ecom-details-image · E-commerce Product Visuals

Plans e-commerce product visuals: hero-shot concepts, scene images, and detail-page visual direction, delivered as executable AI image-generation prompts.

Two modes: Brief/Prompt mode (visual brief + prompts only) and Generate mode (invokes the image script only when rendering is explicitly requested). A full product set defaults to a 5-hero + 7–9 detail-image pack plan.

Example invocation: "Create a detail-page visual plan for this thermos."

Full brief

Positioning

ecom-details-image is the planning entry point for e-commerce visuals: visual strategy plus executable image-generation prompts. It plans; it does not lay out or render. Actual rendering can be handed to the image-generation entry skill.

Core capabilities

25 scene-template matching: matches prompt_template / variants / category_tips by scene type as the prompt skeleton (only the matched template is ever loaded).
Conversion-driver diagnosis: product/detail-page/marketing jobs get diagnosed before sequencing; no invented claims, sales figures, or reviews when evidence is absent.
Campaign Style Lock: multi-image jobs lock palette, color temperature, typography, backgrounds, lighting, layout, and icon style for the whole set — the same lock block opens every prompt verbatim.
PDP image pack: detail-page / PDP / full-set requests default to 5 hero images + 7–9 detail images; detail images must be e-commerce infographic format (titles, icons, labels, benefit points, steps, or trust badges) — not plain multi-angle product photos.
Prompt iron rules: hex colors, numeric ratios, explicit whitespace, negative-constraint lists, anti-AI tips for UGC/social scenes; on-image copy kept short and necessary; users are reminded to zoom to 200% and verify Chinese glyphs after rendering.
Generate mode: calls generate_image.py (apimart endpoint, 14 sizes, 1k/2k/4k resolution); passes --image with the user's reference product photo when one is provided, which keeps the product's appearance accurate.

Workflow

Identify the visual task type and scene, match a template
Collect only the missing inputs that materially affect the image (non-critical gaps are assumed explicitly, never blocking)
Build the visual brief; build the Campaign Style Lock for multi-image jobs
Run conversion-driver diagnosis and sequence for product/detail-page jobs
Write executable prompts against the iron rules (final prompt first, image script second)
Invoke the generation script only when rendering was explicitly requested; otherwise deliver the prompt pack

Inputs & outputs

Minimum inputs (confirm first): goal, usage (hero / ad / social / banner / PDP module / thumbnail), subject, audience and context, style, composition and ratio, whether on-image copy is needed, negative constraints.

Brief/Prompt mode output: matched template, Visual Brief, Final Image Prompt, Negative Constraints, Assumptions; product/marketing jobs add conversion diagnosis, Campaign Style Lock (multi-image), hero sequence, detail-image sequence.

Generate mode adds: Image Pack Plan (number, purpose, size, template, and short copy per image) and generated file paths.

Boundaries with adjacent skills

ai-image-gen: the general image-generation execution entry; this skill is the e-commerce visual planner — strategy + prompts, with actual generation handable to it.
remove-bg: actual background removal for cutout product shots.
card-design: aesthetic alignment — the de-AI-cheapness principles (no blue-purple gradients, no emoji-as-icons, etc.) are written into the prompts' positive/negative constraints; alignment at the aesthetic level, not HTML rendering.

Fit

Hero / scene / detail-page visual planning for products
Prompt production for ad creatives, social marketing images, banners
Visual-consistency locking across multi-image campaigns

Before you start

Brief/Prompt mode needs no configuration at all.
Generate mode needs IMG_API_KEY (apimart / OpenAI-compatible endpoint), metered; without keys the script is never invoked — only the complete prompt pack plus setup guidance is delivered.
Keys are configured by the user; the skill never requests, echoes, uploads, or commits real keys.

ecom-details-image is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.

← All 155 skills