My Product Rendering Workflow After 6 Months
Quick Tip: Click the prompt box above to select it, then press Ctrl+C (Cmd+C on Mac) to copy. Paste directly into Midjourney, DALL-E, or Stable Diffusion!
The Problem with "Photorealistic" as a Goal
Product rendering in generative AI hits a wall that portrait and landscape work often avoids: the client test. A product image must survive scrutiny from art directors who spent decades in physical studios, from pre-press technicians checking CMYK separation, from legal teams verifying label legibility. "Photorealistic" in the prompt doesn't produce this survival-grade output. It produces an image that looks real to casual observation and falls apart under professional examination.
The breakthrough comes from understanding that commercial photography is not a style but a controlled environment with measurable parameters. The original prompt in this workflow contained the seeds of this understanding—studio key light, fill light, specific angles—but scattered them among aesthetic terms that compete for the model's attention. Six months of refinement revealed that hierarchy matters: physical setup first, material specification second, quality parameters last.
Consider the lighting section. "Studio key light from upper left creating specular highlights" describes an effect, not a setup. The model must reverse-engineer: what kind of light, at what distance, through what modifier, would produce this highlight shape? The answer varies wildly across training data. Some photographers use bare heads for hard speculars. Others use large sources for soft, extended highlights. The model averages these into something that looks lit but not professionally lit.
The refined approach specifies: "90cm octabox key light upper left at 45 degrees." This is equipment that exists, with predictable behavior. A 90cm octabox at product photography distance (typically 1-2 meters) produces a specular highlight with specific proportions—wide enough to show product curvature, structured enough to reveal surface texture. The 45-degree angle places the highlight in the upper-left quadrant of reflective surfaces, a convention that reads as intentional rather than accidental. The model no longer guesses at lighting; it applies known physics.
Why Material Description Needs Surface Interaction
Carton packaging seems straightforward: paperboard, plastic film, printed graphics. But each of these materials responds to light differently, and the model's default assumption is plastic—smooth, slightly reflective, uniform. Actual milk cartons use laminated paperboard with specific reflectance: diffuse with subtle specular at grazing angles, texture visible under raking light, print sitting on top of substrate rather than infused.
The original prompt specified "ultra-sharp focus on brand logo and nutritional text." This ensures legibility but ignores material. The refined approach adds nothing explicit about the carton material because the lighting specification does the work. The 90cm octabox at 45 degrees creates raking light across the front plane. This reveals surface texture if present, or confirms smoothness if absent. The fill light ratio (implied by "shadow detail recovery") determines how much of this texture remains visible in shadow areas. Material becomes emergent property of light interaction, not asserted quality.
Strawberries present the opposite problem: the model defaults to generic fruit, and generic fruit in AI training is waxed, flawless, lit for supermarket display. Real strawberries for commercial photography need specific qualities: surface seeds (achenes) in relief, not painted on; calyx with visible veining and serrated edges; flesh that shows subsurface scattering—light penetrating slightly and bouncing back with altered color. The prompt specifies "subsurface scattering on strawberry flesh with translucency at slice edges" because this physical phenomenon distinguishes professional food photography from illustrative rendering. Without it, strawberries read as plastic models.
The distance layering—"three whole strawberries at varying distances from camera"—solves a composition problem that single-plane arrangements create. When all elements share the same focal plane, the image feels like a cutout collage. Varying distances forces the model to apply consistent perspective scaling and atmospheric effects. The nearest strawberry, specified for "macro surface texture," receives full detail resolution. Distant strawberries naturally soften, not through explicit blur instruction but through the model's interpretation of scale and typical optical behavior.
The Physics of Liquid Simulation
Milk splash photography represents one of the most technically demanding product disciplines. Actual high-speed liquid photography requires precise timing: the collision event, the rebound crown formation, the secondary droplet ejection. Each phase has distinct visual characteristics. The original prompt's "explosive creamy pink milk splash forming a perfect liquid crown" captures the aspiration but not the mechanism.
"Perfect" is particularly dangerous. In physics, perfect crowns are rare and often read as artificial. Real splash crowns have asymmetry—primary peak opposite the impact point, secondary peaks where surface tension breaks unevenly. The refined prompt specifies "asymmetric liquid crown with defined primary and secondary splash peaks" because this asymmetry signals authentic physics. The model interprets "asymmetric" as permission to vary element heights and angles, producing visual interest that "perfect" flattens into boredom.
The splash color—"creamy pink" rather than "strawberry milk colored"—matters for material consistency. The model knows milk as white, strawberry as red. Unspecified combination produces saturated pink that reads as artificial flavoring. "Creamy" modifies the pink toward opacity and warmth, suggesting dairy fat content and natural color derivation. This is not aesthetic preference; it's accurate product representation that prevents the "candy" association that oversaturated pink creates.
Caustic light refraction in droplets represents a final quality threshold. Small water or milk droplets in strong light produce focused light patterns—caustics—on surfaces behind them. This effect requires specific conditions: backlighting or strong side light, transparent or translucent liquid, droplet size in the millimeter range. The prompt's "frozen macro droplets with caustic light refraction" establishes these conditions explicitly. Without caustic specification, droplets render as opaque spheres, missing the optical signature that distinguishes liquid from solid particles.
Background as Calibration Environment
Gradient backgrounds in product work serve multiple functions: color harmony with product, separation from competing hues, and neutral environment that doesn't compete for attention. The original prompt's "soft gradient pink background from #FFB6C1 to #FF69B4" provides specific values but no context for their interpretation.
Color values in isolation drift. The model's understanding of #FFB6C1 depends on white balance assumptions, surrounding colors, and intended mood. The refined prompt adds "with 18% gray card neutral reference"—a physical calibration tool that exists in actual studios. This reference forces the model to interpret the pink values relative to known neutral, preventing the warm drift that turns pink into peach or the cool drift that turns it into lavender. The gradient becomes measurable, reproducible, and compatible with downstream color management.
Background separation receives explicit attention through "subtle rim light from behind." This serves two purposes: it creates the light-edge separation that lifts the product from background without artificial drop shadows, and it provides additional highlight structure on the carton edges that reinforces three-dimensional form. The rim light's subtlety—implied by the single word—prevents the halo effect that overdone rim lighting produces, where products appear to glow unnaturally.
The Camera as Final Quality Gate
Medium format specification—"Hasselblad H6D-100c with 120mm macro lens"—does more than invoke expensive equipment. It establishes sensor size (medium format, 53.4×40.0mm), which determines natural perspective and depth of field behavior. It establishes lens characteristics: 120mm macro on medium format produces a specific working distance and magnification ratio, with optical corrections optimized for close focus. It establishes resolution context: 100 megapixels demands optical quality that reveals flaws, forcing the model to generate detail that survives scrutiny.
The aperture specification—f/16—contradicts portrait and landscape conventions where wide apertures create subject separation. In product work, separation comes from lighting and background control, not optical blur. f/16 on medium format extends depth of field sufficiently to render text legible from carton front to back, while remaining below the diffraction limit where sharpness degrades. The 120mm focal length at product distance creates natural perspective without the distortion that shorter macros introduce.
This camera specification is not aspirational. It is a constraint that forces the model to solve for optical plausibility. Images that violate medium format physics—impossible depth of field, wrong perspective for stated focal length, resolution without corresponding detail—read as composite or illustrative rather than photographed. The specification protects against these failures by establishing non-negotiable physical parameters.
Conclusion
Six months of product rendering workflow refinement leads to a single principle: replace descriptive aspiration with physical specification. Every element in the frame—light, material, liquid, background, optics—must be described in terms that could be built in an actual studio. The model's strength is not imagination but interpolation: given sufficient constraint, it finds solutions that satisfy multiple physical requirements simultaneously. The prompt engineer's job is to provide constraints that point toward professional standards, leaving the model to solve for coherence among them.
The resulting image survives the tests that matter: zoom to 100% for texture verification, check CMYK conversion for print viability, examine edge quality for compositing extraction. These are not post-processing fixes. They are emergent properties of a prompt structure that treats commercial photography as engineering problem rather than aesthetic wish.
Label: Product
Key Principle: Replace quality adjectives with physical specifications: every lighting term needs a size and position, every material needs a surface interaction, every spatial relationship needs measurement.