메뉴
引导用户上传同一商品不同面的 1-5 张普通手机照片，综合锁定商品事实后逐张自动精修，重点改善真实材质质感、光泽、微纹理、边缘、曝光、白平衡、透视、噪点、灰尘、背景和反光。用户提到手机拍摄商品图精修、随手拍变电商图、自动修商品照片、提升产品质感或影棚级优化时使用。创建新场景、复刻参考图、只改局部区域或批量改画幅时不要使用本技能。
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/QIYU-JACKMAN/codexQIYU-image-workflow --skill ecommerce-phone-photo-retouch명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
使用预制素材文件夹批量读取 1-20 张原图，并将每张图智能重排到一个或多个目标画幅。用户提到批量改尺寸、多张图片统一适配比例、批量横转竖或平台尺寸适配时使用。单张图片调整和复刻风格不要使用本技能。
把 1-20 张不同 SKU 商品图逐一套入同一个模板，先用内置 gpt-image-2 模块为每个 SKU 优化提示词并请求客户校验，保持整批风格一致，并支持一一对应的商品名称、型号、尺寸或规格文本。用户提到批量 SKU、同模板换产品、批量更新型号名称、统一商品卡片时使用。
根据用户的角色定位和自由指令，把电商内容拆成 1-20 个独立图片任务，使用内置 gpt-image-2 模块优化每屏提示词、请求客户校验后生成统一套图。用户说自定分批、自定义多屏、按一段要求拆若干屏、自由规划详情套图时使用。标准主图详情页、逐参考图复刻、SKU 模板批量和局部替换不要使用本技能。
SOC 직업 분류 기준
SKILL.md 표시 중
| name | ecommerce-phone-photo-retouch |
| description | 引导用户上传同一商品不同面的 1-5 张普通手机照片，综合锁定商品事实后逐张自动精修，重点改善真实材质质感、光泽、微纹理、边缘、曝光、白平衡、透视、噪点、灰尘、背景和反光。用户提到手机拍摄商品图精修、随手拍变电商图、自动修商品照片、提升产品质感或影棚级优化时使用。创建新场景、复刻参考图、只改局部区域或批量改画幅时不要使用本技能。 |
保持内联执行，不使用子代理，不创建素材投递文件夹。
先请用户直接在对话中上传同一商品不同面的 1-5 张手机照片。根据商品特点优先选择正面、背面、左右侧面、顶部/底部、Logo/标签、接口或功能细节、材质纹理、包装与配件中最有信息量的角度。商品结构简单时不强求凑齐五张；已有照片不重复索取。
要求尽量使用原图，不要先加滤镜、截图或压缩。不同 SKU 不要混在同一批；多个 SKU 转到批量 SKU 或分项目处理。
先综合所有角度建立一份商品事实表，再逐张检查曝光、动态范围、白平衡、偏色、透视、水平线、镜头畸变、噪点、压缩痕迹、对焦清晰度、灰尘污点、背景干扰、阴影、反光、材质表现和裁切空间。把问题分为：可安全修复、需要生成式重建、无法从照片确认。
同时登记不可改动的商品事实：外形与比例、部件位置、颜色、材质、表面纹理、光泽类型、Logo、标签文字、包装、接口、按钮、配件数量和任何不对称细节。看不清的事实标记为不确定，不猜测或补造。
材质诊断必须区分并描述真实的光学表现，例如：哑光塑料的细腻漫反射、亮面塑料的受控高光、拉丝金属的方向性纹理、抛光金属的干净反射、玻璃的透光与边缘高光、皮革或织物的细微纹理、木材的自然纹路。只使用照片可见或用户确认的材质事实。
根据照片问题推荐一档并让用户确认；用户未指定时推荐“标准电商精修”。
natural 自然修复：只修曝光、色温、噪点、轻微透视、灰尘和清晰度，保留真实拍摄环境与自然质感。ecommerce 标准电商精修：在自然修复基础上整理背景、优化光影层次和商品边缘，使画面达到干净专业的电商成片水平，但不更换场景或重设计商品。studio 影棚升级：重建为可信的商业影棚光线和干净背景，同时保持原构图意图和全部商品事实。若用户要求新场景或营销创意，转到快捷生成。默认保持原始画幅、主体位置、可见文字和商品数量。裁切、移除道具、更换背景颜色、删除阴影或大幅改变构图必须单独列出并获得确认。
gpt-image-2-prompt-optimizer.md 和质检规则。phone-photo-retouch 项目，每张输入照片登记为一个 product asset，并严格创建一个对应任务。每个任务中 Image 1 是当前待精修照片；其余角度按确认顺序作为商品结构、颜色、材质、Logo 和文字核验参考，不提供当前画面的构图或可见内容。original_request。使用下方内置精修提示词模块生成 precise-object-edit 或 identity-preserve 提示词，分别写明允许修改项和不可修改项。pending。plan.approved=true 并生成。优先使用 Codex 内置图片能力，否则调用公共 run_plan.py。逐张对比原图，检查商品几何结构、颜色、Logo、标签文字、配件数量和原始画幅是否保持；同时检查曝光、白平衡、透视、噪点、边缘、背景、反光和清晰度是否改善。任何商品事实漂移都判定为失败。
只对失败图片重做，并把失败原因改写为一个针对性的修正要求。不得覆盖原始手机照片；最终结果和 QA 记录保存到标准项目目录。
根据每张照片的实际诊断删减空项，不机械堆砌。把 <...> 替换为已确认内容：
Use case: precise-object-edit
Asset type: ecommerce product photo retouch
Primary request: Retouch Image 1 into a clean, premium, photorealistic ecommerce product photograph while preserving the exact product identity and the original photographic intent.
Input images: Image 1 is the only edit target and controls the current view, pose, framing, visible parts, and object count. Images 2-N are cross-angle fact references only for verifying geometry, color, material, surface texture, logos, labels, ports, controls, packaging, and accessories; do not copy their viewpoint or reveal details that are hidden in Image 1.
Corrections: <approved exposure, white balance, perspective, lens distortion, noise, dust, background, reflection, edge, and sharpness corrections>.
Material fidelity: Render physically believable premium material response based only on confirmed evidence. Preserve the product's real base color and material category. Recover fine micro-texture, clean edge definition, subtle tonal separation, natural surface depth, and controlled local contrast. Keep matte surfaces softly diffused without waxy smoothing; glossy surfaces with clean, shaped highlights without blown glare; brushed metal with fine directional grain; polished metal with coherent reflections; glass with realistic transmission and crisp edge highlights; leather, fabric, wood, and coated surfaces with their authentic small-scale texture. Make the product feel tactile, substantial, and professionally photographed, never synthetic or overprocessed.
Lighting: Refine into soft commercial studio-quality light with coherent direction, smooth highlight roll-off, readable shadow detail, natural contact shadow, and dimensional separation. Control harsh phone-flash hotspots and color casts without flattening the object or inventing new reflections.
Detail and sharpness: Improve perceived clarity through restrained micro-contrast and edge acuity. Preserve natural antialiasing and texture variation. Do not hallucinate unreadable text, hidden components, new seams, grain, scratches, embossing, or decorative details.
Background: <preserve and clean the existing environment | simplify to a clean neutral ecommerce background | approved background treatment>. Keep realistic grounding and edge integration.
Constraints: Preserve exact geometry, proportions, silhouette, viewpoint, crop, aspect ratio, product count, part placement, colors, materials, logos, label spelling, packaging, ports, controls, accessories, asymmetry, and all confirmed product facts. Change only the approved photographic defects and background treatment.
Avoid: product redesign, geometry drift, color shift, fake texture, plastic-looking smoothing, excessive denoising, crunchy oversharpening, halos, HDR glow, clipped highlights, crushed shadows, floating product, invented text, invented features, extra objects, watermark, collage, or changing details outside the approved corrections.