Lazada Ecommerce Image Generation and Editing

Build mobile-first catalog images that can be localized across Lazada marketplaces without mixing SKU facts, language or campaign information. Use public_model_nano_banana_pro for generation and editing.

Separate three layers

1. Product layer: SKU shape, variant, packaging, logo and included items.
2. Market layer: customer setting, styling, language-safe space and units.
3. Campaign layer: seller-approved headline, voucher or event information added after generation.

Lock the product layer first. Never bake an unverified campaign price, rating, seller badge or warranty into generated pixels.

Scenarios and commands

1. Mobile catalog hero

python3 "$SKILL_PATH/scripts/imagegen.py" generate \
  --prompt 'Lazada mobile catalog hero for the reference SKU. Preserve exact product shape, variant color, package, logo and included pieces. Show the complete item large on a clean high-contrast background with realistic shadow and room for responsive crop. No price, voucher, rating, LazMall badge or unprovided text.' \
  --image /path/to/sku.png

2. Specification gallery base

python3 "$SKILL_PATH/scripts/imagegen.py" generate \
  --prompt 'Create a 4-image Lazada gallery base: full product, control or material detail, scale-in-use, and package contents. Keep the same exact SKU and lighting system. Reserve clean areas for verified measurements and translated copy; do not generate numbers, compatibility or claims.' \
  --image /path/to/product-front.png \
  --image /path/to/package-contents.jpg \
  --batch 4

3. Campaign tile without fake offer

python3 "$SKILL_PATH/scripts/imagegen.py" generate \
  --prompt 'Lazada campaign tile background for a seller promotion. Accurate product on the right, energetic geometric color system on the left, clear empty zones for an approved event name, voucher and CTA. Do not generate a sale price, countdown, discount percentage, platform logo or seller badge.' \
  --image /path/to/product.png

4. Indonesia and Vietnam localization bases

python3 "$SKILL_PATH/scripts/imagegen.py" generate \
  --prompt 'Create two localized lifestyle bases for the same Lazada SKU: one credible Indonesian household context and one Vietnamese household context. Preserve product, packaging and use method exactly; change environment and caption-safe space only, leaving all localized words and offers blank for native review.' \
  --image /path/to/product.png \
  --batch 2

5. Variant matrix

python3 "$SKILL_PATH/scripts/imagegen.py" generate \
  --prompt 'Generate catalog images for the merchant-confirmed red, navy and cream variants. Lock camera angle, scale, background, shadow, structure and packaging; change only the true variant color. One SKU per image, no mixed bundles or added accessories.' \
  --image /path/to/master-sku.png \
  --batch 3

QA

• SKU and marketplace are encoded in filenames and never mixed.
• Product layer remains identical across localized versions.
• Units, specification text and translated copy are added from approved data.
• Package contents and variant colors match the seller sheet.
• No generated campaign offer, rating, badge, warranty or compliance claim.
• Validate current Lazada image and campaign rules per marketplace.

Run

pip3 install requests
python3 "$SKILL_PATH/scripts/imagegen.py" init --skill-name lazada-ecommerce-image-generation-editing
python3 "$SKILL_PATH/scripts/imagegen.py" task --task-id <taskId>

Use repeatable references, batch generation, model parameters, routing and submit-only mode as needed. Check AI Hive live configuration and cost before a multi-market batch.