photoreal-still-prompt

Installation
SKILL.md

Photoreal Still Prompt

Write a valid JSON image prompt for one believable real-world photograph. This skill produces the prompt, not the image; image rendering uses a separately requested tool or workflow. Return the completed JSON object in one json code block unless the user asks for another format. No markdown or comments inside the object, unresolved placeholders, or fields filled with n/a.

Use references/json-blueprint.md for the JSON structure and the product-packaging example. Adapt its fields to the actual shot; omit people, wardrobe, product, or reference fields when irrelevant. Keep the nested detail relevant to this shot; do not inherit a sample face, pose, beauty treatment, or aspect ratio.

Scope and framing

  • This is a single final image, not a reference sheet, collage, or multi-angle grid. Default to 9:16 portrait whenever the user does not specify an aspect ratio. An explicit user ratio or orientation overrides this default; align crop and camera framing with that choice. Do not inherit a conflicting ratio from a reference image or sample prompt.
  • State what the camera would actually see: height, distance, angle, crop edges, subject scale, foreground/background relationship, and plausible lens or phone perspective. Avoid mutually incompatible optics.
  • In the generated JSON, do not use realistic, photorealistic, or cinematic as a style shortcut. Specify the cues that create the requested look: camera height and handling, focus and exposure, skin/fabric/material texture, light source, falloff, shadows, and scene-appropriate imperfections. If the user says “cinematic,” translate it into those concrete choices rather than repeating the adjective.
  • Anchor objects to real supporting surfaces and positions. Name whether an item sits on a table, counter, shelf, floor, sink ledge, or in a hand; show enough geometry to prove it. A requested kitchen table must not become a floating wall shelf.

Reference control

  • Give each supplied reference one explicit job: person identity, product/packaging identity, location, pose/crop, or lighting/camera. State what to preserve and what not to copy from it. Keep requested changes scoped.
  • If exact product branding matters, use the supplied product image or exact text as authority. Preserve package silhouette, cap, palette, wordmark, label hierarchy, and readable supplied copy. Do not replace it with a blank package, a category-only label such as Hair Serum, invented claims, or random lettering.
  • If no brand or package reference is supplied and a product must be prominent, design a plausible fictional retail identity: an intentional container, material, cap, proper-name wordmark, product-line name, short category descriptor, size, and restrained label hierarchy. Specify the short text exactly; leave fine print indistinct rather than requesting gibberish. If the user needs an exact existing brand and has not provided it, ask for the product reference or exact label copy.
  • Background products should read as ordinary retail objects, not a row of identical blank white bottles. They carry plausible fictional brands too. Describe only the few whose design matters; their short text is readable where in focus, soft where not.
Installs
3
First Seen
Sep 23, 2026