One clothing image → local macro close-up. The few images on a product detail page that "prove this garment is well made".
Why it's needed: High-converting product detail pages usually have 2–3 detail shots (collar, cuffs, fabric texture), but shooting macro requires a dedicated lens and lighting setup. This skill derives these close-ups from regular product images.
Example output
| Input: clothing image |
|---|
garment-flatlay.jpg — Flat-lay of an army-green cable-knit sweater, 800×800 |
Command actually executed:
dlazy gpt-image-2 \
--prompt 'Macro detail shot for an e-commerce detail page. Zoom into the shoulder-and-collar area of this olive-green cable-knit sweater and render a photorealistic close-up that fills the frame. Show the ribbed crewneck collar meeting the raglan-style cable panel, individual yarn plies and the twist of the cable braid, the loft of the wool fibres, and soft directional light raking across the surface to reveal depth. Keep the colour and stitch pattern exactly as in the source. Shallow depth of field with the far edge softly out of focus, clean neutral background bokeh, no person, no text, no watermark.' \
--images docs/clothing-detail/garment-flatlay.jpg \
--size 1024x1024 --quality high --imageFormat jpeg \
--save docs/clothing-detail/example-output.jpg
Output
example-output.jpg — 1024×1024, 60 credits. The junction between the collar ribbing and the cable panel, the twist direction of each yarn, and the fuzziness of the wool fibers are all resolved; side lighting makes the relief of the diamond jacquard three-dimensionally visible, while the far end falls into a soft blur.
1. Capability boundaries
| Capability | Description |
|---|---|
| Framing area | Collar ribbing / cuffs / hem / buttons / zipper / pockets / embroidery / print / knit structure / fabric fibers |
| Style control | Reference image (copy the camera angle and lighting of a certain detail shot) or custom prompt |
| Clothing type | Helps the model determine which parts are worth enlarging |
| Generation ratio | 1:1(square detail slot) / 3:4(vertical product detail page) |
Does not do: does not change color, weave, or structure; does not add craftsmanship that isn't in the original image (embroidery that doesn't exist, a zipper that doesn't exist); does not fabricate fabric composition.
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 recommendations
| Practice | Description |
|---|---|
| ✅ The higher the original image resolution, the better | Macro is enlarging the information in the original image; a blurry original = a fabricated detail shot |
| ✅ The target area is clearly visible in the original image | For areas that can't be seen clearly in the original image, the output is the model's imagination |
| ✅ Enlarge only one area at a time | Cramming three close-ups into one image means none of them are clear |
| ❌ Low-resolution / heavily compressed images | Will enlarge into fake, plastic-looking textures |
| ❌ The target area is occluded | Occluded craftsmanship can only be fabricated |
3. Framing area → prompt wording
| Area | Framing description |
|---|---|
| Collar ribbing | the ribbed crewneck collar meeting the body panel, showing rib wale spacing and the seam join |
| Cuffs | the ribbed cuff and the sleeve seam, showing rib elasticity and stitch density |
| Hem | the hem band and side seam, showing hem width and the finishing stitch |
| Buttons | a single button and its buttonhole, showing button material, thread cross and hole finishing |
| Zipper | the zipper teeth and puller, showing tooth pitch, metal finish and the tape stitching |
| Embroidery / print | the [刺绣/印花] motif filling the frame, showing thread direction / print edge sharpness and substrate texture |
| Knit structure | the [麻花/罗纹/提花] stitch structure, showing individual yarn plies and the twist of each loop |
| Fabric fibers | the fabric surface at extreme magnification, showing fibre halo and weave interlacing |
Each one must add three things：
filling the frame ← 特写要占满画面
shallow depth of field with the far edge softly out of focus ← 微距的景深特征
soft directional light raking across the surface to reveal depth ← 侧光才能显出立体纹理
4. Tool invocation
This skill uses dLazy's gpt-image-2(image editing model + --quality high; the entire value of a detail shot is texture fidelity, and this is the one thing this skill cannot skimp on).
How to call
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 clothing-detail \
--prompt '<见下方 Prompt 模板>' \
--images <按下表顺序> \
--save output/clothing-detail-<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-detail.jpg
Parameter conventions (fixed usage for this skill)
| Parameter | Value | Reason |
|---|---|---|
--images |
[原图]; when including a style reference, [原图, 参考图] |
the order is image 1 / 2 in the prompt |
--size |
1024x1024(square image detail slot) / 1024x1536(vertical detail page) |
corresponds to the original site's 1:1 / 3:4 |
--quality |
high（Do not lower it） |
detail images rely on texture to speak for themselves |
--imageFormat |
jpeg |
general format |
--batch |
2 |
framing position has randomness |
--save |
docs/clothing-detail/output-<sku>-<部位>.jpg |
archive by part |
Command Examples
# basic call: 领口细节
dlazy gpt-image-2 \
--prompt 'Macro detail shot for an e-commerce detail page. Zoom into the collar area of this garment and render a photorealistic close-up filling the frame. Show the ribbed crewneck collar meeting the body panel, individual yarn plies and the twist of each loop, soft directional light raking across the surface. Keep colour and stitch pattern exactly as in the source. Shallow depth of field, clean neutral background bokeh, no person, no text.' \
--images docs/clothing-detail/garment-flatlay.jpg \
--size 1024x1024 --quality high
# complex call: 一个 SKU 批量出 3 个部位的细节图
SRC=docs/clothing-detail/garment-flatlay.jpg
COMMON='Render a photorealistic macro close-up filling the frame. Keep the colour and stitch pattern exactly as in the source image. Shallow depth of field with the far edge softly out of focus, soft directional light raking across the surface to reveal depth, clean neutral background bokeh. No person, no text, no watermark.'
for P in collar cuff stitch; do
case $P in
collar) VIEW='Zoom into the collar area: the ribbed crewneck collar meeting the body panel, showing rib wale spacing and the seam join' ;;
cuff) VIEW='Zoom into the cuff area: the ribbed cuff and the sleeve seam, showing rib elasticity and stitch density' ;;
stitch) VIEW='Zoom into the cable-knit panel: the stitch structure, showing individual yarn plies and the twist of each loop' ;;
esac
dlazy gpt-image-2 \
--prompt "Macro detail shot for an e-commerce detail page. $VIEW. $COMMON" \
--images "$SRC" --size 1024x1024 --quality high --imageFormat jpeg \
--save "docs/clothing-detail/output-sku001-$P.jpg"
done
# 先估价不真跑
dlazy gpt-image-2 --dry-run --prompt '...' --images a.jpg --size 1024x1024 --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 |
5. Prompt templates
Macro detail shot for an e-commerce detail page.
Zoom into [部位] of this [品类 + 颜色 + 面料] and render a photorealistic close-up
that fills the frame. Show [第三节表格里的取景描述].
Keep the colour and stitch pattern exactly as in the source.
Shallow depth of field with the far edge softly out of focus,
soft directional light raking across the surface to reveal depth,
clean neutral background bokeh.
No person, no text, no watermark.
Correction sentences appended by issue
| Issue | Append to the end of the prompt |
|---|---|
| Texture looks like plastic | Resolve individual [yarn plies / weave threads / fibre ends]; the surface must read as real textile, not plastic or CG. |
| Not zoomed in enough | Extreme magnification: the [部位] must occupy at least 70% of the frame. |
| Fabricated a nonexistent craft | Do not invent any construction detail that is not visible in the source image. |
| The whole image is solid, with no macro feel | Only the [部位] is in focus; everything beyond [X] must fall into smooth bokeh. |
| Color changed | Sample the colour directly from the source image; no grading, no saturation boost. |
6. Execution flow
- Pick the original image: the higher the resolution, the better; confirm that the target part is clearly visible and unobstructed.
- List the parts: one SKU usually produces 2~3 images (neckline + fabric + one distinctive craft).
- Each prompt zooms in on only one part, copy the framing description from Section 3.
- Complete three things: fill the frame / shallow depth of field / side light.
--quality high(do not downgrade) →--batch 2pick images, write to disk indocs/clothing-detail/。- Quality check: whether the texture is realistic (not plastic-looking), whether any nonexistent craft was fabricated, and whether the color is consistent.
7. Common issues
| Symptom | Cause | Handling |
|---|---|---|
| Plastic-looking texture | Quality tier is low or the original image is blurry | --quality high + append a texture-parsing sentence; switch to a high-resolution original image |
| Not zoomed in enough, still a half-body shot | No proportion specified | Append a sentence specifying 70% frame proportion |
| Fabricated a zipper/embroidery not in the original image | Model completion | Append a no-fabrication sentence |
| No macro depth of field | No depth of field specified | Append a sentence specifying that only the target part is in focus |
| Color is more vivid than the original | Automatic color adjustment | Append a color-picking constraint sentence |
| Three parts squeezed into one image | One prompt wrote multiple parts | Split into multiple prompts, one part per prompt |
Tips
Visit https://dlazy.com for more information.