I have been testing a few text-to-image workflows for product mockups and social creatives, and the biggest difference is not the prompt length but how consistently the model holds lighting, skin texture, and product geometry across a batch. For Chinese-language prompts I keep a short checklist: describe the camera, the material, the negative space, and one concrete lighting source, then regenerate only the failed regions instead of the whole frame.
A tool I have been using for the Chinese interface is ai生图. It is aimed at photorealistic output rather than cartoon styles, which helps when the asset has to sit next to real photography. I usually start with a reference description of the subject, lock the aspect ratio to the placement (1:1 for feeds, 4:5 for portraits, 16:9 for banners), and only then add style words. Over-prompting style tends to flatten texture.
If you are generating many variants for different storefronts, save the prompt as a template and change only the product noun and background. That keeps color and lighting comparable so the set looks like one shoot. I also export at the largest size the tool offers and downscale in an editor, because resizing up later shows artifacts around hair and type.
Curious how others here structure prompts when they need the same product in several locales without the face or packaging drifting. Happy to compare notes.