Product Photoshoot Prompts
Generated product images go wrong in three ways: the product changes (new label, bigger portion, extra buttons), the scale is off (a mug the size of a bucket), and there is no room left for the headline. Image models follow a spec better than a vibe. This skill turns a short JSON spec into a prompt that pins those three things, for any image model.
Procedure
- Pick a mode.
python3 scripts/shot_prompt.py --list-modesprints all ten with their use. Mode details and when each one fits:references/modes.md. - Write the spec as JSON, one object per shot. Copy
examples/acme-coffee/shot-specs.json. Required:mode,aspect,product.name,product.description,product.size,product.scale_reference. Strongly advised:reference_photo: truewhen you will attach a real product photo,do_not_invent,keep_clearfor text space. - Generate the prompts:
Sample output:python3 scripts/shot_prompt.py shot-specs.json python3 scripts/shot_prompt.py shot-specs.json --format json python3 scripts/shot_prompt.py shot-specs.json --format midjourneyexamples/acme-coffee/outputs/shot-prompts.md. - Attach the real product photo in the image tool whenever it accepts a reference. With a reference, the prompt tells the model to keep the product exactly as photographed; without one, it marks the result as an illustration that cannot be shown as the real product.
- Review every result against the product: label text, colour, shape, size relative to the scale reference, count. Reject anything that changed. Close-ups (
macro-detail) are real photos when they will sell texture or portion. - Composite when in doubt: generate the scene empty, then place the real cutout in
ad-variants.
Rules
- The scene may be invented. The product may not.
- No other brands, logos or readable text besides the product label.
- People shown are clearly generic. Never a look-alike of a real person.
- Label any image that is an illustration, not a photo of the real product, before it goes into an ad.
- Prompts carry no model-specific magic words by default; the Midjourney format only adds the aspect flag.
Output
One prompt per shot, the review checklist, and a note per shot: real-product composite, reference-based or illustration.
Credit
The mode taxonomy is inspired by product-photoshoot workflows in commercial image tools. No text was copied.