| name | lifestyle-stack-generator |
| description | Generate demographically-accurate lifestyle photography for Amazon listing slots 2-7 using Higgsfield. Pulls customer demographic from review-mined photo evidence, optionally trains a Soul Character for cross-slot consistency, then generates 6 lifestyle/use-case images. Use when running `lifestyle-stack-generator {ASIN}` or when the listing image stack needs people-in-context photography. |
lifestyle-stack-generator — Slot 2-7 Lifestyle Photography
Per 02-visual-content/listing-images.md, slots 2-3 are highest-impact conversion slots, and lifestyle photography (people using the product in context) is one of the top image types. This skill generates that stack consistently across all 6 slots using Soul Character training.
Methodology — read before rendering
Read reference/02-visual-content/listing-image-creative-director.md in full before generating any concept. That file is the source of truth for:
- Editorial design philosophy (Monocle / Kinfolk aesthetic, NOT infographic clipart)
- The 5-spinoff methodology (same hero benefit, 5 different visual deliveries via lever rotation)
- The 6-section mandatory prompt structure (Scene / Position / Lighting / Palette / Type Placement / Aspect Ratio)
- The anti-clipart rules — no callout arrows, dimension lines, overlay boxes, flow charts, annotated diagrams, burst stickers
- 4-axis scoring rubric (40/30/15/15)
- Iteration loop: cap at 3 attempts per slot
Hard rules:
- Aspect ratio: 1:1 for listing slots 2–7. Locked in at start AND end of every prompt.
- Reference image always: pull live product photo from
brain/products/{asin}-{geo}.json and pass to Higgsfield
- Brand guidelines respected:
brain.business.brand_guidelines drives palette + visual style keywords; falls back to editorial neutral if empty
- No text generation in image — generate clean photography with type-zone whitespace; brand designer adds type in post
Output emits florence-concepts-{asin} per the artifact protocol.
Prerequisites
- Higgsfield MCP connected (https://mcp.higgsfield.ai/mcp)
- Existing research brief or
asin-deep-research already run
- Product reference image (clean PNG)
Invocation
lifestyle-stack-generator {ASIN}
lifestyle-stack-generator {ASIN} --slots=2,3,4,5,6,7 # default all 6 slots
lifestyle-stack-generator {ASIN} --train-soul # train a Soul Character first for consistency
lifestyle-stack-generator {ASIN} --reference-photos={dir} # use customer photo references for demographic
Output
/tmp/cro-content/{ASIN}-lifestyle-stack-{date}.md + 6 images saved to /tmp/cro-content/{ASIN}-lifestyle/slot-{N}.png.
Phase 1 — Demographic Source
Pull from research brief:
- Photo review evidence — age band, gender split, race distribution, setting (home / office / outdoor / etc.)
- Use context — when/where/how customers report using it
- Lifestyle keywords — "for small apartments" / "as a gift" / "during travel"
If brief missing demographic data: pause, request review-mining --media-only first.
Phase 2 — Soul Character Training (optional but recommended)
If --train-soul flag (or 6+ slots requested): train a Soul Character with 3-5 reference photos matching the demographic, save trained character ID to brand state file at ~/.claude/skills/lifestyle-stack-generator/.brand-characters.json.
Reuse trained characters across ASINs in the same brand → big consistency win.
Phase 3 — Slot-by-Slot Concept Brief
For each slot, derive the concept from research:
| Slot | Concept type | Source signal |
|---|
| 2 | Hero benefit photography | Top driver (reviews + keywords) |
| 3 | Secondary benefit / use case | Second driver |
| 4 | Lifestyle in-context | Photo review evidence + lifestyle keyword |
| 5 | Detail shot or scale ref | Top objection (size / quality) |
| 6 | Edge case / additional use | Positive surprise from reviews |
| 7 | Strong close — gifting / family / pride | Repurchase signal |
Phase 4 — Generate
Default model per plan: Nano Banana Pro for product fidelity, GPT Image for creative angles. No multi-model spread — best-only.
Per slot:
- Pass: prompt + Soul Character (if trained) + product reference + slot-specific composition
- Aspect: 1:1 (Amazon main image) or 4:5 (mobile)
- Resolution: 2000×2000
Phase 5 — Output
# Lifestyle Image Stack — {Title}
**ASIN:** {ASIN} | **Date:** {date} | **Soul Character:** {trained ID or N/A}
## Demographic Anchor (from review photos + research)
- Age: {band}
- Gender: {split}
- Setting: {home/office/outdoor}
- Use context: {when/where}
## Slot Plan
| Slot | Concept | Image | Driver/Objection | Approve? |
|------|---------|-------|-------------------|----------|
| 2 | {} |  | {} | ⬜ |
| ... | ... | ... | ... | ... |
## Designer Brief (handoff)
For polish:
- Maintain consistent character across all slots (Soul ID: {})
- Use trained character for any future variations
- Output spec: PNG, sRGB, 2000×2000 (1:1) or 1600×2000 (4:5 mobile)
## Recommended Next Skill
- Validate stack: `stacked-gallery-test {ASIN}` (Pinion gallery comparison)
- Or full content plan: `/cro-content-plan {ASIN}` (combines all slots + A+ + copy)
Reference Files
- Vault:
CRO-Knowledge-Base/02-visual-content/listing-images.md
~/.claude/skills/cro/listing-image-best-practices.md~/.claude/skills/hero-image/prompt-engineering.md
Quality Bar
Auto-Triggers
- User asks "generate lifestyle images for {ASIN}"
/cro-content-plan calls it for slot 2-7 workcvr-leak-fix calls it when the leak is image-shaped
v0.1.12 — Read brain.image_strategy before generating
Before drafting any Higgsfield prompt, read brain.image_strategy from working memory:
- If null → pause and offer
image-strategy first (~10 min, makes every future render bespoke). If user proceeds without, flag concept cards with "No category research used — generic aesthetic."
- If set + fresh (<90d) → inject
prompt_adjustments.scene_keywords into prompt section 1, palette_keywords + mood_keywords into section 4, and do NOT include {anti_patterns_csv} near the end. Cite the strategy on each concept card.
- If stale (>90d) → flag and recommend
image-strategy --refresh.
This brand-level strategy comes from skills/image-strategy.md's top-15-bestsellers analysis. Same adjustments apply across every render for this brand.
v0.1.13 — Visual verification gate + base64 embedding (NON-NEGOTIABLE)
Florence does NOT present an image she hasn't actually looked at, AND she does NOT use raw Higgsfield URLs in artifact HTML. Two blocking rules added in v0.1.13:
Verification gate
After Higgsfield returns each generated image, BEFORE adding to any artifact OR sending to Pinion:
- Load the image into Florence's multimodal context — paste the URL in chat so Cowork's multimodal Claude loads it natively. Narrate: "Looking at #{N} before I include it."
- Visually inspect against: aspect ratio (1:1 main+listing / 16:9 A+), product fidelity vs reference, technique landed (visibly), background appropriate, no clipart leak (anti-clipart rules), no text-on-main-image (TOS), no model faces.
- If anything fails → re-prompt with specific fix + regenerate. Cap at 3 attempts per concept; surface honestly on attempt 4.
- Only after ALL concepts pass → proceed to artifact emission.
The eye trumps the score. Even if the rubric said 100/100, if visual inspection finds a wrong product or off-aspect output, it fails the gate.
Base64 embedding
Higgsfield URLs are temporary AND Cowork's artifact iframe sandbox blocks external image loads in many builds. For every verified concept:
- HTTP GET the Higgsfield URL → fetch image bytes
- Detect MIME type from response headers (typically
image/png)
- Base64-encode the bytes
- Substitute
{{image-src}} (or {{winner-image-src}} for tests) with data:image/png;base64,<encoded> — NOT the raw Higgsfield URL
This makes the artifact self-contained — survives sandbox + URL expiry. ~1-3 MB per image is fine for Cowork.
The live product image ({{product-image-url}} in concepts.html hero strip / dossier head / cockpit product cards) stays as the live Amazon CDN URL — that's permanent and not affected.
