oldred-byte/ec-visual-skill

Let Codex help you organize “product images + reference images” into a complete e-commerce visual production process.

It’s suitable for making main product images, split-screen detail pages, breaking down reference images, generating prompts, and checking and reworking complete sets of images. When you give it products and references, it acts like a patient e-commerce visual director—first inspecting the product, then unfolding references, and assigning what sales tasks each image should handle.

✨ What can it do?

• Plan 5 main e-commerce maps
• Planned 10-Screen Vertical Page Detail Page
• Analyze the layout, shots, text rhythm, and commercial tone of the reference image
• Re-adapting color and visual systems according to the product
• Write independent generation prompts for each image
• After generation, check product consistency, text readability, and style stability
• Prompt correction and regeneration of failed images
• Save the final image and verification records

🧭 How it works

1. Reading Materials
   Identify product images, reference images, product facts, and target uses.

2. Breaking down reference images (language, not scenes)
   Only extract the visual language of the reference image—light perception, font attitude, density rhythm, commercial temperature, module rhythm. Do not copy specific scene elements from reference images (specific fabric, site, texture).

3. Establish Product Rules
   Product appearance, color, material, structure, and usage should be prioritized for accuracy. Colors are subject to the product/brand and do not copy the main colors from reference images.

4. Unified planning of 5 main images + 10 detail pages
   Plan and review the main image and detail pages together. Each image/screen must first match the message-to-picture pattern, then write a Picture Solo Statement, and finally draft copy from the consumer’s perspective.

5. Unified Verification Levels
   Output one unified review form, containing all information from 5 main images + 10 detail pages (image standalone statement, pattern, product exposure, draft copy). Generation only begins after user approval is approved.

6. Write prompts for each page
   Each image is a prompt, including a standalone image statement, pattern, product exposure rules, consumer-view copy lock, and anti-icon statement.

7. Generate, Inspect, Correct
   Inspection items include: standalone screen testing, consumer-perspective copywriting, no icons, product exposure matching, scene templates, and cluster-level diversity.

8. Isolated Storage
   Each task automatically creates a separate folder ~/Desktop/YYYYMMDD-HHMM- /, including main images, detail pages, records, and verification reports, eliminating mixed batches.

🔒 v2 Core Hard Rules

This edition has upgraded 8 hard rules to address issues exposed in actual use:

• Picture Solo Test: Each image must cover all the text, and the image alone conveys the message. If it can’t be covered, the image will be redone; no text is allowed to fill in the text.
• Product-Presence Rule: Packaging should not appear by default. Multiple flavors, natural ingredients, microscopic materials, mechanisms, sensory messages—these should not be packaged.