Glassware photography is an optics problem wrapped around a geometry problem. Convincing highlights and projected color can make an image feel real while the sold bowl profile, rim, wall, stem, foot, seam, tint, engraving, dimensions, capacity, condition, or set configuration has changed.
This test sent one fictional red-wine-glass brief through GPT Image 2, Seedream 4.5, FLUX.2 Pro, and Nano Banana 2. The four original outputs below show useful differences in displayed projected light, vessel type, bowl/rim crop, table reflection, background-distortion cues, scene, and warmth. No real glass, drawing, dimensions, capacity, material specification, tint standard, seam/finish record, condition standard, set/packaging record, lighting diagram, optical measurement, safety test, or product data was supplied.
Evidence boundary: this is one text prompt and one displayed output per model. The observations describe visible fictional glassware scenes—not physically correct optics, exact geometry, material, dimensions, capacity, tint, hidden condition, safety, model reliability, current price/value, shoot replacement, or preservation of a real glassware SKU.
Quick answer
- Strongest displayed projected red-light pattern: GPT Image 2 in this run.
- Tightest displayed bowl-and-rim macro: Seedream 4.5 in this run; it changed the vessel to a coupe.
- Clearest displayed table reflection: FLUX.2 Pro in this run.
- Fullest displayed window scene: Nano Banana 2 in this run.
- What all four establish: each route made one plausible fictional red-wine vessel scene.
- What none establish: exact glassware geometry, physically correct refraction/caustics, dimensions, capacity, material, tint, condition, set configuration, safety, or repeated reliability.
One output per route does not establish a universal winner or that clear glass and its optics are solved. Plausible light cues can coexist with a different vessel: Seedream changed the briefed stemmed wine glass into a coupe.
First define what the customer receives
| Offer type | Fidelity target | Required evidence |
|---|---|---|
| Exact single-piece SKU | One exact manufactured or one-of-one vessel with defined geometry, dimensions, capacity, tint, finish, marks, and condition. | Approved multi-angle sample, drawing/measurements, capacity/weight, material and finish records, condition, packaging, fulfillment link, and exact-item/SKU mapping. |
| Fixed set or variant | A disclosed count and exact size/shape/tint/pattern configuration. | Approved sample for every variant, set count, matching tolerance, dimensions/capacity, etching/cut/color, accessories, packaging/dividers, thumbnail mapping, and current SKU record. |
| Hand-blown or naturally variable range | A representative vessel within a disclosed artisan variation range. | Multiple representative samples and allowed range for profile, dimensions, capacity, weight, wall/rim/stem/foot, bubbles/inclusions, tint, marks, condition, matching, and representative-photo disclosure. |
A generated hero must not turn one glass into a set, a stemmed wine glass into a coupe, a clear variant into tinted glass, or wine and props into included products.
The controlled concept brief
The original article did not publish the exact prompt. A bounded version matching the displayed subject is:
Fictional brand-free clear stemmed red-wine glass, half-filled with dark red liquid, tall tapered bowl, thin-looking round rim, straight centered stem, circular foot, dark warm wood table beside a window, soft directional light, subtle projected red-light and reflection cues, square barware product photograph. No material claim, capacity claim, logo, engraving, person, bottle, food, second glass, package, price, badge, or text.
This prompt can test art direction. It cannot preserve a real SKU because the models invent the bowl, rim, wall thickness, liquid line, stem, joins, foot, seams, tint, clarity, inclusions, projected light, reflection, background distortion, dimensions, capacity, hidden side, and condition.
Four first-hand outputs
GPT Image 2 produced the strongest visible red-light pattern on the table, alongside a dark shadow and bright highlight region. The displayed crop cuts off the rim and upper bowl, so it cannot support a full-profile, opening-diameter, rim-condition, height, or capacity check. The light pattern is a bounded appearance observation—not proof of a physically correct caustic, refraction, liquid volume, glass index, light source, or product geometry.
Seedream 4.5 made the closest view of the displayed rim, bowl wall, liquid line, highlights, and background-distortion cues. It also made a coupe instead of the requested tall stemmed wine glass. That is product-identity drift: attractive optics do not compensate for the wrong vessel type, bowl profile, stem/foot configuration, dimensions, or capacity.
FLUX.2 Pro emphasized a mirror-like reflection of the glass and liquid on the polished tabletop, plus a luminous-looking liquid region. That supports a scene-direction observation. It does not establish whether the reflection is optically consistent, whether the tabletop belongs in the listing, or whether the bowl, rim, stem, foot, liquid level, dimensions, capacity, and condition match a real SKU.
Nano Banana 2 made the fullest context view, including a window, outdoor background, stemmed vessel, liquid, shadow, and warm projected-light cue near the base. The scene provides more surrounding evidence than a macro, but still cannot establish optical correctness, measured geometry, clarity/tint, capacity, material, hidden defects, or included configuration.
What this comparison supports
| Question | What the four images show | What remains untested |
|---|---|---|
| Can the routes make a red-wine glass scene? | Yes, once each for this fictional brief. | Reliability across seeds, vessel types, references, liquids, lighting, surfaces, views, prompts, and current routes. |
| Do visible light and scene treatments differ? | Yes; projected red patterns, highlights, reflections, background distortion, crop, warmth, and context vary. | Physically correct light transport, refractive index, dispersion, material, measured light source, geometry, and real-surface behavior. |
| Is the vessel type preserved? | Not consistently; Seedream displayed a coupe rather than the briefed tall stemmed glass. | Exact bowl/rim/wall/stem/foot profile, dimensions, capacity, weight, seams, tint, engraving, and view-to-view match. |
| Is a real glassware SKU preserved? | No evidence; no product source was supplied. | Drawing/sample match, set/variant, marks, inclusions, condition, packaging, accessories, inventory, and fulfillment mapping. |
| Are material, safety, or use claims established? | No. | Crystal/lead-free/tempered/borosilicate, food contact, capacity, thermal shock, hot/cold, dishwasher/microwave/freezer, strength, certification, and warranty. |
The old article said glass was solved, described the rendered light as real physics, named each model the best at an optic, promised exactness from references, and made unverified cost comparisons. The defensible conclusion is narrower: four routes made plausible fictional scenes once, and their displayed vessel, projected-light, reflection, background-distortion, crop, and staging treatments differ.
Glassware fidelity checklist
Build the rejection sheet from approved physical samples, front/profile/top/base and detail photography, drawings and measurements, capacity/weight, material and manufacturing records, tint/clarity/finish standards, condition/defect tolerances, set/variant records, safety/use testing, packaging, fulfillment mapping, and current product data:
- Bowl and rim: vessel type, bowl profile, taper/flare, roundness, opening/rim diameter, wall and rim thickness, rolled/cut/fire-polished-facing edge, lip, chips/fleabites, waviness, seams, distortions, etching/cut pattern, decoration, and top-view symmetry.
- Stem, joins, and foot: stem length/thickness/straightness, pulled/molded-facing appearance, bowl and foot joins, knobs, bubbles, seams, foot diameter/thickness/flatness/roundness, rocking-facing contact, centering, tilt, stress/condition, and hidden-side agreement.
- Material and finish: approved clear/tinted/colored/opaque-facing appearance, clarity, bubbles/seeds/cords/inclusions, striae, mold seams, polish, frosting, iridescence, coating, engraving, decal/gilding, recycled-facing variation, and no unsupported crystal/composition/process inference.
- Dimensions and capacity: height, maximum/opening/base diameter, wall/rim/stem/foot dimensions, capacity at defined fill line, overflow capacity where used, weight, balance, center of gravity, scale, stack/nest clearance, and manufacturing tolerance.
- Optics and inspectability: neutral factual view plus approved lifestyle view; front/profile/top/base visibility, transmitted/reflected light, background distortion, projected-light pattern, reflection, highlight, polarization-facing effect, color cast, glare, occlusion, liquid/prop influence, and no hidden geometry or damage.
- Condition and safety-facing inspection: rim chips, cracks, scratches, scuffs, clouding, water spots, residue, bubbles/inclusions versus defect tolerance, stem/foot damage, repairs, temper marks where applicable, packaging damage, seconds status, and current condition.
- Set and fulfillment: exact unit/set count, size/shape/tint/pattern variant, matching tolerance, carafe/decanter/lid/straw/accessories, packaging/dividers, labels/barcode, inventory/lot, substitutions, thumbnail mapping, shipping state, and sold configuration.
- Claims and rights: glass/crystal/composition, lead-free/tempered/annealed/borosilicate, hand-blown/origin, capacity, food contact, hot/cold, thermal shock, microwave/dishwasher/freezer, impact/chip resistance, stackable, recycled/sustainable/certified, delivery, and warranty remain in approved records.
If the scene looks optically convincing but the bowl, rim, wall, stem, foot, tint, pattern, dimensions, capacity, condition, set count, packaging, or claims differ, it is the wrong glassware asset.
A geometry-first glassware workflow
1. Capture authoritative sources
Photograph the approved vessel empty and, when relevant, at a defined fill line: front, back, both profiles, top, base, rim, bowl, stem, joins, foot, seams, decoration, marks, inclusions, condition, set pieces, packaging, and scale. Store drawings, dimensions, capacity/weight, material/process records, tint/finish standards, tolerances, safety/use tests, and current variant data.
2. Build the lighting scene separately
Generate or photograph an empty dark/light sweep, table, bar, shelf, window, or editorial plate at the final crop. Composite approved glassware photography or a production-derived transparent render, then build reflection, shadow, transmitted color, background distortion, and projected light deliberately. Keep at least one neutral empty-vessel view for geometry and condition.
3. Test a constrained reference edit
masonry image "Change only the environment to a dark warm wood table beside a window. Keep the supplied glassware unchanged: exact vessel type, bowl profile, rim, wall, stem, both joins, foot, seams, tint, clarity, inclusions, engraving/pattern, dimensions, fill line, condition, set count, and sold configuration. Add one controlled reflection and subtle projected-light cue. Add no bottle, food, person, second glass, logo, text, price, badge, package, or implied included prop." \ --model gemini-3.1-flash-image-preview \ --ref ./approved-glassware-sku.png \ --aspect 1:1 \ --output glassware-scene-candidate.png
The instruction is not a geometry, material, optical, condition, safety, or SKU lock. Compare the candidate with approved physical/multi-angle sources, drawings, measurements, capacity/weight, condition tolerances, set/variant data, safety/use tests, packaging, and fulfillment records. Composite approved product photography or a production-derived render when exactness matters.
For a fictional concept:
masonry image "Fictional brand-free clear stemmed red-wine glass, half-filled with dark red liquid, tall tapered bowl, thin-looking round rim, straight centered stem, circular foot, dark warm wood table beside a window, soft directional light, subtle projected red-light and reflection cues, no material claim, capacity claim, logo, engraving, person, bottle, food, second glass, package, price, badge, or text" \ --model gpt-image-2 \ --aspect 1:1 \ --output glassware-concept.png
Glassware acceptance sheet
| Area | Pass condition |
|---|---|
| Source match | Candidate is compared with approved front, back, profiles, top, base, rim, bowl, stem/joins, foot, seams, decoration, marks/inclusions, condition, scale, drawings, dimensions, capacity, weight, material/finish, set, testing, packaging, fulfillment, and product data. |
| Geometry and capacity | Vessel type, bowl/rim/wall, stem/joins, foot/base, roundness, centering, straightness, contact, dimensions, defined fill line, capacity, weight, balance, stack/nest clearance, tolerances, and hidden-side logic match. |
| Material, optics, and condition | Approved clear/tint/finish, seams, inclusions, decoration, neutral and lifestyle lighting, reflection/transmission/background distortion/projected-light cues, chips/cracks/scratches/clouding/residue, defect tolerance, and physical inspection match. |
| Set and fulfillment | Exact unit/set count, size/shape/tint/pattern option, matching tolerance, accessories, packaging/dividers, labels, inventory/lot, substitutions, shipping state, sold configuration, thumbnail mapping, and fulfillment record match. |
| Claims and rights | No crystal/composition, lead-free/tempered/borosilicate, hand-blown/origin, capacity, food-contact, thermal-shock/hot/cold, microwave/dishwasher/freezer, strength, sustainable/certified, delivery, or warranty claim is inferred from appearance. |
Bottom line
These four outputs are useful as a vessel, projected-light, reflection, background-distortion, crop, and staging art-direction comparison. They are not evidence that glass optics are solved, rendered light is physically correct, a reference preserves geometry, a fictional glass is ready to list, or one route is always the best value.
Use fictional concepts to choose a lighting direction. Use approved physical and multi-angle product sources, drawings and measurements, capacity/weight, material and finish standards, condition tolerances, set/variant records, safety/use tests, packaging, fulfillment mapping, and product facts to decide whether an asset can ship. Compare the broader AI product photography model review, build a controlled scene in Masonry’s product photography tool, or automate candidates with the Masonry CLI.
