E-commerce Images in Practice: From White-Background Images to Product Videos
💡 Goal: Finish all five lessons in about 60 minutes — cut out a transparent-background product image, upscale it to platform resolution, generate virtual try-on model images, apply your design to physical products, and bring product images to life (model download time not included).
Let's get one thing straight first: the official docs don't have a dedicated "e-commerce images" tutorial — capabilities like background removal, upscaling, try-on, mockups, and video are scattered across various model and tool tutorials. This ComfyUI E-commerce Images in Practice series strings them into five lessons organized by e-commerce tasks. Each lesson starts from an official workflow template — you just swap in your own product images and tweak the prompts.
🎯 Stage Goals
- Cut a transparent-background PNG from product photos: local BiRefNet outputs an RGBA transparent background + mask, or use Qwen-Image-2.1 to directly generate a transparent-background image
- Upscale product images to platform-required resolution: conservative upscaling to preserve detail, or Z-Image-Turbo 2K upscaling to boost quality
- Produce wearing shots without hiring models: AI virtual try-on, hand-held product images
- Apply designs to physical products: mockup images for packaging, mugs, and the like
- Bring product images to life: image-to-video output for hero-image video material
⏳ Estimated Time
- Each lesson takes about 12 minutes, about 60 minutes in total; with buffer time, about 40–80 minutes
- Model download time not included: the try-on lesson runs in the cloud with no model download, while the other four lessons run locally and each requires downloading the corresponding model — extra time depends on your network speed
- Just need to fix one image right now? Use the routes below to jump straight to the relevant lesson — no need to go in order
🚀 Before You Start
- Read Models × Tasks Overview: know how to pick models by task and load official workflow templates (haven't installed ComfyUI yet? Go back to Quick Start first)
- Have product photos on hand: photos with even lighting and clean edges give the most stable results for background removal and upscaling
- Going the local route: make sure your GPU has enough VRAM — each lesson states its requirements at the start
- Going the cloud route (Bria / Kling): have Comfy account credits ready, as cloud nodes are billed by credits
- A cup of water ☕ — perfect for a sip while waiting for model downloads
💡 Every lesson starts from an official template
The workflows used in the five lessons can all be found in the template library: open Workflow → Browse Templates and search for the corresponding name to load it; a popup will prompt you to download any missing models. See the Template Library lesson for more on the template library.
📚 Choose Your Route
First, a multiple-choice question: should this batch of work go local or cloud?
| Local Workflows | Cloud Partner Nodes | |
|---|---|---|
| Representative | BiRefNet, Z-Image-Turbo, Qwen-Image-2.1, FLUX.2, Wan2.2 | Bria, Kling |
| Where it runs | Your own computer | Comfy official cloud API |
| Cost | Models downloaded by you, generation not billed per image | Requires Comfy account credits |
| Best for | Batch image production, keeping product images in-house | Tight VRAM, want to keep things simple |
Now match your task to a route:
Route One: Listing Images — White Background, Transparent Background, and High-Res Upscaling
Platforms often require white-background images, transparent-background PNGs, or minimum resolutions when uploading. The two steps below produce an AI product image ready for listing:
- Product Background Removal and Replacement (12 minutes): local BiRefNet cuts out an RGBA transparent background + mask, excelling at product edges; or use Qwen-Image-2.1 to directly generate a transparent-background image
- High-Resolution Product Image Upscaling (12 minutes): conservative upscaling to preserve product detail, or Z-Image-Turbo 2K upscaling to make up the resolution; if you don't want to use local VRAM, the same lesson also covers cloud upscaling (Magnific Precise, Nano Banana Pro)
Route Two: Marketing Scene Images — Background Replacement, Material Replacement, and Mockups
Click-through rates for AI e-commerce hero images and the mood images on detail pages all depend on scenes. Pick from three tools as needed:
- Background replacement: there's no official workflow for "one-click new background" locally — the route is BiRefNet cutout + compositing a new background; to replace the background directly with a single sentence, use the Bria cloud API (Product Background Removal and Replacement, 12 minutes)
- Material replacement: for needs like turning a leather sofa into fur, use the Qwen-Image-Edit-2511 material replacement workflow, see Qwen-Image-Edit Instruction-Based Editing
- Mockups: to apply designs to physical items like packaging and mugs, use the FLUX.2 Dev Product Mockup workflow, which can also lock in brand colors with hex codes (Product Mockups, 12 minutes; for model details see FLUX.2)
Route Three: Apparel and Wearing Shots — Virtual Try-On and Hand-Held Product Images
Selling clothing, bags, or accessories? Flat-lay shots alone aren't enough, and hiring models for shoots is expensive:
- Virtual Try-On and Hand-Held Product Images (12 minutes): Bria virtual try-on / hand-held product images, Kling try-on — "dress" the product onto a person. Tip: portraits with empty hands work best, and prepare 1–3 images of the same product from multiple angles
Route Four: Hero-Image Videos — Bringing Product Images to Life
- Product Video (12 minutes): Use Wan2.2 image-to-video to bring product images to life; the prerequisite is Wan2.2 14B Image-to-Video
- Want videos with sound? LTX-2 can generate audio-visual synchronized videos in one go, see LTX-2/2.3
- Already have live-action videos and need to change the background or remove the green screen? Bria cloud offers a video background replacement series, with input videos no longer than 60 seconds
🎒 What You Will Gain
- A complete AI product image production pipeline: background removal → upscaling → scene images → model images → video, with an official workflow template for every step that loads in one click
- A feel for choosing between local and cloud: which tasks to run locally without per-image billing, and which to hand off to cloud APIs to save VRAM
- An editing mindset of "change only the background, leave the product untouched": when to use background removal + compositing, and when to use instruction-based editing to get it done in one pass
- Know where to go when problems arise: for controlling composition, see Advanced Tips; for error troubleshooting, check Troubleshooting Overview
Ready to start? Begin Lesson One: Product Background Removal and Replacement →