Brand Shoot Kit
Run this when the user wants a complete ecommerce product photography system, not one-off prompts.
For module boundaries and stage contracts, use SUITE.md plus references/module-contracts/.
Outcome Contract
Produce a runnable shoot packet that includes:
00-brand-analysis.md01-visual-gap-audit.md02-shoot-strategy.md03-shot-list.md04-generation-prompts.md05-qa-report.md06-export-map.md
Also include asset folder targets:
assets/pdpassets/lifestyleassets/modelassets/seasonalassets/socialassets/emailassets/marketplace
Non-Negotiables
- Product truth beats aesthetics.
- Diagnose before generating.
- Keep the default pack to 12 shots unless asked otherwise.
- Keep ecommerce-first framing; this is not an ad-intelligence workflow.
- If browsing/API/image generation is unavailable, still deliver the full planning packet and GPT Image-ready prompts.
Flow (Always This Order)
- Intake: confirm product URL/image, target channels, constraints.
- Brand analysis: brand tone, audience, price tier, visual language.
- Product preservation brief: what must never change.
- Visual gap audit: what is missing and priority.
- Shoot strategy: 2-3 directions, pick one with rationale.
- Shot list: practical 12-shot pack with channels + ratios.
- Prompt pack: one prompt per shot with negative constraints + reroll instruction.
- QA rubric/report: pass/fail gates and rejection triggers.
- Export map: where each approved asset goes.
- Memory: note what to preserve/avoid next run.
When deterministic orchestration is requested (URL/scout JSON -> packet), use:
scripts/run-brand-shoot.pyfor URL/scout-to-packetscripts/validate-packet.pyfor structure validation
Product Preservation Brief (Required)
Always output this block before prompts:
## Product Preservation Brief
- Product type:
- Must preserve:
- Can vary:
- Never change:
- Distortion risks:
- Accuracy confidence:
If confidence is low, explicitly say so and bias toward clean commerce shots over aggressive lifestyle edits.
Anti-Patterns (Block These)
- Starting with moodboard words before diagnosing brand/product.
- 30-50 shot bloated plans by default.
- Fake claims/certifications on labels or props.
- Uncanny face-heavy model scenes when hands/body crop solves the job.
- Generic "luxury minimal" output for every brand.
- Passing pretty but inaccurate images.
- Drifting into ad spy language or competitor ad analysis.
Prompt Quality Rules
Each shot prompt must include:
- use case + ratio
- scene and composition
- product placement and handling
- preservation rules
- negative constraints
- reroll instruction tied to likely failure
Use plain, operational language. Avoid cinematic fluff.
Reroll Logic
If any rejection trigger appears, rewrite prompt and reroll that shot:
- changed product shape/label text/logo
- wrong product count or variant
- malformed hands touching product
- fake badges/claims/certifications
- product too small/obscured for commerce use
Graceful Degradation
When tools/keys are missing:
- still produce full packet
- include "manual generation runbook" section in
04-generation-prompts.md - mark image generation as
Not Runin05-qa-report.md
When URL extraction fails:
- ask for pasted PDP copy + 3-5 product image URLs or uploads
- continue workflow with confidence notes
Calibration Example (Taste)
Weak prompt: "Create a beautiful premium serum image in a modern bathroom with soft light."
Strong prompt: "Square 1:1 PDP-support lifestyle image for a 30ml frosted glass serum bottle with matte white dropper cap and front label text preserved exactly from reference. Place bottle upright front-facing on off-white stone vanity, morning side-light from left, shallow depth of field, folded neutral towel and small green leaf in far background blur only. Product occupies ~38% of frame, label fully legible, no extra text or badges, no second bottle, no warped dropper geometry. If label legibility fails, reroll with camera 15% closer and flatter angle."
Output Style
Write like a creative director + ecommerce merchandiser:
- specific decisions
- explicit tradeoffs
- clear priorities
- no filler sections
For deeper rubrics and templates, use files in references/ and examples/.