Jewelry Photography Glossary

Reference definitions written from a jewelry-specialist perspective. These terms appear frequently in discussions of AI jewelry photography, product fidelity, and jewelry e-commerce.

Macro Photography: Macro photography refers to close-up photography of small subjects at a reproduction ratio of 1:1 or greater. In most product photography categories, "close-up" means capturing an object at arm's length. For jewelry, it means something fundamentally different: the subject is often measured in millimetres, and the camera must resolve features — stone facets, prong tips, millgrain edges — that are invisible to the naked eye at normal distance. Standard product photography lenses are calibrated for objects the size of a shoe or a handbag. Jewelry demands either dedicated macro lenses or extension tubes, and lighting setups that avoid reflections across curved metal surfaces. FormaNova's AI models are trained on macro-scale jewelry imagery, which is why they preserve fine surface detail that general-purpose models — trained on broader product datasets — routinely lose.
Catchlight: A catchlight is the specular highlight — the bright reflection of a light source — visible in a reflective surface. In portrait photography, catchlights appear in the subject's eyes and signal professionalism. In jewelry photography, catchlights appear in cut gemstones and polished metal, and they are structurally non-negotiable: a diamond without catchlight looks dead. The challenge for AI models is that catchlights in faceted gemstones are not simple reflections — they are the result of light entering the stone, refracting at each facet, and exiting at precise angles determined by the cut geometry. Standard image-to-image models often flatten or misplace catchlights during background replacement, producing stones that look plastic. FormaNova preserves catchlight positioning as a preserved feature, not a regenerated one.
Latent Diffusion: Latent diffusion is the underlying mechanism of most modern AI image generation models, including Stable Diffusion. Instead of operating on raw pixel data, these models compress an image into a lower-dimensional latent space using a variational autoencoder (VAE), perform the diffusion process in that compressed space, then decode back to pixels. The compression is the problem for jewelry. A VAE trained on general image datasets learns to compress images efficiently by averaging over common patterns — smooth gradients, organic shapes, broad colour fields. Jewelry facets and prong tips are high-frequency features: sharp edges, specular points, precise geometry. The VAE treats these as noise and smooths them out during encoding and decoding. The result is prongs that look rounded, facet edges that blur into each other, and pavé that becomes a uniform shimmer. FormaNova addresses this through training data and loss functions specifically designed to penalise facet degradation.
Product Fidelity: Product fidelity, in the context of AI photography, refers to the degree to which an AI-generated image accurately represents the original product — unchanged in shape, material, colour, and detail. For most product categories, a small degradation in fidelity is acceptable. If an AI tool slightly alters the shade of a handbag, the brand can live with it. For jewelry, fidelity is a legal and commercial requirement. A ring rendered with slightly different prong geometry is a different ring. A bracelet with altered stone colour may violate a supplier agreement. SSIM (Structural Similarity Index) and LPIPS (Learned Perceptual Image Patch Similarity) are the standard metrics for measuring fidelity. FormaNova provides per-generation fidelity scores so brands can verify output before using it commercially.
SSIM (Structural Similarity Index): SSIM is a perceptual metric that measures the structural similarity between two images on a scale from 0 to 1, where 1.0 means the images are identical. Unlike pixel-level metrics such as MSE (mean squared error), SSIM accounts for luminance, contrast, and structure — making it a better proxy for human perception. In the context of AI jewelry photography, SSIM is used to measure how closely the jewelry piece in the output matches the jewelry piece in the input image. A score below 0.95 on the jewelry region typically indicates visible changes to stone placement, metal geometry, or setting structure. FormaNova surfaces SSIM scores as part of its accuracy verification output, giving jewelry brands a quantitative basis for assessing whether an AI-generated image is commercially usable.
Bezel Setting vs. Prong Setting: A bezel setting encases a gemstone in a continuous ring of metal that holds it flush to the band. A prong setting uses thin metal claws — typically four or six — to grip the stone, leaving its sides and base exposed. The distinction matters for AI photography because these settings behave completely differently under the model's attention mechanism. A bezel setting presents as a clean metal edge — relatively forgiving for AI tools because the geometry is simple. A prong setting presents as four or six thin vertical lines surrounding a stone, which is extremely high-frequency information. General-purpose models frequently merge prongs, eliminate them, or distort their positioning. FormaNova's training includes explicit prong geometry preservation, making it significantly more reliable for prong-set pieces than general-purpose alternatives.
Pavé: Pavé (from the French word for "paved") is a setting technique in which many small diamonds or gemstones are set closely together across a surface, held by tiny beads or prongs, creating the appearance of a surface covered entirely in stones. Pavé is the hardest category for AI fidelity preservation for a specific reason: the individual stones are often just 1–1.5mm in diameter, and the setting requires dozens to hundreds of tiny prongs or beads per piece. At the resolution of a standard product image, these features sit at the boundary of what the model can resolve. General-purpose models trained on diverse image data will routinely smooth pavé into an undifferentiated shimmer. FormaNova was specifically tested against pavé-heavy pieces during development because this was identified early as the hardest test case for structural fidelity.
Ghost Mannequin: Ghost mannequin (also called "invisible mannequin") is a post-production technique in fashion photography where a garment is photographed on a mannequin, then the mannequin is digitally removed, leaving the garment appearing to float in its natural worn shape. It became an e-commerce standard because it conveys fit and structure without distracting from the product. Jewelry has no direct equivalent. A necklace worn by a model and a necklace lying flat are different objects — drape, tension, and the way chains fall are all determined by the wearer's neck and posture. There is no agreed industry standard for "neutral" jewelry presentation equivalent to ghost mannequin in fashion. This is one reason jewelry photography remains more expensive and technically demanding than apparel photography — and one of the core problems FormaNova was designed to solve.
Background Replacement: Background replacement in product photography refers to removing the original background from a product image and substituting a different one — a studio backdrop, a lifestyle scene, or a plain white/grey. In general product photography, this is a straightforward masking task: the product has clear edges, uniform lighting, and limited interaction with the background. Jewelry is different. Metals and gemstones reflect the background they're shot against — a silver ring photographed on a black surface will carry dark reflections in its curves that are visible in the final image. Background replacement that ignores these environmental reflections produces images that look composited. Outpainting — extending the image beyond its borders — is a related but distinct technique. FormaNova's background replacement is specialised for jewelry: it accounts for environmental reflections on metal and adjusts lighting accordingly during generation.
CAD (Computer-Aided Design) in Jewelry: CAD in jewelry refers to the use of 3D modelling software to design pieces before they are physically produced. Programs such as Rhino, MatrixGold, and JewelCAD are standard in the industry, producing STL or OBJ files that can be used to drive CNC mills, 3D printers, or lost-wax casting equipment. CAD models contain exact geometric specifications of a piece — stone dimensions, prong angles, band thickness, millgrain profiles — that are not visible in photography of a finished piece. FormaNova integrates with the CAD workflow at two points: first, FormaNova's Text-to-CAD feature generates 3D CAD models from natural-language descriptions; second, FormaNova's CAD Studio allows users to render photorealistic product images directly from CAD files, bypassing the need to photograph a physical prototype at all.