HUMEN · LIFESTYLE TEST
把一套穿搭,带进生活场景。
2026-10-17 · GPT 内置生图实际生成 · 两轮样本
← 返回换装演练男模特复用 · 填写通用提示词 ↗质感可展示,商品保真未通过
两张的人物、咖啡馆环境和光线目视较连贯,毛衣绒毛与自然光有可展示的效果。领口、门襟纽扣、袖口仍有重绘;整体也比参考图更工整。输入是低分辨率截图,本轮没有验证真实商品精确还原、任意换模特或 API 接入。

三张原始截图分工:整套搭配 / 面孔 / 衣服细节。
下载实际提示词

带上第一张结果固定场景,继续附原始衣服和面孔参考。
下载实际提示词
本轮参考素材



来源:用户提供的截图。截图清晰度有限,没有独立商品平面图。
演练怎么讲
- 给参考图分配用途,逐件列出需要保留的穿搭。
- 写清场景、动作、光线、景别,先生成一张主图。
- 主图合格后,继续附原始商品图,再延展不同动作。
- 发现领口、纽扣等商品细节漂移时,停止扩图,回到高清商品参考修正。
本轮采用开源 clothing-grass-planting 的提示词与流程思路,并适配 GPT 内置生图;没有运行上游 CLI/API。原文与 MIT 许可证随记录保留。
查看完整实测记录 ↗第一轮实际提示词
Use case: identity-preserve. Create ONE new vertical photorealistic lifestyle fashion photograph, full-bleed portrait 2:3, not a collage or screenshot. This is a reference-guided fashion lookbook test. Reference roles: Image 1 establishes the complete outfit, proportions and walking energy. Image 2 establishes the adult woman's face and short dark bob identity. Image 3 is the closest garment reference and establishes knit texture, collar, front button placket, cuffs and hem. Ignore all screenshot borders and UI. Preserve this same adult woman and this same complete outfit, rather than making a generic fashion model. Outfit invariants: rust/cinnamon orange fuzzy wool cardigan, short stand collar, small dark round buttons in the original arrangement, long relaxed sleeves, ribbed cuffs and hem, slightly relaxed cropped hip-length fit; ivory wide-leg trousers with soft natural folds; tan/brown loafers. Preserve garment construction and proportions visible in the references. Do not add a bag, jewelry, logos or new garment details. NEW scene and action: A small real neighborhood cafe with an open timber-and-glass doorway, worn pale plaster, street visible through the window and a wood magazine shelf further inside. The woman is caught taking a relaxed step just across the doorway into the cafe, looking toward the window rather than at the photographer. Hands naturally relaxed; an unforced moment during movement. Show her full body and shoes, a slightly off-center handheld framing at normal standing eye height, approximately a 40mm environmental photograph. Give the setting believable depth and ordinary small imperfections. Quality direction: natural late morning window/street daylight, readable shadows and realistic highlight rolloff, coherent contact shadows beneath the shoes, subtle stray hairs, genuine skin texture, fine wool fibers and uneven wool loft. Knit must look soft and tactile, not smooth plastic or 3D-rendered. Preserve a little movement in loose hair or the moving hand while keeping the cardigan front and face readable. Background is recognizable, only moderately out of focus. Neutral natural colors with warm sunlight, no orange wash. Feels like a well-composed real candid clothing photo, not glossy beauty advertising. Avoid: beauty retouching, wax skin, perfect mannequin pose, forced smile, extreme bokeh, excessive sharpening, fake film overlays, added text, borders, watermarks, composite cutout halos. Output the single new photograph only.
第二轮实际提示词
Use case: identity-preserve. Generate ONE new full-bleed portrait 2:3 natural lifestyle fashion photograph from these references. Not a collage. Image 1 is the just-generated cafe doorway shot: use it to preserve the same adult woman's face, short bob, overall outfit, tan loafers, cafe architecture (wood framed large windows, worn pale plaster, wooden magazine shelf), natural color rendering and sunlight. Image 2 is the original close-up product/outfit reference: it overrides Image 1 for the cardigan construction, button spacing/arrangement, stand collar proportions, cuff and hem ribbing, fuzzy cinnamon-rust wool texture and ivory trousers. Image 3 is the original woman seated on a tram: an additional identity and natural relaxed expression reference, NOT a background reference. Do not reproduce its tram. NEW SHOT: The same woman is now seated on a simple wooden chair beside the window inside the SAME cafe, body turned slightly toward the window, looking out and reacting with a small relaxed unposed smile. Both hands naturally rest separately on her lap, with realistic fingers. Show from above her head down to mid-shin, three-quarter seated view, roomy crop to show how the cardigan sits at the waist and how the ivory wide trousers fold on a seated body. No added bag, jewelry, cups or accessory. Do not change haircut, facial identity, outfit or time of day. Lighting: Same street daylight comes through the tall side window and rakes across the cardigan sleeve and cheek, leaving believable soft room shadows, with natural reflected light from the pale wall. Expose skin naturally, retain highlight detail, no glamorous beauty lighting. Textures: visible soft wool flyaways, genuine subtle facial pores and ordinary skin, realistic seams and natural fabric wrinkles. Keep the cardigan front readable. A handheld 40mm environmental fashion photograph with a moderately recognizable background, natural slight asymmetry, real caught-between-moments energy. Avoid: a rigid straight-on catalog pose, beauty retouching, porcelain skin, excessive bokeh, fake film filter, increasing the number of buttons, lengthening the collar, text, screenshot borders, logos and watermarks. Only the new photograph.