Tongyi Wanxiang Prompt Builder

Structure is what makes a text-to-image prompt reliable

When you prompt Alibaba Tongyi Wanxiang (Wan) — or any modern text-to-image model — the quality of the picture depends far more on how you structure the request than on any single clever phrase. A structured prompt names the subject, sets the scene, fixes the visual style, chooses the composition and camera, defines the lighting, states the intended use case, lists the quality settings, and finally excludes the artefacts you do not want. This builder keeps that structure for you: fill the fields, and it joins them into a clean prompt with the subject as the leading line, followed by clearly headed sections — each prefixed with a Markdown-style heading the model can read at a glance — ready to paste straight into Wan. The result is the kind of prompt a careful e-commerce art director would write by hand, only assembled in seconds, and reproducible every time you tweak one field.

The single highest-leverage line is the subject — and for online retail, that means the product. "A frosted-glass skincare serum bottle with a gold pump, water droplets on the surface" steers the render far more reliably than "a serum bottle", because Wan reproduces the material, colour and key details you actually name. After the subject, the scene and composition do the heavy lifting: a clean seamless background with a centred, front-on framing is the bread and butter of the e-commerce hero shot, while a marble surface or a styled flat-lay suits lifestyle and detail images. Wan is particularly strong on Chinese aesthetic context, so local style words — guochao, ink-wash, neon — translate into posters that feel native rather than generic, which is exactly where many Western-trained models struggle.

Use case, lighting and negatives separate a demo from a sellable image

The fields people skip and regret are lighting, use case and negatives. Lighting is the soul of product texture: soft studio light flatters packaging, a rim backlight carves out silhouette and gloss, and warm ambient light sells a lifestyle mood — the same bottle can look cheap or premium depending on this one field. The use case (e-commerce hero, poster, banner) is what tells the model to leave clean negative space and a text area, so your designer can later drop in a headline, a price tag or a logo without fighting a busy background. And the negatives — no extra text, no watermark, no distortion, no cluttered background — are what turn a promising draft into an image you can actually publish, because product and poster work is judged on cleanliness above all.

Because the output is structured plain text, the same prompt is portable: it works best in Wan but pastes cleanly into other text-to-image tools too, with only the local style words behaving differently. Write it in Chinese when you want idiomatic posters and Chinese-copy scenes; the structure travels regardless of language, and you can mix Chinese and English keywords freely. And because the whole tool runs locally in your browser, you can iterate without limits — adjust the lighting, copy again, regenerate, compare — and nothing you type ever leaves your device, is sent to a model, or is stored. Treat the first prompt as a draft: generate it, see where the render drifts — a busy background, the wrong shadow, a distorted label — and tighten the matching field or extend the negatives. Two or three rounds of that usually turn a rough generation into a clean, on-brand product shot or poster, and you keep a reusable prompt at the end.