agentleFS
dlazy-aidlazy-ai/ecommerce-skills/skills/fission-pattern/skill.md

fission-pattern

Skill71 starsChanged 3 months agoMITView on GitHub

一张商品图裂变成整套素材。商品图 + 卖点 → 多角度多场景成套图,够铺满一屏。当用户说「裂变套图」「一张变一屏」「凑够详情页」「出一套图」时使用。

The file

Files
skill.md

fission-pattern — 一张图裂变完整套图

一张商品图 → 一整套不同角度 / 场景 / 构图的商拍图。

电商主图位通常要 5 张,详情页要十几张。本技能解决的是「只有一张图,要凑满一屏」的问题:同一件商品,多个机位与场景,视觉识别保持一致。


生成效果示例

输入:商品图
<img src="../../docs/fission-pattern/product-watch.jpg" width="280">
product-watch.jpg — 黑色鳄鱼纹皮带钢壳银色太阳纹表盘手表

实际执行的命令(套图第 2 张,其余两张只换第三段镜位):

dlazy gpt-image-2 \
  --prompt 'E-commerce product photography, set image 2 of 3 — in-use lifestyle shot. The subject is the watch from the reference image: a polished stainless-steel watch with a silver sunburst dial, applied baton markers and a black crocodile-embossed leather strap. Keep the product 100% faithful: same case shape and polish, same dial colour and marker layout, same hand shapes, same crown, same strap embossing and stitching — it must be recognisably the identical watch as the reference. Show it worn on a man wrist resting on a wooden cafe table beside a white coffee cup, dark suit sleeve and white shirt cuff visible, warm window light, shallow depth of field with a blurred cafe background. Photorealistic, no text, no watermark.' \
  --images docs/fission-pattern/product-watch.jpg \
  --size 1024x1536 --quality medium --imageFormat jpeg \
  --save docs/fission-pattern/example-output-2.jpg

输出:一套三张

1 · 正面主图2 · 场景使用图3 · 细节微距图
<img src="../../docs/fission-pattern/example-output-1.jpg" width="230"><img src="../../docs/fission-pattern/example-output-2.jpg" width="230"><img src="../../docs/fission-pattern/example-output-3.jpg" width="230">
蓝图纸 + 黄铜直尺,冷调侧光手腕佩戴 + 咖啡桌,暖色窗光表盘/刻度/表冠微距,硬光勾边

三张的商品保真段逐字相同,只有镜位段在变。


1、能力边界

模式说明
商品套图同一商品 → 正面主图 / 45 度图 / 场景使用图 / 细节微距图 / 尺寸对比图
姿势套图同一模特同一穿搭 → 正面 / 侧面 / 背面 / 走动 / 坐姿
输入说明
商品图1 张,主视角最佳
商品名称例:撞色长款风衣
商品卖点例:100% 纯棉,轻盈舒适透气,法式复古撞色元素(用于决定场景与氛围)

不做:不改商品的外形、颜色、材质与结构;不编造商品没有的功能卖点;不生成虚假促销信息。


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

输入建议

做法说明
✅ 主视角 + 纯净背景越干净,整套图的商品一致性越高
✅ 卖点写具体防水防汗 会带出运动场景,法式复古 会带出咖啡馆场景
✅ 商品结构完整可见套图里的细节图要靠这张图推断结构
❌ 已经带营销文字的图文字会被复制到每张套图里,先用 remove-watermark 洗干净
❌ 商品被手/道具遮挡遮住的部分在每张套图里都会不一样

3、套图配方:5 张主图位怎么排

把「一套图」拆成固定的镜位清单,每张一条 prompt,商品描述段完全复用,只换镜位段:

#镜位作用镜位段示例
1正面主图搜索列表首图,要最清楚straight-on hero shot filling the frame, clean seamless background, even studio light
245 度立体图交代体积与厚度45-degree three-quarter angle showing depth and side profile, soft gradient background
3场景使用图建立使用联想in-use lifestyle shot: [场景 + 人物动作], warm window light, shallow depth of field
4细节微距图证明材质与工艺macro close-up of [关键工艺部位] filling the frame, hard rim light, extreme detail
5对比 / 内构图交代尺寸或内部[尺寸对比物 / 内部结构] shown alongside the product, top-down layout, neutral background

一致性的关键:三段结构里,第一段(商品保真描述)在 5 条 prompt 里逐字相同,只有第三段(镜位)在变。

[段1 商品保真:不变]  +  [段2 卖点氛围:不变]  +  [段3 镜位:每张不同]

4、卖点 → 场景的映射

卖点决定第 3 张场景图长什么样,别让模型自由发挥:

卖点类型场景写法
保暖 / 加厚snowy outdoor street, breath visible, cold blue ambient with warm rim light
透气 / 速干gym or running track, dynamic mid-motion, bright daylight
防水 / 三防rainy pavement with water droplets beading on the surface, overcast light
通勤 / 商务office lobby or subway station, dark suit context, cool neutral light
法式 / 复古French cafe interior, marble table, warm window light, film colour grading
亲肤 / 婴童soft nursery bedding, pastel palette, very soft diffused light
精密 / 工艺dark navy blueprint paper with a brass ruler, cool directional side light

姿势套图用的是另一组:正面站姿 / 侧身转头 / 背面回头 / 自然行走 / 坐姿——每条只换姿势段,模特与穿搭段逐字不变(写法可直接复用 creative-scene 的改姿势句式)。


5、工具调用

本技能使用 dLazy 的 gpt-image-2(套图的核心指标是整套图里的商品是同一件,需要最强的参考图保真;实测低价模型在多场景切换时会漂移出另一款商品)。

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 fission-pattern \
  --prompt '<见下方 Prompt 模板>' \
  --images <按下表顺序> \
  --save output/fission-pattern-<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/fission-pattern.jpg

参数约定(本技能固定用法)

参数取值理由
--images[商品图](全套 N 条命令都传同一张)保证整套图的商品同源
--size1024x1536 竖版套图;1024x1024 方图主图位跟随平台主图规范
--qualitymedium 场景图;high 细节微距图微距图靠纹理说话
--imageFormatjpeg电商上架通用格式
--batch1(套图靠多条 prompt,不靠 batch)batch 只会给同一镜位多个版本
--savedocs/fission-pattern/output-<sku>-<序号>.jpg按套编号归档

成本提示:想压成本可换 dlazy seedream-5.0(单张约 1/6 价)。代价是商品保真度下降——实测在场景切换时会漂移成另一款商品,只适合对商品一致性不敏感的氛围图。

Command Examples

# basic call: 套图第 1 张(正面主图)
dlazy gpt-image-2 \
  --prompt 'E-commerce product photography, set image 1 of 5 — hero front shot. The subject is the product in the reference image. Keep it 100% faithful: same shape, colour, material texture and structural details. Straight-on hero shot filling the frame, clean seamless background, even studio light. Photorealistic, no text, no watermark.' \
  --images docs/fission-pattern/product-watch.jpg \
  --size 1024x1536 --quality medium

# complex call: 用 shell 循环一次跑完整套 5 张
PRODUCT='a polished stainless-steel watch with a silver sunburst dial, applied baton markers and a black crocodile-embossed leather strap'
KEEP="Keep the product 100% faithful: same case shape and polish, same dial colour and marker layout, same hand shapes, same crown, same strap embossing and stitching."
i=0
for SHOT in \
  'straight-on hero shot filling the frame, clean seamless background, even studio light' \
  '45-degree three-quarter angle showing depth and side profile, soft gradient background' \
  'in-use lifestyle shot: worn on a man wrist beside a coffee cup on a wooden cafe table, warm window light, shallow depth of field' \
  'macro close-up of the dial edge, applied markers and knurled crown filling the frame, hard rim light, extreme detail' \
  'top-down flat layout beside a brass ruler for scale, dark navy blueprint paper, cool side light'
do
  i=$((i+1))
  dlazy gpt-image-2 \
    --prompt "E-commerce product photography, set image $i of 5. The subject is $PRODUCT from the reference image. $KEEP $SHOT. Photorealistic commercial product photography, no text, no watermark." \
    --images docs/fission-pattern/product-watch.jpg \
    --size 1024x1536 --quality medium --imageFormat jpeg \
    --save docs/fission-pattern/output-sku001-$i.jpg
done

# 先估价不真跑(乘以套图张数就是总价)
dlazy gpt-image-2 --dry-run --prompt '...' --images a.jpg --size 1024x1536

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

6、Prompt 模板

三段结构,前两段整套复用,第三段每张不同:

E-commerce product photography, set image [N] of [总数].

【段1 · 商品保真,整套逐字相同】
The subject is [商品名称 + 颜色 + 材质] from the reference image.
Keep the product 100% faithful: same [外形], same [颜色], same [材质纹理],
same [结构细节:五金/缝线/图案/logo 位置] — it must be recognisably the identical product
across the whole set.

【段2 · 卖点氛围,整套逐字相同】
[从第四节表格取对应的氛围与色调描述]

【段3 · 镜位,每张不同】
[从第三节表格取镜位描述]

Photorealistic commercial product photography, no text, no watermark.

按问题追加的修正句

问题追加到 prompt 末尾
套图里像两件不同商品Cross-check against the reference image: [关键识别特征] must match exactly. Any deviation is a failure.
场景抢了商品的戏The product must occupy at least 40% of the frame and be the sharpest element; keep the environment subordinate and softly defocused.
整套色调不统一Grade the whole set consistently: [色温 + 对比度描述].
细节图糊改 --quality high,并追加 resolve individual [纹理单位:yarn plies / gear teeth / leather pores]

7、执行流程

  1. 洗干净输入:商品图上如有营销文字,先走 remove-watermark。
  2. 写商品保真段:把商品的关键识别特征列全(外形 / 颜色 / 材质 / 五金 / 缝线 / logo 位置)——这段整套复用。
  3. 卖点 → 氛围段:查第四节表格,把卖点翻译成场景与色调,整套复用。
  4. 列镜位清单:查第三节,按平台主图位数量取 5 条(或详情页取 8~12 条)。
  5. 循环跑:用第四节的 shell 循环,每条只换镜位段。
  6. 质检整套:把 N 张图并排看——商品是不是同一件?色调是否统一?有没有哪张场景抢戏?
  7. 不合格的单张重跑,不用整套重来。

8、常见问题

现象原因处理
套图里的商品明显不是同一件商品保真段太笼统,或用了低保真模型列全关键识别特征;换 gpt-image-2
每张色调都不一样未统一氛围段氛围段整套逐字相同,并追加统一调色句
场景太满,商品变小场景描述比商品描述更长追加 40% 画面占比约束句
细节图看不出材质--quality medium改 high 并写明要解析到的纹理单位
套图里出现了营销文字输入图自带文字先洗图;prompt 末尾保留 no text, no watermark
想出 12 张但成本太高张数 × 单价主图位 5 张用 gpt-image-2,详情页氛围图用 seedream-5.0 补量

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.