How to Generate Product Images in Bulk with AI

Last week I generated 47 product lifestyle images for a client's Shopify store in one sitting. Different backgrounds, different angles, different moods. The source material was a single product photo on a white background.

No Photoshop. No photographer. No waiting three days for an agency to deliver assets. I built a flow in PlugNode, hit Run, and had publishable images in minutes.

Here's how to set up the same thing for your store.

What this gets you

• One product photo becomes 10, 20, or 50 variants with different backgrounds and contexts.
• Each image sized for the platform you need: Shopify listings, Instagram posts, TikTok ads, Google Shopping.
• A repeatable flow you can run every time you add new products.
• Total hands-on time: about 5 minutes to build the flow, then seconds per batch.

Who this is for

E-commerce teams tired of paying for product photography every time they need a new angle. Fashion brands that want lifestyle shots without booking a studio. Agencies juggling content for multiple clients who all need "more product images, yesterday."

If you sell physical products online, this saves you hours per week.

What you need

1. A PlugNode account.
2. A Gemini API key (paste it in Settings once, every node uses it automatically).
3. Your product photos. White background works best, but any clean product shot will do.

Step 1: Start a new flow and add your product image

Open PlugNode and create a new flow. Drag a File Input node onto the canvas. This is where your product photo goes.

Upload your image directly, or paste a URL if your photos live on Shopify or a CDN already. The file input accepts PNG, JPG, and WebP.

I usually start with the highest resolution product photo I have. The AI models produce better results when they can see product details clearly.

Step 2: Add a Text node to write image descriptions

Drag a Text node onto the canvas. Connect the File Input's output to it.

This node does something specific: it looks at your product photo and writes a detailed description of what the product looks like. Color, material, shape, size relative to surroundings. The description becomes the foundation for generating variants.

Set the system prompt to something like: "Describe this product in detail. Include color, material, shape, texture, and any distinctive features. Write it as an image generation prompt. Keep it under 80 words."

Why does this step matter? Because raw image generation from a photo alone gives you less control. When the AI model has both the photo AND a text description, you get more accurate variants that look like your product.

I tested this with a leather handbag. Without the text description step, the generated variants sometimes changed the bag's proportions or hardware color. With the description step, accuracy improved noticeably across 20 test generations.

Step 3: Add the Image node for generation

Drag an Image node onto the canvas. Connect the Text node's output to the Image node's prompt input. Also connect the original File Input to the Image node's reference image input.

Now configure the Image node:

• Pick your model: Gemini 2.5 Flash Image or Nano Banana both work well for product shots.
• Set the aspect ratio based on your target platform. 1:1 for Instagram and Shopify thumbnails, 4:5 for Instagram feed posts, 9:16 for Stories and TikTok.
• In the prompt field, add your context. Something like: "Product photography of [this item] on a wooden kitchen counter, warm morning light from the left, shallow depth of field, lifestyle setting."

The combination of reference image plus detailed text prompt gives you variants that keep your product accurate while placing it in completely different contexts.

Step 4: Generate multiple variants

Here's where bulk generation gets interesting. You have two options.

Option A: Multiple Image nodes with different prompts. Duplicate the Image node 3 or 4 times. Give each one a different scene description. "On a marble countertop, studio lighting." "Outdoors on a cafe table, natural light." "Flat lay on white linen, overhead shot." Run the flow once and get all variants at the same time.

Option B: Run the same flow multiple times with different prompts. Change the context in the Text node's system prompt each time you run. This is faster to set up but requires more manual runs.

I prefer Option A for client work. I set up 5 or 6 Image nodes with different scene descriptions, run the flow once, and get a full set of lifestyle variants in one batch.

Step 5: Resize for every platform

Drag an Image Resize node after each Image generation node. Configure it for the exact dimensions your platforms need:

• Shopify product images: 2048x2048
• Instagram feed: 1080x1350
• Instagram Stories / TikTok: 1080x1920
• Google Shopping: 800x800 minimum
• Facebook ads: 1200x628

One generation, multiple crops. No opening Photoshop to manually export sizes.

Step 6: Run and review

Click Run. Watch each node execute in sequence. The execution log shows you what happened at every step: the description the Text node wrote, the image each generation node produced, and the resized outputs.

For a flow with one product photo and 5 scene variants, each with 3 resize outputs, I typically see results in under 90 seconds. That's 15 ready-to-publish images from one click.

Review the outputs. If one variant missed the mark (wrong lighting mood, product positioned oddly), adjust that specific Image node's prompt and run again. The other nodes keep their cached results.

Making it repeatable

Once your flow works for one product, it works for every product. Swap the image in the File Input node, hit Run, and get a fresh batch. Same scenes, same sizing, same quality.

For stores with large catalogs, this turns a 3-day photography project into an afternoon task.

Real numbers from a fashion brand project

I built this flow for a women's accessories brand with 120 SKUs. They needed 5 lifestyle images per product for their new Shopify theme.

Old process: hire a photographer, book a studio, style each product, shoot, edit, crop. Timeline was 3 weeks and the budget was around $4,500.

New process: upload each product photo, run the flow, review outputs, publish. I completed all 120 products in 2 days. API costs came to roughly $85 total using Gemini's published rates with their own key.

Not every generated image was perfect. About 15% needed a re-run with adjusted prompts. But even with that, the total time was under 16 hours versus three weeks.

Tips from running this 500+ times

White backgrounds produce the best source material. The AI model can isolate the product cleanly and place it in new contexts without artifacts.

Be specific about lighting in your scene prompts. "Warm morning light from the left" gives you consistent, usable results. "Good lighting" gives you random results.

Match your brand aesthetic in the prompts. If your brand is minimal and clean, say "minimalist setting, neutral tones, clean background." If your brand is bold and colorful, describe that instead.

Save your best prompts. When a scene description produces great results, copy it. Use it as a template for future products.

Check product accuracy. AI models sometimes alter small details like stitching patterns, hardware colors, or logo placement. Always compare the generated image to the original product photo before publishing.

Cost breakdown

At Gemini's current pricing with your own API key:

• Text description (Gemini 2.5 Flash): roughly $0.0001 per call
• Image generation (Gemini 2.5 Flash Image): see Google's published per-image rates
• Image resize: free (runs on PlugNode's server, no API call)

For a batch of 50 product images with 5 variants each (250 generated images), you're looking at a few dollars in API costs. Compare that to a photographer's day rate.

You pay your AI providers directly through your own keys. No markup from PlugNode. Check your Google AI Studio dashboard to track exact spend.

FAQ

Can I use this for video ads too?

Yes. Replace the Image node with a Video node. PlugNode's Video node calls Gemini Veo and produces clips up to 8 seconds. Same flow structure, different output type. Video generation takes longer (30-60 seconds per clip) but the process is identical.

What if my product photos aren't on white backgrounds?

It still works, but results are less consistent. The AI model has to separate the product from its existing background before placing it in a new scene. For best results, use a background removal tool first, or upload photos with clean, simple backgrounds.

Can I connect this directly to my Shopify store?

Yes. Publish your flow as an API endpoint (one click in PlugNode). Then use Shopify Flow or a Zapier integration to call your PlugNode endpoint whenever a new product is created. Images generate automatically and you can push them back to your product listing.

How does image quality compare to real photography?

For product lifestyle shots and social media content, AI-generated images are production-ready. For hero images on your homepage or print catalogs where every pixel matters, you may still want professional photography. The sweet spot is using AI for volume (social content, ad variants, listing images) and photography for hero assets.

Do I keep full rights to the generated images?

You own the outputs. Gemini's terms grant commercial usage rights for generated images. Check your specific provider's terms for confirmation, but standard commercial use is covered.

For more on how PlugNode compares to other visual workflow tools, see Top 7 AI Workflow Builders. For the BYOK pricing comparison, see 5 AI Pipeline Tools That Let You BYOK. For the same publish-as-API pattern applied to cinematic video, see Cinematic AI Video, Then an API.