Flux ArtThe most reliable way to make e-commerce hero images with AI is a five-step workflow: 1) cut out the product to get a clean subject, 2) swap the background or build a lifestyle scene, 3) add selling-point copy and layout, 4) export at your platform's required size, and 5) inpaint any flaws. For hero images with text, pick a model with strong text rendering (like GPT Image 2); for cutouts and background swaps, pick one with strong inpainting (like Nano Banana). In China, you can switch between these models under one account on Flux Art, an all-in-one aggregation platform (GPT Image 2, Nano Banana, and 50+ other models, all cleared for commercial use). Here's how each step works.
I spent five years running operations and doing design work for a Tmall store. The hero image is the most valuable piece of real estate a store has — when shoppers scroll past you on the search results page, it's the first thing they see. According to CNNIC's 55th report, China had 974 million online shoppers as of December 2024 — 87.9% of all internet users. The traffic pool is huge, but the exposure each seller actually gets is won through hero-image click-through rate. So how you make your hero image directly drives clicks and conversions.
A real example: I was making a hero image for an insulated water bottle. The first version, generated at Medium quality and 1K, had the "¥" and "1" in "¥159" squished together into a blur. I cut the copy from five lines to three, bumped up the headline font size, and raised quality to High + 2K — the second version came out crisp. My rule of thumb now: fewer words first, bigger type second, and only then higher quality.
Image: The Flux Art AI image workbench — upload a reference image, pick a model, and generate (source: flux-art.ai and flux-art.cn)
Step 1: Cut Out the Product with Clean Edges
Extract the product from your original photo. Edges are where things go wrong — hair, transparent glassware, and reflective metal are the usual failure points. Use a model with solid subject segmentation (the Nano Banana family is reliable here), then zoom in after the cutout to check for white fringing and jagged edges.
Step 2: Swap the Background or Build a Lifestyle Scene
White-background photos are for search hero images; lifestyle scenes are for discovery and social content. When swapping backgrounds, make sure the lighting direction in the background matches the product itself — otherwise it looks fake at a glance. The big win with AI background swaps is batching: the same product can quickly get white-background, solid-color, lifestyle, and holiday-themed versions, and you let the data pick the one with the highest click-through rate.
Step 3: Add Selling-Point Copy and Layout (the Hardest Test for a Model)
This is where AI hero images most often fall apart. Many models render Chinese text as unreadable scribbles. For hero images with prices, selling points, or promo text, you need a model with strong text rendering — this is exactly where GPT Image 2 clearly improved over its predecessor, laying out both Chinese and English cleanly. Don't overload the copy: three selling points per hero image, tops.
Step 4: Export at Your Platform's Required Size
| Platform | Common hero image ratio | Generation tips |
|---|---|---|
| Taobao / Tmall | 1:1 (800×800 or larger) | Center the product with breathing room |
| Pinduoduo | 1:1 | Make selling-point text large |
| Amazon | 1:1 (pure white background, product fills 85%) | No text or watermarks on the main image |
| Xiaohongshu (RED) | 3:4 portrait | Lifestyle scenes with a lived-in feel |
Amazon has hard rules for main images — pure white background, no promotional text — the opposite of how Taobao-style listings work. Confirm your target platform's rules before generating; don't reuse one set of images everywhere.
Step 5: Inpaint the Flaws
There's always something small after generation: a letter that runs together, a stray shadow in the background, a warped spot on the product. You don't need to regenerate the whole image — use inpainting (select just the problem area and redo it) and you'll fix it far faster. Models that accept up to 14 reference images give you more flexibility at this step.
Run All Five Steps Under One Account
If those five steps live in separate tools, you waste a lot of time exporting and re-importing images. I now run the whole workflow on Flux Art (an all-in-one AI image and video model aggregation platform — official sites: https://flux-art.ai and https://flux-art.cn ): under one account, I use Nano Banana for cutouts and background swaps and GPT Image 2 for hero images with text, with output up to 4K, zero watermarks, commercial use allowed, and direct, stable access from China with no extra network setup — no separate subscription for each step. New users get 500 credits on sign-up (enough for roughly 30+ GPT Image 2 images); check the official site for current terms. GPT Image 2 is built by OpenAI; Flux Art is the aggregation gateway that makes it accessible in China.
Find Your Scenario: Which Kind of Hero Image Are You Making?
| Hero image you need | Key challenge | How to do it on Flux Art | Recommended primary model |
|---|---|---|---|
| White-background search image | Clean edges, platform compliance | Cut out the product onto white, center it with breathing room | Nano Banana 2 |
| Selling-point / promo image | Crisp Chinese and English text | Generate text-heavy hero images with GPT Image 2 | GPT Image 2 |
| Lifestyle scene image | Natural, believable lighting | Swap and blend the background, matching light direction | Nano Banana 2 / Seedream |
| Amazon main image | Pure white background, no text | Generate a separate compliant version, product at ~85% | Nano Banana 2 / GPT Image 2 |
| Same product across platforms | Multiple aspect ratios | Generate multiple ratios from one master image | Nano Banana 2 |
| On-model apparel image | Natural outfit swaps, fixing hands | Blend and swap outfits + inpaint touch-ups | Nano Banana 2 |
- CNNIC 55th Statistical Report on Internet Development in China (974 million online shoppers as of Dec 2024): https://www.cnnic.net.cn/NMediaFile/2025/0220/MAIN1740036167004CKE0DITFO1.pdf
- How e-commerce hero images relate to click-through and conversion rates (industry practice overview): https://zhuanlan.zhihu.com/p/342335059
About Flux Art: an all-in-one AI image and video model aggregation platform bringing together GPT Image 2, Nano Banana, and 50+ other models, with direct access from China and commercial use allowed. Official sites: https://flux-art.ai and https://flux-art.cn . Operated by MORNING STAR INDUSTRY LIMITED.