The outfit stays the same; the person / scene / pose all change, switched to a social-platform grass-planting style.
Difference from one-shot : one-shot targets e-commerce main images (maintains studio-shoot conventions, composition unchanged), while this skill targetscontent grass-planting— what it wants is a sense of life, a candid feel, and ambient light; both composition and shot size can change.
Generation effect examples
| Input: original outfit image | Input: scene/model reference image |
|---|---|
source-outfit.jpg — light gray knit dress + pearl necklace + tote bag + silver high heels, fountain courtyard |
scene-reference.jpg — street shot at a café entrance, hand touching hair, dappled tree shadows and warm light |
Command actually executed:
dlazy gpt-image-2 \
--prompt 'Lifestyle social-commerce photo. Image 1 shows the outfit to keep: a light-grey textured sleeveless knit mini dress with a mock neckline, a pearl choker, a cream-and-tan tote bag and silver pointed heels. Image 2 is the model, pose, scene and lighting reference: a young woman on a sunlit tree-lined street outside a coffee shop, hand in her hair, warm dappled daylight, shallow depth of field. Put the complete outfit from image 1 onto the model from image 2. Keep every garment detail faithful: same grey knit texture, same neckline and hem length, same pearl choker, same tote bag colour blocking. Reproduce image 2 for the model identity, pose, camera angle, crop, street background and colour grading. Photorealistic influencer-style photo, no text, no watermark.' \
--images docs/clothing-grass-planting/source-outfit.jpg docs/clothing-grass-planting/scene-reference.jpg \
--size 1024x1536 --quality medium --imageFormat jpeg \
--save docs/clothing-grass-planting/example-output.jpg
Output
example-output.jpg — 1024×1536. The dress's light gray knit texture, stand collar, waistline, and skirt length, the pearl necklace's bead diameter, and the tote bag's off-white/brown color-blocking are all preserved; the model, the hair-touching pose, the café street scene, the dappled tree shadows and warm light, and the shallow depth of field are copied from the reference image.
1. Capability boundaries
| Stays unchanged | Will change |
|---|---|
| The color, pattern, material, cut, and pairing relationships of tops / bottoms / shoes / bags / accessories | Model identity, pose and action, scene and background, lighting and color tone, camera position and shot size |
| Control method | Description |
|---|---|
| Reference image | Copy the model, scene, pose, and lighting of a certain grass-planting image |
| Custom prompt | Describe the target scene and action in text |
Do not do: do not change any single item of the outfit; do not generate face-swapped images of specific real people; do not use it to forge another person's likeness endorsement or fake usage experience.
2. Input material rules
Before generating, self-check these hard constraints:
- Size:20KB ~ 15MB
- Resolution:greater than 400×400
- Format:jpg / jpeg / png / webp
Input suggestions
| Practice | Description |
|---|---|
| ✅ Full-body or mostly full-body outfit image | Grass-planting images usually need to show the whole outfit |
| ✅ Every single item is clearly distinguishable | Items that cannot be seen clearly will be redrawn after the scene change |
| ✅ Single person | With multiple people in frame, the model cannot tell which is the subject |
| ❌ Product is severely occluded | The occluded parts will inevitably deform after the pose change |
| ❌ Already has filters/text stickers | They will be carried into the new image together; clean them off first |
3. Scene recipe for grass-planting images
The persuasiveness of a grass-planting image comes from "looking like it was shot casually". Four elements are indispensable:
| Element | Writing example |
|---|---|
| Scene | a sunlit tree-lined street outside a coffee shop / a bright bedroom with sheer curtains / a seaside boardwalk at golden hour |
| Action | hand in her hair / mid-stride walking toward the camera / sitting on a step holding a coffee cup |
| Lighting | warm dappled daylight through leaves / soft window light with visible haze / low golden-hour backlight with rim glow |
| Camera language | shallow depth of field, shot on 50mm, slight film grain, natural colour grading |
One outfit spread across multiple posts: fix the outfit description paragraph, and only change these four elements → one outfit produces 5 scene versions, covering different content themes (commute / date / travel / home / sport).
Outfit lock sentence(required, name each item one by one):
Keep every garment detail faithful: same [上装描述], same [下装描述],
same [鞋], same [包], same [配饰] — colour, pattern, texture and fit unchanged.
4. Tool invocation
This skill uses dLazy's gpt-image-2(image editing model; this skill requires "faithful preservation of every item in the full outfit + complete change of person and environment", which is the most demanding type of task for instruction following).
How to invoke
Two equivalent ways of writing it; choose one. The unified entry point automatically selects the backend, retries on failure, creates directories and writes to disk, and estimates 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 clothing-grass-planting \
--prompt '<见下方 Prompt 模板>' \
--images <按下表顺序> \
--save output/clothing-grass-planting-<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/clothing-grass-planting.jpg
Parameter conventions (fixed usage for this skill)
| Parameter | Value | Reason |
|---|---|---|
--images |
[原穿搭图, 场景/模特参考图]; when describing the scene purely in text, pass only the original image |
The order is image 1 / 2 in the prompt |
--size |
1024x1536(vertical format for social platforms) |
Xiaohongshu/Douyin are mainly 3:4, 9:16 |
--quality |
medium conventional;high when fabric is the selling point |
For种草 images, the texture requirement is slightly lower than for main images |
--imageFormat |
jpeg |
General format |
--batch |
3 ~ 4 |
Poses and hands are highly random |
--save |
docs/clothing-grass-planting/output-<场景>.jpg |
Archive by content theme |
Command Examples
# basic call: 用参考图换模特+场景+姿势
dlazy gpt-image-2 \
--prompt 'Lifestyle social-commerce photo. Image 1 shows the outfit to keep. Image 2 is the model, pose, scene and lighting reference. Put the complete outfit from image 1 onto the model from image 2. Keep every garment detail faithful — colour, pattern, texture and fit unchanged. Reproduce image 2 for model identity, pose, camera angle, crop, background and colour grading. Photorealistic influencer-style photo, no text, no watermark.' \
--images docs/clothing-grass-planting/source-outfit.jpg docs/clothing-grass-planting/scene-reference.jpg \
--size 1024x1536 --quality medium
# complex call: 一套穿搭铺 5 个内容场景（纯文字描述场景）
SRC=docs/clothing-grass-planting/source-outfit.jpg
OUTFIT='Keep the complete outfit from the source image faithful: same light-grey textured sleeveless knit mini dress with mock neckline, same pearl choker, same cream-and-tan tote bag, same silver pointed heels — colour, pattern, texture and fit unchanged.'
CAM='Photorealistic influencer-style photo, shallow depth of field, shot on 50mm, slight film grain, natural colour grading, no text, no watermark.'
for S in street cafe home travel commute; do
case $S in
street) SCENE='on a sunlit tree-lined street, mid-stride walking toward the camera, warm dappled daylight through the leaves' ;;
cafe) SCENE='sitting at a marble cafe table holding a coffee cup, soft window light with visible haze' ;;
home) SCENE='standing by a bright bedroom window with sheer curtains, hand resting on the frame, soft diffused morning light' ;;
travel) SCENE='on a seaside boardwalk at golden hour, hair moving in the wind, low backlight with rim glow' ;;
commute) SCENE='in a modern office lobby, walking past glass panels, cool even daylight' ;;
esac
dlazy gpt-image-2 \
--prompt "Lifestyle social-commerce photo. Replace the model, pose, scene and lighting: a young woman $SCENE. $OUTFIT $CAM" \
--images "$SRC" --size 1024x1536 --quality medium --imageFormat jpeg \
--batch 3 --save "docs/clothing-grass-planting/output-$S.jpg"
done
# 先估价不真跑
dlazy gpt-image-2 --dry-run --prompt '...' --images a.jpg b.jpg --size 1024x1536
Further reading
| What to look up | Where to look |
|---|---|
| Authentication, multi-backend configuration, output structure, error codes | references/provider-cli.md |
gpt-image-2 all available parameters of |
references/model-flags.md |
| All options of the unified entry point | node scripts/gen.mjs --help |
5. Prompt templates
Lifestyle social-commerce photo.
Image 1 shows the outfit to keep: [逐件描述：上装 / 下装 / 鞋 / 包 / 配饰].
Image 2 is the model, pose, scene and lighting reference. ← 用参考图时保留这行
Put the complete outfit from image 1 onto the model from image 2.
Keep every garment detail faithful: same [上装], same [下装], same [鞋], same [包],
same [配饰] — colour, pattern, texture and fit unchanged.
Reproduce image 2 for the model identity, pose, camera angle, crop, background
and colour grading.
（纯文字版改为：Replace the model, pose, scene and lighting: [场景 + 动作 + 光线]）
Photorealistic influencer-style photo, shallow depth of field, shot on 50mm,
slight film grain, natural colour grading, no text, no watermark.
Correction sentences appended by problem
| Problem | Append to the end of the prompt |
|---|---|
| A certain item was replaced | Name it separately once more, and append This item must appear unchanged; substituting it is a failure. |
| Looks too much like a studio shoot, no sense of life | Candid snapshot feel: slightly off-centre framing, motion in the hair or fabric, imperfect natural light. |
| Pose too stiff | Natural mid-action pose, weight on one leg, relaxed hands, not posing straight at the camera. |
| Hands are broken | Hands must be anatomically correct — five distinct fingers, natural knuckles. |
| Scene steals the spotlight | The outfit must remain the visual subject; keep the background softly defocused. |
6. Execution flow
- Clean the input thoroughly: remove filters and text stickers (remove-watermark）。
- List the outfit item by item: tops / bottoms / shoes / bags / accessories—one sentence per item; reuse this whole block.
- Set the content theme: commute / date / travel / home / sport; each theme corresponds to a set of scenes + actions + lighting (recipes in section 3).
- Add camera language: 50mm, shallow depth of field, slight grain—this is the key to "looking like a casual shot".
--batch 3~4Generate multiple images and pick the one with a natural pose, then save to disk atdocs/clothing-grass-planting/。- Quality check: whether all items are present, whether any were replaced, hands, whether there is enough sense of life, and whether the background steals the spotlight.
7. Common problems
| Symptom | Cause | Handling |
|---|---|---|
| Bag or shoes replaced with something else | Did not name each item | Name each item + append the sentence "replacement means failure" |
| Still looks like a studio shoot | Missing sense-of-life description | Append a candid-feel sentence + camera language |
| Pose is stiff like an ID photo | No action described | Append a natural dynamic pose sentence |
| Fingers are broken | Change pose and redraw the hands | Append a hand-fixing sentence + --batch 4 |
| Skirt length/neckline changed | Fidelity items written too vaguely | Specify the neckline shape and hem length explicitly |
| A set of images has inconsistent style | The camera language paragraph is changed every time | The entire camera language paragraph remains unchanged word for word |
Tips
Visit https://dlazy.com for more information.