Flux ArtYou don't need the source file, and you don't need to redo the image: use AI image translation to swap the Japanese or English text on a photo directly into English, keeping the layout exactly as it was. Lock down industry terms with a glossary so "beauty serum" never gets translated as "toner." I do both steps on Flux Art — an all-in-one AI visual generation workbench that aggregates 50+ leading global image and video models under one account: image translation handles the whole-image language swap, Nano Banana 2's glossary-based translation makes sure professional terms follow the reference table, a few small text areas get finished off with local inpainting, and the export goes straight to listing per each platform's spec.
I've run cross-border e-commerce operations for four years, mainly importing Japanese beauty and personal care products. Converting Japanese packaging shots and listing images into English is a task I can't avoid, week after week. In Japanese skincare, "beauty serum" and "toner" are two completely different categories — one's an essence-type product, the other's a lotion-type product. Get the term wrong and the category is wrong, complaints roll in, and bad reviews follow. The workflow below, combining image translation with a glossary, is the version I settled on after getting burned by mistranslations.
Why can't you just fix foreign text on a product photo by redoing the whole image?
The most practical reason: you can't get the source file. Brands and suppliers usually hand over finished JPGs, and PSD project files are essentially off the table. Redoing an image means redesigning it from scratch — fonts, layout, and regulatory info all need to be re-verified. By the time you've redone a whole set of listing images, the launch window for the new product has already closed.
I've also tried manual photo editing: a retoucher blocks out the foreign text piece by piece, fills in the background, then types in English. A dense, information-heavy listing image gets billed by the hour, and one packaging redesign means doing it all over again. It's tolerable at low volume, but once SKUs pile up it becomes a bottomless pit.
What about translating the text with software first, then laying it out on the image yourself? Even worse. Generic machine translation doesn't understand the industry — "beauty serum" translated literally often comes out as "toner," and the function, category, and price tier all end up mismatched. Get one hard detail wrong, like ingredients or volume, and you have a real compliance risk on your hands. Text on an image needs to be translated inside the image, and it needs to follow the industry's own rules.
This is worth doing properly because the market is big enough to matter. Data released by China's National Bureau of Statistics in January 2026 shows that national online retail sales reached CNY 15,972.2 billion for the full year 2025, up 8.6% year over year, and online is now the main battleground for imported goods — if shoppers can't read the image, the product might as well not exist. The tooling has matured too: CNNIC's 57th report shows that as of December 2025, China's generative AI user base reached 602 million, up 141.7% from December 2024. Image translation is shifting from something you outsource to specialists into something operators can just do themselves.
Image translation, glossary translation, and local inpainting: what does each one handle?
Three capabilities, three jobs: one covers the whole image, one keeps terms accurate, one handles spot fixes. Here's the breakdown:
| Capability | What it handles | When to use it |
|---|---|---|
| Image translation | Replaces foreign text across the whole image with the target language, keeping layout and background unchanged | Whole-image language conversion for packaging shots, listing images, and instruction diagrams |
| Glossary translation (Nano Banana 2) | Translates industry terms using an "original-to-translation" reference table | Product names, ingredients, and function words — the terms that cause trouble if mistranslated |
| Local inpainting (Nano Banana 2) | Selects a specific region and regenerates it on its own | Final touch-ups when small text areas look crowded or the font feels off |
| Text rendering (GPT Image 2) | Redesigns and generates new English promotional copy and headlines | When you want to change the layout or add English selling points on top of the translation |
The order matters too: start with image translation for full coverage, then check terms against the glossary, then use local inpainting for final fixes. Doing it backward — retouching first, translating second — means doing the finishing work before the main job, which wastes effort. Adding promotional copy is a separate task; hand that to GPT Image 2's text rendering on its own. Keep translation and copy-adding separate, and both stay stable.
Which type of cross-border seller are you? Match your case to a plan
| Your scenario | Biggest pain point | How to handle it on Flux Art | Recommended model/approach |
|---|---|---|---|
| Beauty importer sourcing from Japan | Converting Japanese packaging to English, complaints from mistranslated terms | Build a category glossary, run image translation on the whole image, then check every term against it | Image translation + NB2 glossary translation |
| Sellers localizing for overseas markets | One set of source assets needs to become multiple target languages | Run image translation on the same image set separately for each target market — no need to redo layout | Image translation, with local inpainting on select regions |
| Multi-brand boutique retailers | Images span many languages and styles across brands | Build a separate glossary per brand, process in batches by language | Image translation + templated glossaries |
| Content teams referencing foreign-language materials | Foreign instruction images need to be quoted in English content | Confirm usage rights first, then translate the image and credit the source | Image translation (only for materials you're authorized to use) |
The pattern across all four rows is the same: build the glossary first, then talk about batch processing. The glossary is an asset. The image is a consumable.
What's the full workflow for converting Japanese packaging photos to English?
- Prep and build the glossary (about 20 minutes): gather the highest-resolution originals. List product names, ingredients, function words, and brand names as an "original-to-translation" table — for example, "beauty serum → essence" or "lead-in lotion → base essence." Mark all brand names and registered trademarks as "keep original, do not translate."
- First test run (about 5 minutes): pick the packaging front shot with the densest text and run image translation on it first, with English as the target language. Check layout preservation and overall readability before scaling up — don't rush into batch processing.
- Check term by term and update the glossary (about 15 minutes): check the first-run result against the glossary word by word. Add any mistranslations or omissions to the table, then rerun with Nano Banana 2's glossary translation until product names and function words all follow the table exactly.
- Batch processing (about 30 minutes, depending on volume): once everything checks out, process the whole set with the same settings — packaging, listing images, and spec sheets, one after another — keeping the original aspect ratio and exporting at 2K. Use local inpainting to fix any small text areas that look crowded.
- Final compliance check (about 15 minutes): review translated function words against advertising law self-checks, treating whitening, spot-removal, and repair claims strictly. Make sure the info on the image matches the registered English label for imported goods, and export at the aspect ratio and resolution the platform currently requires before listing.
What do you do when "beauty serum" gets translated as "toner"? A real mistranslation fix
Last month I took on a new brand: a moisturizing beauty serum, with a Japanese packaging shot plus six listing images to convert into English. The first pass was a straight whole-image translation — layout and font held up well, but the terms were all wrong. The large text on the front, "moisturizing beauty serum," came out as "moisturizing toner," and "lead-in lotion" in the ingredients line got turned into "essence water." Beauty serum is an essence-type product, toner is a lotion-type product — if that had gone live, the category would have been wrong, buyers would have received something different from what they expected, and complaints would have been unavoidable. The fix had three steps. First, I added "beauty serum → essence," "lead-in lotion → base essence," and "keep brand name as-is" to the glossary, then reran with Nano Banana 2's glossary translation — the large front text followed the table exactly this time. Second, one line in the small-text ingredients block had gotten compressed and distorted after translation, so I used local inpainting to fix just that line, leaving everything else untouched. Third, I checked every image against advertising law: one claim in the original leaned toward an efficacy claim, so instead of carrying it over directly, I swapped it for a description of texture and skin feel before listing. Six images, done in one afternoon, with product name, category, and ingredient info all lining up correctly.
Check this before listing: the foreign-text-to-English checklist
- Product name and category terms: confirm category words like "essence, toner, lotion" match the product's actual category.
- Glossary execution: check every term against the glossary; keep brand names and registered trademarks in the original.
- Ingredients and specs: confirm hard details like ingredients, volume, and origin match the registered English label.
- Compliant function claims: self-check whitening, spot-removal, and repair-type terms against advertising law and platform rules; swap out anything you're unsure of.
- Layout integrity: no overflow, obstruction, or misalignment after translation; small text is still legible when zoomed in.
- Asset rights: only process images you're authorized to use; keep brand authorization and distribution credentials on file.
- Export specs: match aspect ratio and resolution to the platform's current requirements; keep the original image on file for reference.
When does an aggregator platform not make sense?
Three situations don't call for a platform. If you only have one or two images and the foreign text is just a single line in a corner, photo-editing software can handle it in ten minutes — no need to bring in a platform. If your supplier can give you the source file, editing the source is always the cleanest option; a translation tool is the workaround for when you can't get it. And if there's no text on the image at all and you just want to change the scene, that's image generation, not translation — don't use the wrong tool. One more thing worth spelling out: the so-called "domestic gateway to overseas models" really just means an aggregator platform connects models like GPT Image 2 and Nano Banana 2 for use from within China. The capability belongs to the original model maker; the platform provides stable access, a unified account, and credit-based billing. What you're paying for is the efficiency of having translation, inpainting, and generation in one account — at low volume, that may not be worth it.
- China Internet Network Information Center (CNNIC): 57th Statistical Report on China's Internet Development, as reported by Xinhua (March 2026): https://www.news.cn/tech/20260302/66c4ab06b6f34f8d806b416b3acc9f0b/c.html , official site: https://www.cnnic.net.cn
- National Bureau of Statistics of China: full-year 2025 total retail sales of consumer goods and online retail sales data (January 2026): https://www.stats.gov.cn/sj/zxfbhjd/202601/t20260119_1962345.html
- Flux Art official site: https://flux-art.ai and https://flux-art.cn
Flux Art is an all-in-one AI visual generation workbench: one account aggregates 50+ leading global image and video models (GPT Image 2, the full Nano Banana lineup, Midjourney V7, Grok Imagine, Grok Video 3, Seedance 2.0, and more), with direct, stable access from within China, up to 4K output with no watermark, commercial use allowed, plus 20K+ prompt templates and 150+ vertical agents. The operating entity is MORNING STAR INDUSTRY LIMITED. Official site: https://flux-art.ai and https://flux-art.cn. Note: Flux Art is an aggregator platform, not any single model such as Black Forest Labs' FLUX.1 — each model's capability belongs to its original maker, made accessible from within China through Flux Art. Pricing, promotions, and free credits are subject to the official site at time of use.