Zapytanie AI #362

{
    "text": {
        "format": {
            "name": "product_data_batch",
            "type": "json_schema",
            "schema": {
                "type": "object",
                "required": [
                    "products"
                ],
                "properties": {
                    "products": {
                        "type": "array",
                        "items": {
                            "type": "object",
                            "required": [
                                "ean",
                                "images"
                            ],
                            "properties": {
                                "ean": {
                                    "type": "string"
                                },
                                "images": {
                                    "type": "array",
                                    "items": {
                                        "type": "object",
                                        "required": [
                                            "url",
                                            "source_page",
                                            "type"
                                        ],
                                        "properties": {
                                            "url": {
                                                "type": "string"
                                            },
                                            "type": {
                                                "enum": [
                                                    "main",
                                                    "gallery",
                                                    "detail",
                                                    "lifestyle"
                                                ],
                                                "type": "string"
                                            },
                                            "source_page": {
                                                "type": "string"
                                            }
                                        },
                                        "additionalProperties": false
                                    }
                                }
                            },
                            "additionalProperties": false
                        }
                    }
                },
                "additionalProperties": false
            },
            "strict": true
        }
    },
    "input": [
        {
            "role": "system",
            "content": "You are an aggressive image harvester for a sports e-commerce store. Your PRIMARY GOAL is to find **10 unique, high-resolution product images**.\n\nCRITICAL RULES:\n1. **MANDATORY: Search MULTIPLE sources.** Never stop after searching just the manufacturer site.\n2. **Quantity Target: 10 images minimum.** Only return fewer if you have exhausted ALL sources.\n3. **Resolution:** Target 1200px width. Acceptable: 800px-2000px. Reject < 800px.\n4. **Direct Links Only:** Return direct image file URLs (.jpg, .png, .webp).\n\nSEARCH ORDER (you MUST follow this):\n1. First: Official brand website\n2. Then: ALWAYS search retailers: {8a.pl, sportano.pl, decathlon.pl, intersport.pl, bergfreunde.eu, trekkinn.com}\n3. If still < 10: Search more retailers and general web\n\nDO NOT STOP until you have 10 images or have searched at least 5 different sources."
        },
        {
            "role": "developer",
            "content": "## MANDATORY SEARCH PROCEDURE\n\nFor EACH product, you MUST execute ALL of these searches in sequence:\n\n### STEP 1: Brand Website + Preferred Sources\nSearch the brand website(s) from `brand_website` field:\n```\nsite:{brand_website} \"{product_name}\" \"{color}\"\n```\nAlso search any URLs in the product's `preferred_sources` array.\nOpen product pages. Extract ALL image URLs from galleries.\n\n### STEP 2: Major Retailers (ALWAYS search these)\nSearch each shop from the `preferred_shops` list using site: operator:\n```\nsite:{shop_domain} \"{brand} {product_name}\"\n```\nOpen product pages found. Extract ALL gallery images.\n\n### STEP 3: General Search (if still < 8 images)\n```\n\"{brand} {product_name} {color}\" product images high resolution\n```\n\n### STEP 4: Lifestyle Images (if still < 10 images)\nWhen product shots are insufficient, supplement with lifestyle images:\n```\n\"{brand} {product_name}\" worn model photo\n\"{brand} {product_name}\" in use action shot\n```\nLifestyle images show the product being worn\/used by a person - valuable for apparel, footwear, helmets, backpacks.\n\n## IMAGE EXTRACTION TECHNIQUE\n\nWhen you open a product page:\n1. Look for the image gallery\/carousel - extract ALL images (product + lifestyle)\n2. Find the CDN\/media URL pattern (e.g., `cdn.shopify.com`, `media.*.com`)\n3. Extract ALL unique image URLs from that gallery\n4. Modify size parameters: change `?w=400` to `?w=1200` or `_400x` to `_1200x`\n\n## IMAGE PRIORITY ORDER\n1. **Product shots** (white\/neutral background) - HIGHEST priority\n2. **Detail shots** (close-ups of features, materials, labels)\n3. **Lifestyle shots** (product worn\/used by person) - supplement when product shots < 8\n\n## QUANTITY CHECK\n\nAfter each search step, count your images:\n- < 5 images → CONTINUE to next step (include lifestyle search)\n- 5-7 images → CONTINUE, add lifestyle images to reach 10\n- 8-9 images → Add 1-2 lifestyle if available\n- 10+ images → STOP, you have enough\n\n## RESOLUTION RULES\n- Target: 1200px width\n- Minimum: 800px width\n- If URL has size params, modify for larger size\n- Reject thumbnails and previews\n\n## IMAGE TYPES TO FIND (aim for 10 total)\n- **main**: 1 front view product shot (white background preferred)\n- **gallery**: 4-6 different angles (back, side, top, detail shots)\n- **detail**: 1-2 close-ups (zippers, straps, labels, materials, technology)\n- **lifestyle**: 1-2 images of product worn\/in use (if product shots < 8)\n\n## OUTPUT FORMAT\n```json\n{\n  \"products\": [{\n    \"ean\": \"...\",\n    \"images\": [\n      {\"url\": \"https:\/\/direct-image-url.jpg\", \"source_page\": \"https:\/\/product-page\", \"type\": \"main\"},\n      {\"url\": \"...\", \"source_page\": \"...\", \"type\": \"gallery\"},\n      {\"url\": \"...\", \"source_page\": \"...\", \"type\": \"lifestyle\"},\n      ...\n    ]\n  }]\n}\n```\n\n## FORBIDDEN\n- Amazon, eBay, Allegro marketplace\n- Pinterest, forums, blogs\n- Watermarked images\n- Images < 800px\n- Wrong color\/variant"
        },
        {
            "role": "user",
            "content": "## Task\nFind **10 high-resolution images** for each product below.\n\n## Products\n{\n    \"batch\": {\n        \"products\": [\n            {\n                \"ean\": \"3515720099038\",\n                \"name\": \"Millet Buty trekkingowe męskie Hike Up Mid GTX czarne (46)\",\n                \"brand\": \"MILLET\",\n                \"brand_website\": \"https:\\\/\\\/www.millet.com\\\/\",\n                \"color\": \"czarny\",\n                \"preferred_sources\": [\n                    \"8a.pl\",\n                    \"sportano.pl\",\n                    \"decathlon.pl\",\n                    \"intersport.pl\",\n                    \"bergfreunde.eu\",\n                    \"trekkinn.com\"\n                ]\n            }\n        ]\n    }\n}\n\n## MANDATORY SEARCH SEQUENCE\nFor each product, search in this EXACT order:\n\n1. **Brand website** (use `brand_website` field) + any URLs in `preferred_sources`\n2. **Major retailers** from preferred_shops: {8a.pl, sportano.pl, decathlon.pl, intersport.pl, bergfreunde.eu, trekkinn.com}\n3. **General web search** (if still < 8 images)\n4. **Lifestyle images** (if still < 10 images): \"{brand} {product}\" worn\/in use\n\nOpen EACH product page found and extract ALL images from the gallery (product + lifestyle).\n\n## Requirements\n- **Quantity:** 10 images per product (minimum 5)\n- **Resolution:** 800px+ width (target 1200px)\n- **Variant:** Must match exact color from product data\n\n## Image Types (priority order)\n1. **main**: 1 front view product shot (white background)\n2. **gallery**: 4-6 different angles (back, side, detail)\n3. **detail**: 1-2 close-ups (features, labels, materials)\n4. **lifestyle**: 1-2 images worn\/in use (if product shots < 8)\n\n## IMPORTANT\n- Do NOT stop after finding 2-3 images\n- Search ALL sources listed above\n- Extract ALL images from each gallery (including lifestyle shots)\n- Modify URL size params for max resolution (e.g., ?w=1200)\n- Lifestyle images are valuable supplements for apparel, footwear, backpacks\n\n## Output\nReturn JSON with ALL found images (aim for 10 per product)."
        }
    ],
    "model": "gpt-4o",
    "store": false,
    "tools": [
        {
            "type": "web_search"
        }
    ],
    "temperature": 0,
    "tool_choice": "auto"
}