The image can change.
The product cannot.
A shared production and quality-control process for AI-assisted product imagery — the full e-commerce catalogue, collection heroes, campaign imagery, advertising assets, social content and product-led motion.
The objective is not simply to produce visually convincing images. Every final asset must represent the original jewellery accurately, follow the approved visual direction and remain consistent with the wider brand system. AI output is production material requiring comparison, correction and approval — never a final decision on its own.
Every approved asset must preserve
Required source material
Before generating an image, gather all available product information. Production should not begin when a critical part of the product cannot be understood from the available references.
Prompting system
The references define the product. The prompt defines how it is presented. A production prompt should be structured, measurable and easy to revise.
Be specific
"Seven round-cut red stones" is more useful than "several red gemstones."
Avoid contradictions
Do not request an extremely shallow depth of field while requiring the entire piece to remain sharp.
Correct one variable at a time
Geometry, materials, lighting, background and composition should be refined separately.
Repeat critical constraints
Repetition is useful when it protects the product, but useless if not.
Don't solve everything in one generation
Lock the jewellery first. Build the scene around it afterwards.
AI is a great prompter
Use any LLM to generate the prompt. Make sure you say what model you are generating on.
Prompt builder
Fill in the eight fields below — every field maps to one non-negotiable part of the prompt. Edit freely; the assembled prompt updates live on the right.
Product photography workflow
A structured workflow separates product reconstruction from visual styling. Each stage moves from reference analysis and product accuracy to lighting, composition, refinement and final approval.
Confirm what's being produced before anything is generated.
Create the most accurate neutral representation of the jewellery before introducing models, expressive environments or campaign lighting. Once approved, it becomes the base for the full image set.
Neutral views should be approved before producing the more complex scale and model assets. Recommended production order — check off as each is produced:
Do not discard a correct product because one secondary element failed. Correct isolated failures without unnecessarily regenerating the entire image. Use local correction for:
Never approve from memory. Compare the final image directly with:
Error severity
Not every issue has the same impact. This system distinguishes between errors that misrepresent the product, issues that must be corrected before approval and minor imperfections that do not affect accuracy.
- Wrong product or variant
- Altered silhouette or proportions
- Missing or added components
- Incorrect gemstone count
- Wrong metal
- Incorrect chain or clasp
- Incorrect engraving
- False scale
- Major geometry failure
- Incorrect background
- Inaccurate roughness or finish
- Wrong gemstone colour or setting
- Broken or fused links
- Floating product
- Incorrect contact
- Contradictory lighting
- Unnatural reflections
- Visible AI artefact
- Incorrect model interaction
- Slight background variation
- Small edge halo
- Minor reflection inconsistency
- Local sharpness mismatch
- Isolated artefact outside the jewellery
Full QA checklist
Every final asset must be inspected directly against the original references before delivery. Run this checklist per asset — it covers product accuracy, materials, lighting, background, scale, artefacts and technical requirements. 71 checks across 10 categories.