your-storage.com/uploads/product-raw.jpg", "product_name": "Ceramic Mug Blue", "background_preferences": ["white", "kitchen", "flat_lay"]
}

Step 2: Call PickOrCraft's model training endpoint

PickOrCraft lets you train a custom model on your existing product photography. You'll do this once per product category (mugs, furniture, clothing) rather than per individual item. After training completes, you get back a model ID that you use for all subsequent variations of products in that category.

CopyPOST https://api.pickorcraft.com/v1/models/train { "api_key": "your_pickorcraft_api_key", "model_name": "ceramic_mugs_branded_style", "training_images": [ "https://your-storage.com/training/mug-1.jpg", "https://your-storage.com/training/mug-2.jpg", "https://your-storage.com/training/mug-3.jpg" ], "style_description": "bright natural light, product centred, minimalist composition"
}

PickOrCraft responds with a status URL. Use n8n's polling node to wait for training to complete (this usually takes 5-10 minutes). Store the returned model_id in n8n's variable storage.

CopyGET https://api.pickorcraft.com/v1/models/{model_id}/status?api_key=your_pickorcraft_api_key

Step 3: Generate variations using the trained model

Once the model trains, call PickOrCraft's generation endpoint to create variations of your product. Set it to generate multiple angles or slight compositional tweaks.

CopyPOST https://api.pickorcraft.com/v1/generate { "api_key": "your_pickorcraft_api_key", "model_id": "trained_model_xyz", "prompt": "ceramic mug, product photography, bright natural lighting", "num_variations": 3, "output_format": "jpg"
}

Step 4: Remove and swap backgrounds with AI Boost

For each generated variation, pass the image to AI Boost's background removal and replacement endpoint. AI Boost handles both steps in one call. You'll need to prepare your background images beforehand (white backdrop, kitchen setting, etc.) and store them in your cloud storage.

CopyPOST https://api.aiboost.com/v1/background-swap { "api_key": "your_aiboost_api_key", "image_url": "https://your-storage.com/generated/mug-variation-1.jpg", "background_image_url": "https://your-storage.com/backgrounds/white-studio.jpg", "edge_refinement": "high", "output_format": "png"
}

In n8n, add a Loop node here. For each background preference submitted in the initial webhook payload, make a separate API call to AI Boost. This means a single product photo becomes three (or five) different variations, one for each background style.

Step 5: Polish with Pixelcut AI

Once backgrounds are swapped, pass each variation to Pixelcut AI for final retouching. Pixelcut excels at colour correction, shadow adjustment, and ensuring product edges look natural after the background swap.

CopyPOST https://api.pixelcut.ai/v1/enhance { "api_key": "your_pixelcut_api_key", "image_url": "https://your-storage.com/swapped/mug-white-bg.jpg", "enhancement_type": "product_photo", "auto_color_correct": true, "shadow_depth": 2, "output_resolution": "2048x2048"
}

Step 6: Save outputs and notify completion

Store all final variations in your product management system or cloud folder. Use n8n's file storage node to save images locally or push them directly to Google Drive, Dropbox, or AWS S3. Include metadata (product ID, background type, generation timestamp) alongside each image. Finally, send a notification (email, Slack, webhook back to your e-commerce platform) confirming that variations are ready. Include download links or a folder location.

CopyPOST https://hooks.slack.com/services/YOUR/WEBHOOK/URL { "text": "Product variations ready", "attachments": [ { "title": "Ceramic Mug Blue - 5 variations generated", "text": "White, kitchen, flat lay, detail, seasonal backgrounds", "fields": [ { "title": "Output Folder", "value": "https://your-storage.com/output/SKU-12345/" } ] } ]
}

The Manual Alternative

If you prefer more control over composition and styling, skip PickOrCraft's training step. Instead, upload your product photo directly to AI Boost and manually select from available background templates. Pixelcut AI has a web interface where you can adjust colour correction and shadow depth before saving. This approach takes 10-15 minutes per product instead of two minutes, but gives you finer artistic control. For teams with a dedicated e-commerce designer, this hybrid approach often works better: use the automated workflow for bulk variations, then the manual interface for hero shots or seasonal campaigns.

Pro Tips

Rate limiting and queuing:

PickOrCraft's training endpoint has a queue system; you're typically processed within 30 minutes even during peak hours, but the API documentation specifies a limit of two concurrent training jobs per account.

In n8n, set up a Queue node before the training step so multiple products don't pile up simultaneously. This prevents failed requests from exhausted API limits.

Background image library:

Create a folder in your cloud storage with pre-edited background images at least 2048x2048 pixels. Store them with consistent naming (white-studio-v1.jpg, kitchen-lifestyle-v1.jpg) so your n8n workflow can reference them reliably. Test each background once before scaling; some backgrounds with heavy textures cause AI Boost to struggle with edge detection.

Cost optimisation:

Train your model once per product category, not per product. A category might be "stoneware mugs" or "dining tables"; reuse that trained model for all similar products in that category. This cuts API costs by 70-80% compared to training individual models. Store the model ID in a spreadsheet so future uploads can reference it immediately instead of retraining.

Error handling for image quality:

Before passing images to AI Boost, add a validation node in n8n that checks resolution (minimum 1200 pixels on the shortest edge) and file size (under 10MB). Low-resolution source images produce blurry background swaps. If an image fails validation, log it and send a manual review request instead of failing the entire workflow.

Batch processing on schedule:

Instead of triggering on individual uploads, set the workflow to run nightly on all products uploaded that day. This consolidates API calls and gives you bulk processing rates. Many vendors, including PickOrCraft, discount volume, so batching ten products together often costs less than processing them separately throughout the day.

Cost Breakdown