Product Visuals

SkillMedia

Generate product imagery - packshots, lifestyle scenes, e-commerce sets, mockups, and short product videos from reference photos. Use for product pages, catalogs, marketplaces, and launches.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Product Visuals skill

What this skill tells your AI

The instructions your AI receives, as published by navinspire-ia/navin in navin/skills/product-visuals/SKILL.md and read by ahel’s review.

Produce a consistent visual set for one product: packshots, lifestyle scenes, detail shots, and short videos - from a text description or from the user's real product photos.

AWS media templates

If a media template is attached, treat the downloaded file under .navin/resources/media-templates/ as the master reference (format, lighting, crop). Generate variations from that file. If the download is missing, stop and tell the user.

Tooling

  • generate_image - all still visuals. Pass the user's product photo (or a validated generated packshot) in reference_images to keep the product identical across the whole set.
  • generate_video - 360 turns, hero animations, unboxing-style clips; use reference_image with the validated packshot.
  • visual_qa - mandatory final gate for every still. Pass the authoritative product photo in references and claim product_fidelity; use marketplace, ratio, dimensions, alpha and safe-zone requirements for the placement.
  • exec + Pillow - exact marketplace sizes, white-background compliance checks, batch renaming.
  • Interactive 3D (user asked to orbit the product on a page): three or @react-three/fiber + @react-three/drei on that web page, with the validated packshot as the still fallback. Never put that canvas on a PowerPoint slide.
  • Missing tools → deliver the shot list + exact prompts and point to Settings → Image / Video.

The standard e-commerce set

For one product, produce in this order:

  1. Master packshot - white/neutral background, soft studio lighting, slight shadow. Validate with the user before declining anything.
  2. Angles - 3/4 left, 3/4 right, back, top (each via reference_images from the master).
  3. Detail shots - texture, label, ports/seams - whatever sells the product.
  4. Lifestyle scenes - product in real context, matching the brand's world (2-3 scenes).
  5. Scale shot - product next to a familiar object or in-hand.
  6. Short video - 6-8s rotation or hero animation from the master packshot.

Prompt patterns

Packshot:

generate_image(
  prompt="Professional e-commerce packshot of [product], pure white background, soft even studio lighting, subtle contact shadow, centered, no props, ultra sharp product photography",
  reference_images=["/path/to/user-product-photo.jpg"],
  aspect_ratio="1:1", image_size="2K"
)

Lifestyle:

generate_image(
  prompt="Use the reference image. Place this exact product on a rustic oak kitchen table, morning sunlight through a window, shallow depth of field, warm lifestyle photography, keep the product and its label strictly unchanged",
  reference_images=["<master packshot path>"],
  aspect_ratio="4:3"
)

360/hero video:

generate_video(
  prompt="Slow elegant 360-degree rotation of this exact product on a seamless white background, studio lighting, keep product and label unchanged",
  reference_image="<master packshot path>",
  aspect_ratio="1:1", duration_seconds=8
)

Always include "keep the product and its label strictly unchanged" when working from references.

Marketplace compliance (check before delivering)

PlatformMain imageNotes
Amazonpure white (RGB 255,255,255), product ≥85% of frame, ≥1600pxno text, logo, watermark
Shopify/sitefree, 1:1 or 4:5 recommended, ≥2048pxconsistent set
Instagram Shop1:1, lifestyle allowedproduct visible, low text

Use Pillow via exec to verify pixel dimensions and background whiteness, and to export exact sizes. Then call visual_qa for each deliverable. Deliver only PASS assets. Rework WARN assets and never deliver BLOCK assets automatically. A product-fidelity claim without an authoritative reference is BLOCK, never PASS.

Rules

  • One master, validated early - everything declines from it via reference_images.
  • Never alter the product itself: shape, colors, label, and branding stay exact.
  • Deliver via the message tool with artifact paths in media; keep raw paths internal.
  • Name files predictably: sku_packshot_front.png, sku_lifestyle_kitchen.png, etc.
  • For ad-oriented declinations (headline space, platform ratios, CTA), continue with ad-creative-generator.

Signals

GitHub stars
22
Forks
4
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
product-visuals
Source
github.com/navinspire-ia/navin