Provide at onceup to 8 item images, fused intoa complete Look。
and flat-lay 's "multiple garments worn" (only supports two images: top + bottom) is different; this skill isfree combination of any categories: sweater + wide-leg pants + hat + necklace + bag + shoes, producing one complete styling image at a time.
Generation result examples
| Input item 1 | Input item 2 | Input item 3 |
|---|---|---|
item-sweater.jpg — Army green cable-knit sweater |
item-hat.jpg — Colorful woven bucket hat |
item-necklace.jpg — Pearl necklace |
Command actually executed:
dlazy seedream-5.0 \
--prompt '电商搭配商拍图。将参考图中的多件单品组合到同一个模特身上：图1 的军绿色麻花针织圆领毛衣作为上装，图2 的彩色编织渔夫帽戴在头上，图3 的珍珠项链戴在颈部。每件单品必须与参考图完全一致——颜色、织法纹理、图案、材质与细节都不能改。下装自动补一条米白色阔腿长裤，脚穿白色运动鞋。青年亚洲女模特，正面站姿，全身入画，纯浅灰色摄影棚背景，柔和顶光，真实照片质感，无文字无水印。' \
--images docs/image-fusion/item-sweater.jpg docs/image-fusion/item-hat.jpg docs/image-fusion/item-necklace.jpg \
--size 3:4 --resolution 2k \
--save docs/image-fusion/example-output.jpg
Output
example-output.jpg — 3:4 / 2K, 5 credits. All three items are in place at the same time; the sweater's cable-knit pattern, the bucket hat's woven color scheme, and the necklace's bead diameter are all preserved; the bottom and shoes not provided are filled in according to the prompt.
1. Capability boundaries
| Capability | Description |
|---|---|
| Number of items | Up to 8 images |
| Category combination | Tops / bottoms / outerwear / dresses / shoes / bags / hats / scarves / jewelry, any mix and match |
| Reference image (optional) | Determines the model's pose, shooting angle, shot size, scene, and lighting |
| Model image (optional) | Locks the face and body shape, keeping the same model across multiple Looks |
| Automatic filling of missing items | When only a top is provided, the bottom and shoes are automatically styled by the model according to the style |
Not supported: does not change an item's color, pattern, material, or fit; does not provide size recommendations; not for forging others' likeness for endorsements.
2. Input material rules
Before generating, self-check these hard constraints:
- Size:20KB ~ 15MB
- Resolution:Greater than 400×400
- Format:jpg / jpeg / png / webp
This skill's material constraints are looser than other skills:
- Quantity:Up to 8 images
- Size:20KB ~ 15MB
- Resolution:No more than 8192×8192
- Format:jpg / jpeg / png / heic / webp
Input suggestions
| Practice | Description |
|---|---|
| ✅ Put only one item per image | If a top and pants are mixed in one image, the model will not be able to tell which one to wear |
| ✅ Flat-lay images on solid-color/transparent backgrounds | Least interference, most accurate colors |
| ✅ Upload images in wearing order | Top → bottom → outerwear → shoes → bag → accessories; image N in the prompt corresponds to these |
| ❌ Duplicates in the same category | If you upload two tops, the model will only choose one or mix them together |
| ❌ Images of someone already wearing it | Will bring in the original model's body as well |
3. Item list → prompt mapping
The key to image fusion isName each item explicitly: Each image corresponds to a specific wearing position; items that are not named will be ignored.
图1 → 上装：军绿色麻花针织圆领毛衣（作为上装穿在身上）
图2 → 帽子：彩色编织渔夫帽（戴在头上）
图3 → 项链：珍珠项链（戴在颈部）
未提供 → 下装：自动补米白色阔腿长裤
未提供 → 鞋：自动补白色运动鞋
Write as prompt:
将参考图中的多件单品组合到同一个模特身上：
图1 的<单品描述>作为上装，图2 的<单品描述>戴在头上，图3 的<单品描述>戴在颈部。
每件单品必须与参考图完全一致——颜色、织法纹理、图案、材质与细节都不能改。
下装自动补一条<描述>，脚穿<描述>。
Layer conflicts must be explicitly ordered: When providing both an outerwear and a top, write 图2 的外套敞开穿在图1 的上装外面，露出内搭, otherwise the model will choose one of the two.
4. Unify the visual style of the entire set
A shop often needs to produce a dozen or more looks, and the visual style must be unified, otherwise the product detail page looks pieced together. Fix these four items:
| What to fix | How to write it |
|---|---|
| Model | Use the same model image, or fix it in the prompt 青年亚洲女模特，鹅蛋脸，中长黑直发 |
| Pose and shot size | Use the same reference image, or fix 正面站姿，全身入画 |
| Background and lighting | Fix 纯浅灰色摄影棚背景，柔和顶光 |
| Composition and negative space | Fix 人物居中，头顶留白约画面高度 8% |
Only change the item list; do not alter a single word of the rest of the prompt—only then will the whole set of images look like one series.
5. Tool invocation
This skill uses dLazy's seedream-5.0(supports up to 10 reference images + 2K/3K/4K output; the most cost-effective in multi-item constraint scenarios: 5 credits per image, suitable for a shop running dozens of looks).
How to invoke
Two equivalent ways to write it; choose one. The unified entry point will automatically select the backend, retry on failure, create directories and write to disk, and estimate cost:
# 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 image-fusion \
--prompt '<见下方 Prompt 模板>' \
--images <按下表顺序> \
--save output/image-fusion-<sku>.jpg
# B. 直接用 dLazy CLI（不想引入 Node 依赖时，效果等价）
dlazy seedream-5.0 --prompt '...' --images ... --save output/image-fusion.jpg
Parameter conventions (fixed usage for this skill)
| Parameter | Value | Reason |
|---|---|---|
--images |
[单品1, 单品2, …, (参考图), (模特图)], up to 10 images |
The order is the 图1 / 图2 / … in the prompt |
--size |
3:4(main image for full-body looks);1:1(main image for square images) |
Mainly e-commerce vertical format |
--resolution |
2k for everyday;3k/4k for print or large detail images |
2K is already enough for listing |
--batch |
2 ~ 4 |
Multi-item combinations have high randomness; generate a few extra to pick from |
--save |
docs/image-fusion/output-look<N>.jpg |
Write directly to disk |
Command Examples
# basic call: 三件单品融合
dlazy seedream-5.0 \
--prompt '电商搭配商拍图。将参考图中的多件单品组合到同一个模特身上：图1 的毛衣作为上装，图2 的帽子戴在头上，图3 的项链戴在颈部。每件单品必须与参考图完全一致——颜色、织法纹理、图案、材质与细节都不能改。青年亚洲女模特，正面站姿，全身入画，纯浅灰色摄影棚背景，柔和顶光，真实照片质感，无文字无水印。' \
--images docs/image-fusion/item-sweater.jpg docs/image-fusion/item-hat.jpg docs/image-fusion/item-necklace.jpg \
--size 3:4 --resolution 2k
# complex call: 6 件单品 + 姿势参考图 + 固定模特 + 出 4 张挑图
dlazy seedream-5.0 \
--prompt '电商搭配商拍图。图1 的上装、图2 的下装、图3 敞开外穿的外套、图4 的鞋、图5 的手提包（左手提）、图6 的帽子，全部穿戴到同一个模特身上，外套敞开露出图1 的内搭。每件单品与参考图完全一致，颜色、图案、材质与版型不能改。图7 是姿势与场景参考，图8 是模特脸，人脸与身材沿用图8。真实照片质感，无文字无水印。' \
--images docs/image-fusion/top.jpg docs/image-fusion/bottom.jpg docs/image-fusion/coat.jpg docs/image-fusion/shoes.jpg docs/image-fusion/bag.jpg docs/image-fusion/item-hat.jpg docs/image-fusion/pose.jpg docs/image-fusion/model-face.jpg \
--size 3:4 --resolution 3k \
--batch 4 --save docs/image-fusion/output-look01.jpg
# 先估价不真跑
dlazy seedream-5.0 --dry-run --prompt '...' --images a.jpg b.jpg --size 3:4
Further reading
| What to look up | Where to go |
|---|---|
| Authentication, multi-backend configuration, output structure, error codes | references/provider-cli.md |
seedream-5.0 All available parameters of |
references/model-flags.md |
| All options of the unified entry point | node scripts/gen.mjs --help |
6. Prompt template
电商搭配商拍图。将参考图中的多件单品组合到同一个模特身上：
图1 的[单品描述]作为[穿着位置]，
图2 的[单品描述][穿着位置]，
图3 的[单品描述][穿着位置]。
[层次关系，例：图3 的外套敞开穿在图1 的上装外面，露出内搭。]
每件单品必须与参考图完全一致——颜色、织法纹理、图案、材质与细节都不能改。
[缺件补齐，例：下装自动补一条米白色阔腿长裤，脚穿白色运动鞋。]
[模特描述]，[姿势与景别]，[背景]，[光线]，真实照片质感，无文字无水印。
Correction sentences appended by problem
| Problem | Append to the end of the prompt |
|---|---|
| An item is missing | 务必同时出现全部 N 件单品：<逐件重述一遍>，缺一件视为失败。 |
| Two items have mixed colors | 图1 与图2 是两件独立单品，颜色不得互相污染：图1 保持<色>，图2 保持<色>。 |
| Layering is worn incorrectly | 穿着层次由内到外为：图1 → 图3；图3 必须敞开且能看到图1 的领口与前襟。 |
| Pattern position drifts | 图案定位：<单品>的<图案>位于<位置>，尺寸约<描述>，不得移动或重复。 |
| The visual style of the whole set is not unified | Keep the four fixed descriptions from Section 4unchanged word for word, and only replace the item list |
7. Execution process
- Break down the list: List every item in this look, remove duplicates within the same category, and confirm there are no more than 8 items.
- Order the uploaded images: In the order of top → bottom → outerwear → shoes → bag → accessories, the 图N in the prompt and
--imagesmust strictly correspond in order. - Name every item: One sentence per item, clearly state the wearing position; explicitly order any layering relationships.
- Fill in missing items: If no bottom or shoes are provided, specify them in the prompt; do not let the model improvise.
- Fix the visual style of the whole set(four items in Section 4), between multiple Looks only the item list is changed.
--dry-runEstimate → actually run →--batch 2~4select images, save to disk todocs/image-fusion/。- Quality check: whether the item count is complete, whether colors are mixed up, whether the layering is correct, whether the pattern positions have drifted.
8. Common Issues
| Symptom | Cause | Handling |
|---|---|---|
| One item missing | Many items, prompt did not name each one individually | Name each item individually, and add the sentence "missing one item counts as failure" |
| Hat and hair clipping | Missing interaction description | Add 头发从帽檐下自然垂落，帽子与头顶贴合无缝隙 |
| Two items' colors mixed up with each other | Similar colors | Add the sentence "colors must not contaminate each other", and hard-code the color name of each item |
| Outerwear completely covers the inner layer | Layering not specified | Add a layering order sentence, requiring the inner layer's collar and front placket to be visible |
| The whole set of images has inconsistent style | Background/lighting was changed in every prompt | Keep the four descriptions in Section 4 fixed word for word |
| Number of items exceeds 8 | Beyond capability | Split into two Looks, or first use image-fusion to generate the main subject, then use wear-everything to add accessories separately |
Tips
Visit https://dlazy.com for more information.