Turn a product image ofshoes / bags / watches / glasses / hats / scarves / necklaces / earringsinto a commercial photo ofcorrect real-person wearing.
Difference from flat-lay : flat-lay handles "clothes worn on the body", while this skill handles "things worn on the body" — the product occupies only a small area of the frame, so **selection (where the product appears in the reference image)** is the key to success or failure.
Generation effect examples
| Input: product image | Input: reference image |
|---|---|
product-sunglasses.jpg — tortoiseshell square-frame sunglasses, 768×1024 |
model-reference.jpg — young woman, front-facing, night street shot, 768×1024 |
Command actually executed:
dlazy gpt-image-2 \
--prompt 'On-model accessory product photography. Image 1 is the product: a pair of tortoise-brown rectangular sunglasses with dark grey lenses. Image 2 is the model/scene reference. Put the sunglasses from image 1 onto the face of the model in image 2, correctly seated on the nose bridge and ears with natural perspective, realistic lens reflections and a soft shadow on the cheekbones. Keep the product 100% faithful: identical frame shape, tortoise-brown acetate color and grain, hinge and temple design, lens tint. Change nothing else — face, hair, scarf, coat, bag chain, night street background, colour grading and crop must stay pixel-identical to image 2. Photorealistic, no text, no watermark.' \
--images docs/wear-everything/product-sunglasses.jpg docs/wear-everything/model-reference.jpg \
--size 1024x1536 --quality medium --imageFormat jpeg \
--save docs/wear-everything/example-output.jpg
Output
example-output.jpg — 1024×1536. The sunglasses are placed according to the correct nose bridge/ear hook relationship, and the tortoiseshell texture of the frame and lens color are preserved; the model's facial features, the scarf's houndstooth pattern, the night street, and the color tone remain unchanged.
1. Capability boundaries
| Capability | Description |
|---|---|
| Single-view image | Upload 1 product image (front or main view) |
| Multi-view images | Upload multiple angles of the same product to help the model understand the 3D structure (shoe side + sole, bag front + interior) |
| Reference image | Provides the real-person model, pose, scene, and lighting |
| Selection | Frame the position where the product should appear in the reference image (eyes / wrist / neck / feet / shoulder-back / top of head) |
| Supported categories | Shoes, bags, watches, glasses and sunglasses, hats, scarves, necklaces, earrings, belts, gloves, etc. |
Not done: does not change the product's shape, color, hardware, logo, or lens color; does not handle clothing (use flat-lay); not used to forge another person's portrait endorsement.
2. Input material rules
Before generation, self-check these hard constraints:
- Size:20KB ~ 15MB
- Resolution:greater than 400×400
- Format:jpg / jpeg / png / webp
Recommended input types (✅)
| Type | Description |
|---|---|
| Fashion leather shoes | A single shoe or a pair fully in frame, main view |
| Elegant watch | Watch face clear from the front, strap complete |
| Pearl necklace | Laid out or hung, chain complete |
| Bucket hat | Hat shape complete, texture clear |
| Sunscreen sunglasses | Frame unfolded, lens color realistic |
Inputs that will noticeably reduce the effect (❌)
| Problem | Description |
|---|---|
| Incomplete product | A shoe with only half captured, a necklace with the clasp cropped out |
| Too many hanging items | A bag covered with charms and silk scarves, making it hard for the model to distinguish the main subject |
| Unclear product | Blurry, overexposed, metal reflections blurred into a single patch |
3. Region selection: the most important parameter of this skill
The original site draws a box directly on the reference image; this skill instead uses the prompt to explicitly specify the anatomical location + wearing pose, which is equivalent and more controllable:
| Category | Region description (written into the prompt) |
|---|---|
| Glasses / sunglasses | seated on the nose bridge and hooked over both ears, temples visible along the cheekbones |
| Watch | on the left wrist, dial facing the camera, strap closed with the buckle on the underside |
| Necklace | around the neck, clasp at the nape, pendant centred on the collarbone |
| Ear jewelry | on the visible earlobe, hanging naturally with correct scale |
| Hat | on the head, brim angle matching the head tilt, hair falling naturally around it |
| Scarf | wrapped twice around the neck, fringe hanging over the chest |
| Shoes | on both feet, soles contacting the ground with correct perspective and grounded shadows |
| Bag | held in the right hand / on the left shoulder, strap resting on the shoulder line with fabric compression |
You must also hard-code "no changes allowed in other regions", otherwise the model will casually redraw the face and background:
Change nothing else — face, hair, clothing, background, colour grading and crop
must stay pixel-identical to image 2.
4. Reference image selection dimensions
- Category:
女装 / 男装 / 童装; Region:国内 / 海外; Type:电商 / 种草 - Filter: Gender
男 / 女; Age婴儿 / 小童 / 大童 / 青少年 / 青年人 / 中年人 / 老年人; Skin tone欧美人 / 非洲人 / 亚洲人 / 其他肤色
Selection advice:
- The wearing area must be exposed. For glasses, the face must be frontal or slightly turned; for a watch, the wrist must be in frame and not covered by the cuff; for shoes, the feet must be fully visible and not blocked by the hem.
- It is best if there is no similar product in the original image. When the model in the reference image is already wearing another pair of glasses, the model tends to stack two pairs—prefer reference images with empty hands and a bare face, or explicitly state in the prompt
replace the existing sunglasses。 - The lighting direction should be close to that of the product image, otherwise metal reflections will look unnatural.
5. Tool invocation
This skill uses dLazy's gpt-image-2(an image editing and compositing model with multi-image reference, up to 5 reference images; accessory wearing is local replacement and requires strong instruction following and region preservation capabilities).
Invocation method
Two equivalent ways to write 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 wear-everything \
--prompt '<见下方 Prompt 模板>' \
--images <按下表顺序> \
--save output/wear-everything-<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/wear-everything.jpg
Parameter conventions (fixed usage for this skill)
| Parameter | Value | Reason |
|---|---|---|
--images |
[商品图, 参考图]; for multiple views, [商品图1, 商品图2, 参考图] |
the order is image 1 / 2 / 3 in the prompt |
--size |
1024x1536(half-body/full-body wearing);1024x1024(close-ups such as wrists and ear jewelry) |
Choose according to shot size |
--quality |
high(metal, lenses, leather reflections);medium(fabric accessories) |
Reflective materials require a high level |
--imageFormat |
jpeg |
Common format for e-commerce listings |
--batch |
2 ~ 4 |
Wearing position is random, generate a few extra images to pick from |
--save |
docs/wear-everything/output-<sku>.jpg |
Write directly to disk |
Command Examples
# basic call: 单视角（商品图 + 参考图）
dlazy gpt-image-2 \
--prompt 'On-model accessory photo. Image 1 is the product, image 2 is the model/scene reference. Put the product from image 1 onto the model in image 2, seated on the nose bridge and hooked over both ears. Keep the product identical in shape, colour, material and logo. Change nothing else — face, hair, clothing, background and crop stay pixel-identical to image 2.' \
--images docs/wear-everything/product-sunglasses.jpg docs/wear-everything/model-reference.jpg \
--size 1024x1536 --quality high
# complex call: 多视角商品 + 一次出 4 张挑图 + 直接落盘
dlazy gpt-image-2 \
--prompt 'On-model accessory photo. Image 1 is the product front view, image 2 is the product side view, image 3 is the model/scene reference. Put the shoes onto both feet of the model in image 3, soles contacting the ground with correct perspective and grounded shadows. Preserve the patent finish, brogue perforation, lacing and lug sole exactly. Change nothing else — the model, clothing, pose, background and colour grading stay identical to image 3. Photorealistic, no text, no watermark.' \
--images docs/wear-everything/shoe-front.jpg docs/wear-everything/shoe-side.jpg docs/wear-everything/model-reference.jpg \
--size 1024x1536 --quality high --imageFormat jpeg \
--batch 4 --save docs/wear-everything/output-sku001.jpg
# 先估价不真跑
dlazy gpt-image-2 --dry-run --prompt '...' --images a.jpg b.jpg --size 1024x1536 --quality high
Further reading
| What to look up | Where to go |
|---|---|
| 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 |
6. Prompt template
On-model accessory product photography.
Image 1 is the product: [品类 + 颜色 + 材质，例：a pair of tortoise-brown rectangular sunglasses with dark grey lenses].
Image 2 is the model / scene reference.
Put the product from image 1 onto the model in image 2, [选区描述，见第三节表格],
with natural perspective, correct scale, realistic [反光/投影描述] and a soft contact shadow.
Keep the product 100% faithful: identical [外形], [颜色与材质], [五金/镜片/纹理], [logo 位置].
Change nothing else — face, hair, clothing, accessories, background, colour grading
and crop must stay pixel-identical to image 2.
Photorealistic, no text, no watermark.
Correction sentences appended by problem
| Problem | Append to the end of the prompt |
|---|---|
| Wearing position is off | The product must be anatomically correctly placed — re-check the [部位] alignment against the model's [参照点]. |
| Product proportions are wrong | Match real-world scale: the product's width must be about [X] of the [参照部位] width. |
| Two similar products appeared | Remove the existing [同类商品] the model is wearing and replace it with the product from image 1. |
| Face/background was altered | Change only the [部位] region. Keep every other pixel identical to image 2. |
| Metal/lens reflections look fake | Render physically plausible reflections: [环境] reflected on the [材质], consistent with the light direction in image 2. |
7. Execution flow
- Validate input: size / resolution / format; remove incomplete, overly cluttered with pendants, or blurry product images.
- Determine the view: Simple structure (glasses, necklace) → a single viewpoint is enough; complex structure (shoes, bags) → upload 2~3 angles.
- Choose reference images: Confirm that the wearing area is fully exposed, not blocked by similar products, and that the light direction is close to that of the product image.
- Write the selection: Take the corresponding sentence from the table in Section 3, and be sure to also write "no other areas may be changed."
--dry-runEstimate, confirm the credits, then actually run it.--batch 2~4Generate multiple images and select, save to disk atdocs/wear-everything/。- Quality check: wearing position, product proportions, plausibility of reflections, whether two products are overlaid, and whether the face has been altered.
8. Common Issues
| Symptom | Cause | Solution |
|---|---|---|
| Glasses are crooked / floating | Anatomical position not specified | Add the selection sentence from Section 3 to clarify the relationship between the bridge of the nose and the ear hooks |
| Shoes are not standing on the ground | Missing ground-contact constraint | Append soles contacting the ground with correct perspective and grounded shadows |
| Product size is obviously too large or too small | The model has no scale reference | Append a proportion sentence giving the width ratio to the reference body part |
| Two pairs of glasses appear in the image | The model in the reference image is already wearing them | Append replace the existing …, or switch to a blank-face reference image |
| The face was replaced | The model redrew the entire image | Append Change only the [部位] region. Keep every other pixel identical to image 2. |
| Metal reflections look like plastic | --quality medium |
Change --quality high |
| The bag's interior structure is messed up | Only the front was provided | Switch to multi-view images and add interior/side views |
Tips
Visit https://dlazy.com for more information.