skills/pixverse-product-stills/references/shoot-recipes.md

# Product Photography Recipes

Select by deliverable intent before environment: a kitchen Pinterest image is a pin,
an in-use header is a banner, and a sequence is a carousel. Apply the image quality
policy; source-specific old image model names do not replace the user's Sunburst default.

## Ten Modes

| Mode | Composition and sequence | Useful directions |
| --- | --- | --- |
| Product shot | Whole readable product, deliberate support/contact/reflection, neutral or styled studio; generic 1:1 | Clean soft studio; hard-rim dramatic; pastel minimal; handmade linen/wood; luxury reflective; bold color; suspended hero; genuine ingredient flatlay |
| Lifestyle scene | Specific usable environment and event; context supports rather than hides the hero | Morning kitchen; vanity; nightstand; desk; cafe; outdoors; living room; gym; social table; pour; curated flatlay; hand-held; gift reveal |
| Closeup with person | Tight product plus hands/partial face/skin; product is the anchor, person secondary | Actual application; vertical hold; pour into palm; eye/lip detail; swatch; hovering dropper; supported jar; material/contact detail |
| Pinterest pin | Vertical 2:3, strong upper hook, warm layered utility, crop-aware focal point | Aspirational lifestyle; vertical feature; recipe cover; editorial flatlay; supported before/after; moodboard; requested quotation |
| Hero banner | Wide, usually 16:9 or requested website crop; one anchor readable after responsive cropping | Product launch; environmental lifestyle; founder/editorial portrait; abstract brand field; wide studio; seasonal; landscape; explicitly split comparison |
| Social carousel | 3–10 connected slides, hook→body→payoff/CTA; common palette/surface/light/type/spacing | Launch; tips; real before/during/after; roundup; brand story; comparison; feature deep-dive |
| Static ad pack | Independent ordered test variants, locked brand, a different truthful hook per image | Problem/solution; supported transformation; real social proof; curiosity; aspiration; feature zoom; offer only if supplied |
| Virtual model tryout | Real product reference worn/used at the correct body site, plausible fit and intersections | Clean studio; editorial; street; outdoor; home; detail; runway; overhead on-body |
| Conceptual product | Deliberately unreal staging with exact product structure and materially coherent light | Levitation; frozen splash; CGI material scene; sculpture; suspended liquid; supported exploded arrangement |
| Restyle | Existing subject and substantially the same composition; change mood, season or both | Aesthetic/material/palette transformation plus optional seasonal props/light; preserve untargeted identity, pose, branding and geometry |

Restyle, closeup-with-person and virtual tryout need a concrete image/product reference.
Text-only product creation is valid when form, category, material, color, packaging and
distinctive features are sufficiently specified. Never invent a real SKU's details.
Marketplace main images, infographics and A+ modules belong to listing-images; moving
try-on/unbox/review belongs to its video recipe. Reuse an accepted wearer reference
when the user requires the same person.

For an unspecified ordinary shoot, use three distinct variants, clean studio, 1:1 and
product-derived/neutral colors. Explicit count always wins. A carousel uses its requested
3–10 slides; an ad pack's count is variant count, not one image containing all variants.
Use platform aspect from the brief (feed 4:5, story 9:16, square 1:1, pin 2:3 as creative
starting points); verify live specifications when promising platform compliance.

## Compose Before Generating

Record exact identity, label/branding, visible surfaces, hand-relative scale and real claims.
For a multi-view text-only product, make the first requested deliverable first and use it
as the remaining variants' product reference; do not add a separate unrequested control
image. With a supplied product, reuse that same source across all first passes. A set's
variations change composition, light, preset or environment materially, not just adjectives.

For carousel, outline slide number, role, subject/action, exact copy and continuity first.
For an ad pack, outline variant, tested hook, medium/composition, audience and aspect first.
Do not invent proof badges, users, endorsements or offers to fill the table. Compare the
set's shared visual system before correcting an individual image.

## Provider Prompt Sections

Write concrete English image directions while preserving literal requested text in its
original language. Use the applicable named sections, fully resolved:

`SUBJECT` — exact product and visible identity; `COMPOSITION` — framing, camera height,
angle, occupancy and support; `ENVIRONMENT / SURFACE` — real scene or conceptual stage;
`PERSON / INTERACTION` when needed — body crop, plausible contact and weight;
`LIGHTING` — source size/direction, contrast, shadow/reflection and highlight roll-off;
`CAMERA / MATERIAL DETAIL` — depth of field, focus plane and surface behavior;
`COLOR / FINISH` — actual palette, photographic register and texture;
conditional `TYPOGRAPHY`; `PRESERVE / EXCLUDE` — targeted fidelity and artifact constraints.

Use the selected mode's logic: tight person crop for closeup, crop-safe anchor for a banner,
same visual-system clauses for every carousel slide, true wearable placement for tryout,
and exact preserve/change separation for restyle. Photography references become tangible
light, lens perspective, composition and surface decisions, not photographer/publication/
competitor names or internal preset codes sent to the model. Real product branding stays.

Typography has three cases:

1. Exact on-image words requested: integrate that literal copy with named font character,
   position and contrast; verify all glyphs after generation, repair deterministically.
2. User explicitly plans later text: leave a naturally calm tonal area within the scene,
   not a hard empty rectangle or artificial blank band.
3. No text request: do not mention overlay space or force an empty region. Compose freely.

Use targeted negatives: no altered label/geometry, invented accessory, extra finger,
impossible grip, floating contact shadow, waxy material, blown reflective edge or unwanted
text. Conceptual mode may intentionally suspend the product; do not negate its concept
with a universal realism ban. Only show ingredients/components supported by input.

## Refinement And Delivery

This photography recipe refines only an observed failed quality gate or a specific user
correction, at most two passes per affected index. It does not inherit mandatory UGC-board
refinement. Use the latest accepted result as the edit source, same aspect and focused
change preserving passing properties. No new crop or model downgrade without the relevant
choice. A vague rejection without a visible defect needs its actual correction intent.

Check whole silhouette, real color/labels/materials, focus, support, scale, human anatomy,
requested text and set coherence with available vision. Deliver unchanged first passes
when no defect exists; say when visual inspection was unavailable. Return the requested
images individually with semantic names and editable exact-copy layers when requested.