修图与风格迁移

多张照片合成宠物合影

把几张照片里的宠物放进同一个场景,对齐光线并保住每只动物的花纹

更新日期:

完整提示词

英文提示词包含建议的示例需求。生成前请修改主体和细节。原始示例输入和生成参数尚未完全核实。

推荐模型: GPT Image 2.5 Flare

Combine the pets named in the request into one believable photograph. When two or more images are uploaded, treat the earliest as the destination and each following image as an animal to bring across, unless the request names another order. Preserve breed cues, coat pattern, markings, ear shape, and any collar exactly as shown. Place each animal on the surface the user names, with paws or hooves touching it and a shadow that follows the light already in the destination. Match color temperature, grain, lens softness, and time of day so the arrival does not look cut out. Leave people, furniture, architecture, and the original crop as they are, except where an animal must overlap them. If only one photo is uploaded, use it as the destination and build any missing animal from the written description; markings will only be reliable when a source photo is present. If nothing is uploaded, construct the room and the animals from the brief and keep them consistent inside the frame. Do not invent brand marks, readable collar lettering, or watermarks. The user may name a short label; render that text once, exactly, or omit lettering when none is given. Return one still photograph. User request: Use the living-room photo as the setting and add the greyhound from the later upload, lying on the empty rug in front of the sofa. Match the warm lamp light and leave the books and the plant where they are.

提示词技巧

  • 分享前对照每张原图,检查花纹、耳形和项圈是否一致
  • 看爪子和影子;若动物悬空或光线方向不对,补写接触面和灯在哪一侧

来源与署名

由 OmniArt 改编和重写,以下列出来源署名、修改说明及使用条件

  • Combine pets across images · OpenAI

    RaphaelAI-OpenLab/awesome-gpt-image-2.6

    Public example from OpenAI's image prompting guide; no redistribution license was verified and the collection's MIT license does not cover it. This template is independently authored; no source text or media is copied.

    Generalized the composite so any pets the user names can move into a destination photo, with upload order, coat identity, contact shadows, and a path when a photo is missing. The source collection labels this case GPT Image 2.6; GPT Image 2.5 Flare is a recommendation, not a verified example model.

来源与署名

相关提示词配方