agentleFS
dlazy-aidlazy-ai/ecommerce-skills/skills/material-enhancement/skill.md

material-enhancement

Skill71 starsChanged 3 months agoMITView on GitHub

材质质感增强与纹理重建。糊掉的图 + 高清商品图 → 纹理清晰可信的图。当用户说「增强质感」「图糊了」「补纹理」「提清晰度」时使用。

The file

Files
skill.md

material-enhancement — 优化服装材质和细节质感

把一张面料糊掉的商拍图修成面料清晰真实的图,构图和人一动不动。

这是一个后处理技能:AI 生成的服装图最容易崩的就是面料——远看还行,放大一看针织变成一片糊。给它一张真实的高清商品图当参照,把表面重建回来。


生成效果示例

输入:原图(待增强)输入:高清商品图(真实面料)
<img src="../../docs/material-enhancement/source-image.jpg" width="240"><img src="../../docs/material-enhancement/hires-product.jpg" width="240">
source-image.jpg — 军绿麻花毛衣上身图,1024×1536hires-product.jpg — 同款毛衣高清平铺图,800×800

实际执行的命令:

dlazy gpt-image-2 \
  --prompt 'Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo that defines the true fabric. Rebuild the sweater surface in image 1 using the real texture from image 2: crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, correct fold shadows and a matte fibre sheen. Do not change anything else — the model face, hair, hands, pose, brown trousers, background, framing and colour grading must stay identical to image 1. The garment silhouette and colour must not shift; only the material fidelity and micro-detail improve. Photorealistic, sharp, no text, no watermark.' \
  --images docs/material-enhancement/source-image.jpg docs/material-enhancement/hires-product.jpg \
  --size 1024x1536 --quality high --imageFormat jpeg \
  --save docs/material-enhancement/example-output.jpg

输出

<img src="../../docs/material-enhancement/example-output.jpg" width="320">

example-output.jpg — 1024×1536,60 credits。麻花辫的立体起伏、菱形提花的凹凸、羊毛的绒毛感与褶皱处的暗部层次被重建;模特五官、发型、手部、姿势、棕色长裤、灰墙背景与整体色调保持不变,毛衣轮廓与军绿色未偏移。


1、能力边界

输入作用
原图要改善的商拍图,决定构图、模特、姿势、背景与色调
高清商品图提供真实面料信息(织法、纹理、绒感、光泽)
服装类型帮助判断哪些表面特征该被强化
只改不改
面料表面纹理、微观细节、褶皱阴影层次、纤维光泽构图、模特(脸/发/手/姿势)、其他服饰、背景、色调、服装轮廓与颜色

不做:不改服装轮廓与颜色;不改模特与背景;不把一种面料换成另一种(换面料请用 fabric-on-body)。


2、输入素材规则

生成前先自检这几条硬性约束:

  • 大小:20KB ~ 15MB
  • 分辨率:大于 400×400
  • 格式:jpg / jpeg / png / webp

原图(✅):构图与人物都满意、只有面料不行的图。

高清商品图(✅):同一件商品的高分辨率平铺图或特写,能看清织法。

注意事项

情况说明
⚠️ 两张图必须是同一件商品面料不同的两张图会导致材质张冠李戴
⚠️ 原图颜色如果已经偏了本技能只管纹理,颜色偏差要在生成阶段解决
❌ 原图崩坏严重(版型错、结构乱)材质增强救不了结构问题,重跑生成
❌ 原图分辨率极低没有足够信息定位纹理该长在哪

3、材质增强的两条铁律

铁律一:只改表面,别的一个像素都别动。

Do not change anything else — the model face, hair, hands, pose, other garments,
background, framing and colour grading must stay identical to image 1.
The garment silhouette and colour must not shift; only material fidelity
and micro-detail improve.

铁律二:把「好质感」翻译成具体的表面特征,否则模型只会整体加锐化。

面料要重建的具体特征
粗针织crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, matte fibre sheen
精纺羊毛fine even weave, subtle fibre halo, soft directional sheen
缎面smooth continuous specular gradients along the folds, no texture noise
灯芯绒parallel wale ridges with correct pitch, matte pile catching light on the ridge tops
牛仔diagonal twill lines, slub irregularity, whiskering at stress folds
皮革grain pores, broad soft creases, low-frequency specular sheen
摇粒绒/绒毛dense short pile, fuzzy silhouette edge, light scattering rather than specular

再加一条褶皱层次:correct fold shadows with proper ambient occlusion in the creases——这是「看起来有厚度」的来源。


4、工具调用

本技能使用 dLazy 的 gpt-image-2(图像编辑模型 + --quality high;材质增强要求「几何与色彩零变化、表面高频信息重建」,是典型的双图参考定向修复)。

调用方式

两种等价写法,选一种。统一入口会自动选后端、失败重试、建目录落盘、估算成本:

# A. 统一入口(推荐):可切任意后端,加 --dry-run 不计费空跑
node scripts/gen.mjs --task material-enhancement \
  --prompt '<见下方 Prompt 模板>' \
  --images <按下表顺序> \
  --save output/material-enhancement-<sku>.jpg

# B. 直接用 dLazy CLI(不想引入 Node 依赖时,效果等价)
dlazy gpt-image-2 --prompt '...' --images ... --save output/material-enhancement.jpg

参数约定(本技能固定用法)

参数取值理由
--images[原图, 高清商品图]顺序即 prompt 中的 image 1 / 2
--size与原图比例一致增强不该改构图
--qualityhigh(必须)本技能的产出就是高频细节
--imageFormatjpeg通用格式
--batch2纹理重建有随机性
--savedocs/material-enhancement/output-<sku>-enhanced.jpg与原图分开归档便于对比

Command Examples

# basic call
dlazy gpt-image-2 \
  --prompt 'Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo defining the true fabric. Rebuild the garment surface in image 1 using the real texture from image 2: crisp knit relief, visible loops and yarn twist, natural loft, correct fold shadows. Do not change anything else — model, pose, background, framing and colour grading stay identical to image 1. Only material fidelity improves.' \
  --images docs/material-enhancement/source-image.jpg docs/material-enhancement/hires-product.jpg \
  --size 1024x1536 --quality high

# complex call: 批量给一组生成图做材质后处理
HIRES=docs/material-enhancement/hires-product.jpg
LOCK='Do not change anything else — the model face, hair, hands, pose, other garments, background, framing and colour grading must stay identical to image 1. The garment silhouette and colour must not shift; only material fidelity and micro-detail improve. Photorealistic, sharp, no text, no watermark.'
FEAT='crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, matte fibre sheen, correct fold shadows with proper ambient occlusion in the creases'
for f in docs/material-enhancement/raw/*.jpg; do
  dlazy gpt-image-2 \
    --prompt "Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo that defines the true fabric. Rebuild the garment surface in image 1 using the real texture from image 2: $FEAT. $LOCK" \
    --images "$f" "$HIRES" \
    --size 1024x1536 --quality high --imageFormat jpeg \
    --save "docs/material-enhancement/enhanced/$(basename $f)"
done

# 先估价不真跑
dlazy gpt-image-2 --dry-run --prompt '...' --images a.jpg b.jpg --size 1024x1536 --quality high

延伸阅读

要查什么去哪
认证、多后端配置、输出结构、错误码references/provider-cli.md
gpt-image-2 的全部可用参数references/model-flags.md
统一入口的全部选项node scripts/gen.mjs --help

5、Prompt 模板

Texture enhancement pass.
Image 1 is the on-model photo to improve.
Image 2 is the high-resolution product photo that defines the true fabric.

Rebuild the [服装部位] surface in image 1 using the real texture from image 2:
[第三节表格里对应面料的具体特征],
correct fold shadows with proper ambient occlusion in the creases.

Do not change anything else — the model face, hair, hands, pose, other garments,
background, framing and colour grading must stay identical to image 1.
The garment silhouette and colour must not shift; only material fidelity
and micro-detail improve.

Photorealistic, sharp, no text, no watermark.

按问题追加的修正句

问题追加到 prompt 末尾
只是整体变锐利了This is not sharpening: reconstruct the actual [织法单位] structure, loop by loop, matching image 2.
人脸/背景也变了Only pixels inside the garment region may change. Everything outside it must be identical to image 1.
颜色被改了Preserve the exact garment colour from image 1; do not shift hue, saturation or brightness.
纹理密度不对Match the texture scale to the garment size in image 1: about [N] knit loops per [尺寸].
褶皱变平Deepen the crease shadows with ambient occlusion; the garment must read as [厚度描述] fabric.

6、执行流程

  1. 确认两张图是同一件商品——这是最容易犯的错。
  2. 判断原图值不值得救:结构崩坏的重跑生成,只有面料糊的才用本技能。
  3. 写面料特征:查第三节表格,用具体特征替代「质感更好」。
  4. 写两条铁律:只改表面 + 别的不许动。
  5. --quality high(必须) → --batch 2 挑图,落盘到 docs/material-enhancement/。
  6. 并排对比原图与输出:面料是否真的重建了(不是加锐化)、脸和背景是否没动、颜色是否没偏。

7、常见问题

现象原因处理
只是变锐利了未要求重建结构追加「这不是锐化」句,要求逐个线圈重建
人脸/背景也被改未锁定区域追加只允许服装区域变化句
颜色偏了模型顺手调色追加保色句
纹理密度不对(线圈太大/太小)未给尺度参照追加纹理尺度句
褶皱变平、没厚度缺 AO 描述追加加深褶皱暗部句
结构崩坏没被修好超出能力本技能只管表面,结构问题回到生成环节重跑

Tips

Visit https://dlazy.com for more information.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.