Xinghe Amazon Visual Suite

Xinghe Amazon Visual Suite

Create a complete Amazon visual image suite in one run: product analysis, selling-point strategy, layout planning, prompts, generation, deterministic size export, QA, and delivery.

Default output count is fixed unless the user explicitly asks otherwise:

• 1 main image, square 1:1.
• 6 secondary images, vertical 3:4.
• 7 desktop A+ modules, final 1464x600.
• 7 mobile A+ modules, final 1600x1200.

Hard count rule: the default deliverable is exactly 21 final images. Do not reinterpret this skill as a 6-image Amazon A+ generator. Do not downgrade, merge, skip, or collapse the output into 6 A+ modules, a collage, or a partial A+ set unless the user explicitly requests a smaller output count in the current task.

Required Workflow

1. Require at least one real product white-background image before producing the main image. Do not invent a main product image.
2. Read references/workflow.md before planning.
3. Read references/image-specs.md before writing prompts or resizing outputs.
4. Read references/page-task-table.md before creating the 21-image task table.
5. Read references/quality-rules.md before generation and again before delivery QA.
6. Read references/generation-backend.md before choosing the generation route.
7. Build a complete Product Master Description and reuse it in every generated visual prompt.
8. Generate every final visual as a separate image. Do not use a collage as a substitute for an individual final image.
9. Prefer the Xinghe deployment helper or fixed API route when available. If unavailable or failed, use built-in image_gen as the final fallback.
10. Use scripts/visual_size_helper.py only for deterministic resize, crop, padding, format conversion, and dimension checks.
11. Use scripts/make_contact_sheet.py only to create a review overview from finished images.
12. Treat fewer than 21 final images as incomplete execution unless the user explicitly requested fewer images. Report incomplete execution and continue generating the missing groups rather than stopping at 6 A+ modules.

Inputs

Collect these if available, but do not block except for the real white-background product image needed for the main image:

• Product name, category, platform, marketplace site, and target buyer.
• White-background product images and optional side/detail/packaging/lifestyle images.
• Listing title, bullet points, description, A+ copy, product facts, parameters, certificates, and forbidden claims.
• Brand assets, colors, reference visuals, competitor pages, and style preferences.

If platform or marketplace is missing, default to Amazon US and English image copy.

Planning Contract

Before generation, output a concise plan containing:

• Product Image Analysis.
• Information Confidence: confirmed, reasonable inference, or needs confirmation.
• Buyer Demand Map.
• Demand-to-Selling-Point Match.
• Product Master Description.
• Campaign Style Lock and Product Consistency Lock.
• Complete 21-image task table with image id, asset type, final size, buyer question, selling point, visual proof, headline, subcopy, layout, required source material, and risk note.

Use conservative copy for unproven points. Do not put needs confirmation claims on images.

Generation Contract

Generate in this order:

1. Main image.
2. Six secondary images.
3. Seven desktop A+ modules.
4. Seven mobile A+ modules.

Completion requires all four groups. If any group is missing, continue with the missing group. A run that only creates A+ images, only creates 6 modules, or only creates a contact sheet is not complete.

Desktop and mobile A+ modules may share the same selling logic, but must be independently composed for their final aspect ratios. Do not stretch, squeeze, or crop desktop modules into mobile modules as the final design.

Delivery Contract

Deliver:

• Generation method used: Xinghe helper, fixed API script, or built-in image_gen fallback.
• Save directory.
• Numbered file list grouped by main, secondary, desktop A+, and mobile A+.
• Contact sheet path when created.
• QA notes covering product consistency, physical logic, text legibility, language, unsupported claims, dimensions, and repeated template risk.
• Missing evidence list for claims, parameters, certifications, compatibility, performance, or compliance language.

Include an explicit final count line in delivery: Final kept image count: 21/21. If the count is lower, state Incomplete and list the missing IDs.