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

Your product photography budget just got cut by 40%. It's January, spring campaigns launch in six weeks, and your e-commerce team needs images in five different colour variations and two seasonal themes. Hiring a professional photographer costs £3,000 per shoot, and you need at least three shoots to cover everything. Alternatively, you could generate these variations from a single hero shot using AI model training and automated background swaps. This workflow uses specialised AI photography tools to train custom models on your existing product images, then generates dozens of variations without touching a camera. Each variation gets automatic background replacement, colour adjustments, and quality enhancement. Everything runs on a schedule; new campaign assets appear in your cloud storage folder every week. The real time save isn't in generation time, it's in the elimination of manual handoffs. No downloading files, no opening Photoshop, no exporting and uploading. Just a scheduled automation that takes your source images, processes them, and deposits finished assets directly into your product management system.

The Automated Workflow

We'll use n8n as the orchestration backbone because it handles conditional logic, file processing, and API calls well. The workflow runs weekly and does the following: polls your source image folder, trains a model in PickOrCraft if new images exist, generates variations with different aesthetics, swaps backgrounds using Pixelcut AI, enhances the final output with SellerPic AI, then uploads everything to your e-commerce platform.

Step 1: Trigger on New Source Images

Set up n8n to monitor a cloud storage folder (Google Drive, Dropbox, or AWS S3) for new product images. This trigger fires whenever you upload a new hero shot.

CopyTrigger: Folder Poll
Path: /product-photography/source
Check Interval: Daily (Monday, 08:00 UTC)
File Types: .jpg, .png

Step 2: Train a Custom Model in PickOrCraft

PickOrCraft's API accepts your source images and trains a model in roughly 30 minutes. You send the training request with your image URLs and specify the aesthetic style you want (e.g. "bright white studio background", "minimal lifestyle", "luxury packaging display").

CopyPOST https://api.pickorcraft.com/v1/models/train
Content-Type: application/json { "api_key": "your_api_key_here", "model_name": "spring_2026_product_line", "training_images": [ "https://storage.example.com/source/product_hero_01.jpg", "https://storage.example.com/source/product_hero_02.jpg", "https://storage.example.com/source/product_hero_03.jpg" ], "aesthetic": "studio_bright", "training_duration": "standard"
}

n8n will poll the training status endpoint every 5 minutes until the model is ready. Once complete, you'll get a model_id that you use in the next step.

CopyGET https://api.pickorcraft.com/v1/models/{model_id}/status

Step 3: Generate Variations with Different Aesthetics

Now use the trained model to generate variations. Create four variations per source image: one for spring campaigns, one for summer, one with a white background, and one with a lifestyle setting.

CopyPOST https://api.pickorcraft.com/v1/generate
Content-Type: application/json { "api_key": "your_api_key_here", "model_id": "spring_2026_product_line", "prompt": "product photography, spring collection, natural lighting, green botanical background", "num_variations": 4, "resolution": "1024x1024", "output_format": "jpg"
}

Each generated image gets saved to a temporary n8n variable. Don't download yet; you'll batch process these next.

Step 4: Swap Backgrounds with Pixelcut AI

Take each generated variation and use Pixelcut's background removal and replacement API. You can either remove backgrounds entirely or swap them for branded backdrops. Pixelcut's API is simpler than most; you send the image URL and a background specification.

CopyPOST https://api.pixelcut.ai/api/v1/edit
Content-Type: application/json { "api_key": "your_api_key", "image_url": "https://temp-storage.example.com/generated_variation_01.jpg", "operation": "replace_background", "background_type": "colour", "background_value": "#FFFFFF"
}

The response includes a download URL for the processed image. n8n stores this URL in an array for bulk download later.

Step 5: Enhance with SellerPic AI

Run each background-swapped image through SellerPic AI for final enhancement: colour correction, sharpness, contrast adjustment, and automatic product optimisation for e-commerce platforms.

CopyPOST https://api.sellerpic.ai/enhance
Content-Type: application/json { "api_key": "your_api_key", "image_url": "https://pixelcut-output.example.com/background_swapped_01.jpg", "enhancement_profile": "ecommerce_standard", "auto_adjust": true, "output_size": "product_thumbnail"
}

Step 6: Bulk Download and Upload to E-commerce Platform

Once all images are processed, n8n downloads each one concurrently (set worker threads to 4 or 5 to avoid rate limits) and uploads them directly to your e-commerce platform via API. Most platforms (Shopify, WooCommerce, custom APIs) accept batch image uploads.

CopyPOST https://api.shopify.example.com/graphql
Content-Type: application/json { "query": "mutation { productImageCreate(input: {productId: \"gid://shopify/Product/123\", image: {src: \"https://final-storage.example.com/spring_variation_01.jpg\"}}) { image { id src } } }"
}

Alternatively, use SFTP or AWS S3 to drop the files into a monitored folder that your e-commerce system ingests automatically. n8n Workflow Diagram The workflow looks like this: 1. Folder Poll trigger (Google Drive) 2. Format URLs for PickOrCraft 3. Train model (wait for completion) 4. Generate 4 variations 5. Loop: Pixelcut background swap for each variation 6. Loop: SellerPic enhancement for each variation 7. Batch download all processed images 8. Upload to Shopify/WooCommerce/custom API 9. Log completion and send Slack notification

The Manual Alternative

If you prefer control over the timing or want to cherry-pick which images get processed, you can use the same tools without orchestration. Train your model in PickOrCraft's web interface, generate variations manually, then download and process them in AI Boost or Pixelcut's apps. This takes about 45 minutes per batch but gives you more creative oversight and the ability to tweak prompts between runs. AI Boost is particularly useful for this manual approach because it consolidates face swapping, background changes, and body reshaping into a single interface. You can upload a generated variation, adjust the background interactively, and save the result without touching another tool.

Pro Tips

Monitor PickOrCraft training failures.

Model training sometimes fails on images with extreme aspect ratios or very low resolution.

Add a conditional check in n8n: if training doesn't complete within 45 minutes, trigger an alert and skip that batch. Don't let a single bad image hang your entire workflow.

Rate limit Pixelcut and SellerPic.

Both services enforce rate limits (typically 10-20 requests per minute for standard plans). Build in 3-5 second delays between API calls, and use n8n's throttle node to manage request queues. You'll avoid 429 errors and won't need to retry failed calls.

Resize generated images before uploading.

PickOrCraft generates at 1024x1024 by default, but most e-commerce platforms want multiple sizes. Add an image resize step before uploading: create 400x400 thumbnails, 800x800 for product pages, and 1600x1600 for detail views. This adds 2 minutes to the workflow but prevents platform-side resizing, which degrades quality.

Test with two source images first.

Before scheduling this for your full product catalogue, run it on two sample images and manually inspect the outputs. Check that PickOrCraft's aesthetic matches your brand, that Pixelcut's background swaps look natural, and that SellerPic's enhancement doesn't over-saturate colours.

Store model IDs in a config file.

If you train multiple models (one per product category), store the model IDs and aesthetic preferences in a JSON config that n8n reads. This lets you manage model versions without editing the workflow itself.

Set up a quarantine folder for failed images.

If any image fails processing, have n8n move it to a separate folder for manual review rather than letting it block the entire batch. This keeps your workflow resilient.

Cost Breakdown