Take acommercial photo with blurry fabricand retouch it intoclear and realistic fabric, with the composition and the person completely unchanged.
This is apost-processing skill: the part of AI-generated clothing images that most easily falls apart is the fabric—it looks fine from a distance, but zoom in and the knit turns into a blurry mess. Give it a real high-definition product image as a reference, and rebuild the surface.
Generation effect examples
| Input: original image (to be enhanced) | Input: high-definition product image (real fabric) |
|---|---|
source-image.jpg — worn shot of an army-green cable-knit sweater, 1024×1536 |
hires-product.jpg — high-definition flat-lay image of the same sweater, 800×800 |
Command actually executed:
dlazy gpt-image-2 \
--prompt 'Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo that defines the true fabric. Rebuild the sweater surface in image 1 using the real texture from image 2: crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, correct fold shadows and a matte fibre sheen. Do not change anything else — the model face, hair, hands, pose, brown trousers, background, framing and colour grading must stay identical to image 1. The garment silhouette and colour must not shift; only the material fidelity and micro-detail improve. Photorealistic, sharp, no text, no watermark.' \
--images docs/material-enhancement/source-image.jpg docs/material-enhancement/hires-product.jpg \
--size 1024x1536 --quality high --imageFormat jpeg \
--save docs/material-enhancement/example-output.jpg
Output
example-output.jpg — 1024×1536, 60 credits. The three-dimensional relief of the cable braids, the embossing of the diamond jacquard, the fuzziness of the wool, and the shadow depth in the folds are reconstructed; the model's facial features, hairstyle, hands, pose, brown trousers, gray wall background, and overall color tone remain unchanged, and the sweater's silhouette and army-green color do not shift.
1. Capability boundaries
| Input | Effect |
|---|---|
| Original image | The commercial photo to be improved; determines composition, model, pose, background, and color tone |
| High-definition product image | Provides real fabric information (weave, texture, fuzziness, sheen) |
| Clothing type | Helps determine which surface features should be strengthened |
| Only change | Do not change |
|---|---|
| Fabric surface texture, microscopic details, fold shadow depth, fiber sheen | Composition, model (face/hair/hands/pose), other clothing, background, color tone, clothing silhouette and color |
Do not do: do not change the clothing silhouette or color; do not change the model or background; do not replace one fabric with another (to change fabric, use fabric-on-body）。
2. Input material rules
Before generating, self-check these hard constraints:
- Size:20KB ~ 15MB
- Resolution:greater than 400×400
- Format:jpg / jpeg / png / webp
Original image (✅): an image where the composition and person are both satisfactory and only the fabric is not.
High-definition product image (✅): a high-resolution flat-lay or close-up of the same product, where the weave can be seen clearly.
Notes
| Situation | Description |
|---|---|
| ⚠️ The two images must be of the same product | Two images with different fabrics will cause the material to be mismatched |
| ⚠️ If the original image's color is already off | This skill only handles texture; color deviation must be resolved at the generation stage |
| ❌ The original image is severely broken (wrong fit, chaotic structure) | Material enhancement cannot save structural problems; rerun generation |
| ❌ The original image resolution is extremely low | There is not enough information to locate where the texture should go |
3. The two iron rules of material enhancement
Iron rule one: only change the surface; do not move a single pixel of anything else.
Do not change anything else — the model face, hair, hands, pose, other garments,
background, framing and colour grading must stay identical to image 1.
The garment silhouette and colour must not shift; only material fidelity
and micro-detail improve.
Iron rule two: translate "good texture" into specific surface features, otherwise the model will just apply global sharpening.
| Fabric | Specific features to reconstruct |
|---|---|
| Chunky knit | crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, matte fibre sheen |
| Worsted wool | fine even weave, subtle fibre halo, soft directional sheen |
| Satin | smooth continuous specular gradients along the folds, no texture noise |
| Corduroy | parallel wale ridges with correct pitch, matte pile catching light on the ridge tops |
| Denim | diagonal twill lines, slub irregularity, whiskering at stress folds |
| Leather | grain pores, broad soft creases, low-frequency specular sheen |
| Fleece/fuzz | dense short pile, fuzzy silhouette edge, light scattering rather than specular |
Add one moreFold layers：correct fold shadows with proper ambient occlusion in the creases—this is the source of "looking thick."
4. Tool invocation
This skill uses dLazy's gpt-image-2(image editing model + --quality high; material enhancement requires "zero change in geometry and color, reconstruction of high-frequency surface information," which is a typical two-image reference targeted repair).
How to invoke
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 material-enhancement \
--prompt '<见下方 Prompt 模板>' \
--images <按下表顺序> \
--save output/material-enhancement-<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/material-enhancement.jpg
Parameter conventions (fixed usage for this skill)
| Parameter | Value | Reason |
|---|---|---|
--images |
[原图, 高清商品图] |
The order is image 1 / 2 in the prompt |
--size |
Same aspect ratio as the original image | Enhancement should not change the composition |
--quality |
high（Required） |
The output of this skill is high-frequency detail |
--imageFormat |
jpeg |
General format |
--batch |
2 |
Texture reconstruction has randomness |
--save |
docs/material-enhancement/output-<sku>-enhanced.jpg |
Archived separately from the original image for easy comparison |
Command Examples
# basic call
dlazy gpt-image-2 \
--prompt 'Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo defining the true fabric. Rebuild the garment surface in image 1 using the real texture from image 2: crisp knit relief, visible loops and yarn twist, natural loft, correct fold shadows. Do not change anything else — model, pose, background, framing and colour grading stay identical to image 1. Only material fidelity improves.' \
--images docs/material-enhancement/source-image.jpg docs/material-enhancement/hires-product.jpg \
--size 1024x1536 --quality high
# complex call: 批量给一组生成图做材质后处理
HIRES=docs/material-enhancement/hires-product.jpg
LOCK='Do not change anything else — the model face, hair, hands, pose, other garments, background, framing and colour grading must stay identical to image 1. The garment silhouette and colour must not shift; only material fidelity and micro-detail improve. Photorealistic, sharp, no text, no watermark.'
FEAT='crisp cable-braid relief, visible knit loops and yarn twist, natural wool loft, matte fibre sheen, correct fold shadows with proper ambient occlusion in the creases'
for f in docs/material-enhancement/raw/*.jpg; do
dlazy gpt-image-2 \
--prompt "Texture enhancement pass. Image 1 is the on-model photo to improve. Image 2 is the high-resolution product photo that defines the true fabric. Rebuild the garment surface in image 1 using the real texture from image 2: $FEAT. $LOCK" \
--images "$f" "$HIRES" \
--size 1024x1536 --quality high --imageFormat jpeg \
--save "docs/material-enhancement/enhanced/$(basename $f)"
done
# 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 |
references/model-flags.md |
| All options of the unified entry point | node scripts/gen.mjs --help |
5. Prompt template
Texture enhancement pass.
Image 1 is the on-model photo to improve.
Image 2 is the high-resolution product photo that defines the true fabric.
Rebuild the [服装部位] surface in image 1 using the real texture from image 2:
[第三节表格里对应面料的具体特征],
correct fold shadows with proper ambient occlusion in the creases.
Do not change anything else — the model face, hair, hands, pose, other garments,
background, framing and colour grading must stay identical to image 1.
The garment silhouette and colour must not shift; only material fidelity
and micro-detail improve.
Photorealistic, sharp, no text, no watermark.
Correction sentences appended by problem
| Problem | Append to the end of the prompt |
|---|---|
| It just became sharper overall | This is not sharpening: reconstruct the actual [织法单位] structure, loop by loop, matching image 2. |
| The face/background also changed | Only pixels inside the garment region may change. Everything outside it must be identical to image 1. |
| The color was changed | Preserve the exact garment colour from image 1; do not shift hue, saturation or brightness. |
| The texture density is wrong | Match the texture scale to the garment size in image 1: about [N] knit loops per [尺寸]. |
| Folds became flat | Deepen the crease shadows with ambient occlusion; the garment must read as [厚度描述] fabric. |
6. Execution workflow
- Confirm that the two images are the same product—this is the easiest mistake to make.
- Judge whether the original image is worth saving: rerun generation for structurally broken ones; use this skill only when the fabric alone is blurry.
- Write the fabric features: check the table in section 3, and replace "better texture" with specific features.
- Write the two iron rules: only change the surface + nothing else may be touched.
--quality high(Required) →--batch 2Select images, write to disk atdocs/material-enhancement/。- Compare the original image and the output side by side: whether the fabric was truly reconstructed (not just sharpened), whether the face and background were untouched, and whether the color is unbiased.
7. Common problems
| Symptom | Cause | Handling |
|---|---|---|
| It just became sharper | Structure reconstruction was not requested | Append the "this is not sharpening" sentence and require stitch-by-stitch reconstruction |
| The face/background were also changed | The region was not locked | Append a sentence allowing only the clothing region to change |
| The color drifted | The model adjusted the color along the way | Append a color-preservation sentence |
| Texture density is wrong (coils too large/too small) | No scale reference given | Add a texture scale sentence |
| Wrinkles flattened, no thickness | Missing AO description | Add a sentence to deepen the dark areas of the wrinkles |
| Structural collapse was not fixed | Beyond capability | This skill only handles the surface; structural issues go back to the generation stage and are rerun |
Tips
Visit https://dlazy.com for more information.