Eyewear is a paired product and an optical product. A listing image has to preserve lens shape, eyewire, bridge, hinges, temples, frame pattern, tint, markings, and the relationship between reflections and what is visible through each lens.
This test sent one fictional tortoiseshell-frame and brown-gradient-lens brief through SeedDream 4.5, GPT Image 2, Nano Banana 2, and FLUX.2 Pro. The four original outputs below show useful differences in frame shape, angle, material cues, tint, transmission, and reflection. No real eyewear SKU was supplied.
Evidence boundary: this is one prompt and one displayed output per model. The sunglasses were fictional, no approved frame, lens specification, or measurement set was supplied, and no optical, UV, polarization, impact, fit, safety, compliance, or repeated-reliability test was possible. The observations describe visible cues in four images; they are not model pass rates or proof of lens performance.
Quick answer
- Strongest displayed transmitted-tint cue: the Seedream candidate shows the background through the lenses with a visible brown tint.
- Cleanest displayed front view: GPT Image 2 made the most directly comparable bilateral presentation in this run.
- What all four prove: each route made one plausible fictional sunglasses concept.
- What none prove: preservation of a real frame, hinge, temple, lens curve, tint specification, coating, marking, or optical performance.
- Safer listing path: generate the set and light, then composite approved frame-and-lens photography.
The controlled brief
The original article summarized the prompt. A reproducible version matching the tested subject is:
Fictional brand-free tortoiseshell acetate sunglasses, medium rounded-rectangular lenses, centered keyhole bridge, brown gradient-tinted lenses darker at the top, small gold-tone hinges and rivets, front three-quarter view on pale limestone, soft camera-left studio light, visible but restrained lens reflections, natural contact shadow, square product photograph. No text, logos, face, hands, case, cloth, extra glasses, or watermarks.
This text-only brief can test art direction and visible optical cues. Without a real source, the frame, pattern, hardware, and lenses are inventions.
Four first-hand outputs
Seedream 4.5 made the most obvious transmission cue: the stone remains visible behind the lenses with a brown cast, and the shadow carries some tint. The frame and gold-tone hardware also read clearly. Those pixels support an art-direction observation; they do not verify lens material, base curve, transmission percentage, UV performance, polarization, or coating.
GPT Image 2 provided the easiest view for a left-right comparison. The lens outlines, brow, bridge, and hinge regions look broadly aligned at article scale. “Broadly aligned” is not a dimensional pass: overlay the approved frame, compare endpoints and angles, and inspect intentional left-right differences in pattern, hardware, and markings.
Nano Banana 2 made a conventional product composition with a readable frame and consistent-looking tint. The tortoiseshell pattern should not be expected to mirror perfectly; for a real acetate SKU, compare pattern placement, translucency, edge polish, frame thickness, rivets, and temple details with the actual product.
FLUX.2 Pro chose a noticeably rounder silhouette. Material and tint cues remain readable, but frame identity changed because no product source constrained it. The result is useful for concept exploration and irrelevant as evidence that a real frame has been preserved.
What the comparison actually supports
| Question | What these four images show | What remains untested |
|---|---|---|
| Can the routes make a sunglasses concept? | Yes, once each for this brief. | Reliability across seeds, frames, crops, and prompts. |
| Do both sides look coherent? | No obvious article-scale bilateral break in the displayed candidates. | Dimensional overlays, hinge alignment, temple angles, lens seating, and behavior on other frames. |
| Do tinted-lens cues differ? | Yes; transmission, gradient, reflections, and shadows vary visibly. | Lens material, base curve, VLT, polarization, UV, coatings, prescription, and optical quality. |
| Is a real SKU preserved? | No evidence; no source frame was supplied. | Every dimension, pattern, hardware part, lens spec, mark, and sold configuration. |
| Are safety or optical claims supported? | No. | UV, polarization, prescription, impact, coatings, compliance, fit, and safety. |
The earlier version called symmetry largely solved and said one lens behaved like real glass. The defensible finding is narrower: the four candidates look broadly coherent at article scale, and one shows a stronger transmitted-tint cue. Neither observation establishes geometry or optics beyond these pixels.
Product fidelity checks for real sunglasses
Build the rejection sheet from approved packshots, measurements, drawings, artwork, and lens specifications:
- Front geometry: overall width, lens and eyewire shape, A and B dimensions, bridge width and style, brow line, rim thickness, lens spacing, pantoscopic appearance, and frame tilt in the shot.
- Nose and fit hardware: bridge, nose pads or saddle, pad arms, pad shape, contact points, and any adjustable metalwork.
- Hinges and temples: hinge type and position, rivets, screws, decorative hardware, temple angle, length, taper, core wire visibility, tips, and left-right markings.
- Frame material and pattern: acetate or other material cues, tortoiseshell pattern placement, translucency, polish, edge treatment, color, inclusions, and approved material language.
- Lenses: shape, seating, edge thickness, curvature cues, base color, gradient direction and strength, mirror or flash treatment, reflections, transmission, shadows, and relationship to the background.
- Artwork and marks: logos, model name, size code, color code, country, compliance and certification copy, temple printing, lens marks, stickers, and every visible character.
- Sold configuration: case, cloth, spare parts, packaging, tags, lens option, prescription status, and included accessories must match what the customer receives.
- Product facts: UV category, VLT, polarization, prescription, filtering, coatings, impact performance, material, fit, compliance, and safety remain in approved data and test records.
If a frame looks plausible but its bridge, hinge, temple, pattern, lens seating, or tint differs, it is the wrong SKU.
A reference-first eyewear workflow
1. Prepare authoritative sources
Use straight front, three-quarter, top, underside, side, hinge, temple-inside, temple-tip, lens-edge, markings, case, and packaging views plus dimensions, artwork, frame color target, and lens specification. Name each source's role.
2. Generate the environment separately
Create an empty stone, mirrored, beach, dashboard, or editorial plate with controlled reflections and the final crop. Composite approved eyewear photography, then construct contact shadow and reflected environment deliberately. This preserves frame and lens truth.
3. Test a reference-conditioned scene
masonry image "Change only the environment to pale limestone with soft camera-left studio light. Keep the supplied sunglasses unchanged: lens and eyewire shapes, bridge, brow, rim thickness, nose hardware, hinges, rivets, temples, tips, frame pattern, lens seating, tint gradient, reflections, logos, size marks, and every visible character. Add one natural contact shadow. Add no face, hands, case, cloth, text, or extra glasses." \ --model gemini-3.1-flash-image-preview \ --ref ./approved-sunglasses-three-quarter.png \ --aspect 1:1 \ --output sunglasses-scene-candidate.png
The instruction is not an eyewear lock. Compare the candidate with every approved view and composite the source product when exact geometry, pattern, or optics matter.
For a fictional concept:
masonry image "Fictional brand-free tortoiseshell acetate sunglasses, rounded-rectangular frame, keyhole bridge, brown gradient lenses darker at top, small gold-tone hinges, three-quarter view on pale limestone, soft camera-left light, restrained reflections, no text or logos, square product photograph." \ --model seedream-4-5 \ --aspect 1:1 \ --output sunglasses-concept.png
Sunglasses acceptance sheet
| Area | Pass condition |
|---|---|
| Source match | Candidate is compared with approved front, top, side, hinge, temple, lens edge, markings, and accessory sources. |
| Front geometry | Lens and eyewire shapes, bridge, brow, rims, spacing, dimensions, tilt, and seating match. |
| Hardware and temples | Nose hardware, hinges, rivets, screws, temples, core cues, tips, angles, and markings match. |
| Material and pattern | Frame material cues, pattern placement, translucency, polish, edges, and target color match. |
| Lenses | Shape, seating, edge, curvature cues, base color, gradient, reflections, transmission, and shadows follow the approved source and brief. |
| Artwork and marks | Every approved character and mark matches; no invented logo, size, compliance line, sticker, or background text appears. |
| Product facts | No UV, VLT, polarization, prescription, coating, impact, material, fit, compliance, or safety fact is inferred from the image. |
Bottom line
The four visuals belong here because they show useful differences in frame invention, viewing angle, tint, transmission, and reflection. Their value is the controlled comparison—not a claim that symmetry or lens physics is solved.
Use the images to choose a scene direction. Use approved frame sources, measurements, lens specifications, and test records to decide whether an asset can ship. Compare other controlled tests in the AI product photography model review, or build a reference-first workflow in Masonry's product photography tool and automate candidates with the Masonry CLI.
