Sunglasses looked like a symmetry trap. They are a paired object, two lenses, two temples, two hinges, a centered bridge, all of which have to match, and matching paired parts is exactly the kind of spatial consistency image models are documented to struggle with. A frame with one lens slightly bigger than the other, or temples at different angles, reads as broken instantly. So I expected asymmetry to be the story.
It mostly was not. I ran one brief, a classic tortoiseshell acetate frame with gradient brown-tinted lenses and gold hinges, shot front-on, no logos, through four of the strongest image models with the same prompt: Nano Banana 2, GPT Image 2, Seedream 4.5, and FLUX.2 Pro. All four held the symmetry, even on the straight front-on view. The real test turned out to be the lens itself. This is the sunglasses entry in our product-photography series, alongside the skincare, jewelry, supplements, makeup, food and beverage, footwear, candles, clothing, furniture, electronics, handbags, glassware, flowers, watches, perfume, packaging, pet products, toys, textiles, cookware, stationery, drinkware, soap, ceramics, art prints, earbuds, houseplants, knives, and automotive wheels tests and the broader best AI image model for product photography roundup.
Quick answer
- Best overall, and cheapest photoreal: Seedream 4.5. The most convincing acetate and a tinted lens that behaves like real glass.
- Symmetry held: all four kept the two lenses, temples, and hinges matched, including GPT Image 2 on the hardest straight-on angle.
- The real test is the lens: only Seedream rendered the tint as an optical element that transmits and colors what is behind it. The others painted a convincing but flat gradient.
If you only remember one thing: the frame is easy now, so judge the lens. A real tinted lens is glass you can see through, not a colored shape.
The test, model by model
One brief, four models, same prompt. I judged symmetry first (the expected weak point), then the lens, then the acetate and hardware.
Seedream 4.5 made the most beautiful image and, more importantly, the most physically correct lens. On a tinted-lens product the lens is the hard part, because a real one transmits light and colors whatever sits behind it. Seedream rendered exactly that: the stone surface is visible through the gradient lens, tinted brown, with a believable tinted cast shadow. The tortoiseshell acetate has real depth and the gold hinge detail is intricate. For a sunglasses hero where the lens has to read as glass, this is the one.
GPT Image 2 won the symmetry test outright, on the angle most likely to expose a mismatch. Shot dead front-on, the two lenses are identical, the frame is even, the hinges match, and the tortoiseshell pattern mirrors believably. Its lens is the tradeoff: a clean, convincing gradient that reads more as a colored surface than as glass you can see through. For a catalog front view where symmetry and a clean look matter most, it is excellent.
Nano Banana 2 produced the most everyday-usable result: a clean tortoiseshell wayfarer with matched lenses, a consistent gradient, and tidy gold hinge accents. The symmetry holds and the acetate is believable. Like GPT, its lens is a convincing flat gradient rather than a transmitting optical element, which is fine for most catalog uses and short of the Seedream hero for a close-up.
FLUX.2 Pro gave a clean, symmetric, slightly rounder frame at the lowest cost, with believable acetate and a consistent tint. As across this series, it is a strong overall image, and its lens is the convincing flat gradient rather than the transmitting glass Seedream achieved. It also produced a noticeably different frame shape, a reminder that each model invents its own silhouette.
The comparison
| Model | Symmetry | Tinted lens | Acetate / material | Rough cost/image |
|---|---|---|---|---|
| Seedream 4.5 | Held (angled hero) | Best, transmits like glass | Best, rich depth | ~4.8 credits |
| GPT Image 2 | Best, perfect front-on | Flat painted gradient | Clean | ~26.4 credits |
| Nano Banana 2 | Held | Flat painted gradient | Believable | ~9.3 credits |
| FLUX.2 Pro | Held | Flat painted gradient | Good | ~3.6 credits |
Credit costs are first-hand from this test on Masonry; per-image rates move, so check current pricing.
Why eyewear is a lens problem, not a symmetry problem
I expected to write about asymmetry. The more useful finding is that for a standard frame, symmetry is largely solved, and the real test is the optical behavior of the lens and the exact frame you get.
Symmetry held, which was the surprise. Matching paired parts is a known weak spot for image models in the general case, but modern models have seen so many glasses that for common frame shapes they keep the two sides consistent, even front-on. If you sell conventional frames, the broken-symmetry fear is mostly outdated. It is still worth checking on unusual or ornate frames.
The lens is an optical element, and most models paint it. A tinted lens is not a color, it is glass that transmits light and tints what is behind it. Only Seedream rendered that physics here; the others produced a convincing flat gradient that looks right head-on and less right the moment the lens should be showing something through it. For sunglasses specifically, that is the detail that separates a real-looking hero from a sticker, so judge the lens as glass.
The exact frame is still an identity problem. As with furniture and handbags, a text prompt gives you a category, not your SKU. Each model produced a different silhouette, a wayfarer, a rounder keyhole, a squarer shape, with different tortoiseshell patterning. For your actual product, the frame shape and the pattern are the design, so generate from a reference photo of the real glasses.
How to shoot your eyewear line without a studio
The workflow is the roundup approach, tuned for a product whose hard part is glass. Run your frame through two models and judge the lens as an optical element, does it transmit and tint what is behind it, not just whether the color looks nice. Check symmetry on a front-on shot, the angle most likely to expose a mismatch. And for the actual SKU, feed a reference photo so the frame shape and tortoiseshell pattern are yours rather than a plausible invention.
With the Masonry CLI you can fire the same frame prompt at every model and compare the lens behavior side by side, or pass your real glasses as a reference to keep the exact frame:
masonry image "tortoiseshell acetate sunglasses, gradient tinted lenses, front-on, soft studio light, photoreal" --model seedream-4-5 masonry image "place these exact sunglasses on a sunlit stone surface, keep the frame and tint" --ref ./real-glasses.png --model gpt-image-2
The bottom line
Sunglasses are not the symmetry story I expected. All four models kept the frame symmetric, even front-on, so the choice comes down to the lens and the frame: Seedream 4.5 for a lens that behaves like real tinted glass and the richest acetate at the lowest cost, GPT Image 2 for the cleanest perfectly symmetric front view. Judge the lens as glass, check symmetry head-on, and use a reference photo when the frame has to be yours. See how the same fidelity-first logic plays out across every product type in our best AI image model for product photography roundup, or run your own frames from one place with the Masonry CLI.
