product-photography

Product Photography Studio

Skill for generating professional AI product photography and video of any product. The user provides real photos of their product (even low-quality ones) or a store URL, and the skill produces professional images and simple videos that are always faithful to the real product.

Always reply to the user in their own language — the language of these instructions does not define the conversation language.

The non-negotiable principle: total product fidelity

This skill exists to sell REAL products. A beautiful image of a product that doesn't exist is worse than useless: it misleads the customer and causes returns. Therefore:

• Never invent anything about the product. If the original photo doesn't show the sole of the shoe, don't describe it in the prompt or let the model imagine it as a prominent detail.
• Don't elaborate on what is unclear. A blurry logo, a doubtful seam, illegible text: they do NOT go into the prompt and, where possible, stay out of frame or away from focal areas.
• Emphasize what IS visible. Exact colors, shape, materials, visible finishes: that is what the prompt must pin down precisely.
• Verify before delivering. Every generated image is compared with the original product attribute by attribute. If it isn't identical, it gets corrected. Never silently deliver something that doesn't match.

Per-user configuration

Before generating anything, read ~/.claude/product-studio/config.json. If it doesn't exist, tell the user to run /product-setup first (or walk them through configuration right then).

{
  "provider": "higgsfield | fal | custom",
  "fal_api_key": "only if provider = fal",
  "custom": { "description": "...", "notes": "..." }
}

The details of how to call each provider are in references/providers.md — read it before the first generation of the session.

Local learnings

The file ~/.claude/product-studio/learnings.md accumulates the corrections that have worked in the past (what tends to fail and how to fix it).