Pottery had a specific knock against AI: the handmade details. When people tested image models on glazed mugs, the models clearly understood glaze, but they fumbled the construction, especially the handle, attaching it awkwardly or unconvincingly, because they had not fully learned how a real pulled handle joins a thrown body. So pottery was a good place to check whether that limitation still holds.
It does not. I ran one brief, a reactive-glaze stoneware mug, teal breaking to cream over a speckled clay body, a pulled handle, an unglazed foot, 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 rendered a convincing drip glaze, and all four got the handle right, the detail that used to give them away. This is the ceramics entry in our product-photography series, alongside the skincare, jewelry, supplements, makeup, food and beverage, footwear, candles, clothing, furniture, electronics, handbags, sunglasses, glassware, flowers, watches, perfume, packaging, pet products, toys, textiles, cookware, stationery, drinkware, soap, 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 vivid reactive drip glaze and the most authentic thrown details.
- Glaze and handle both held: all four rendered a coherent drip glaze and a believable, properly attached handle, the construction weak spot has caught up.
- The caveat: a reactive glaze is unique to each firing, so the generated one is a plausible invention, not your specific piece.
If you only remember one thing: pottery is now a reliable category, glaze and handle both, so choose on glaze detail and use a reference when the exact piece has to be yours.
The test, model by model
One brief, four models, same prompt. I judged the reactive glaze first, then the handle, the documented weak spot, then the thrown form.
Seedream 4.5 rendered the glaze the way a potter hopes a photo will. The teal breaks and drips over the speckled cream body with real reactive-glaze behavior, the glossy glaze pools and runs, the matte clay shows iron speckle, the unglazed foot reveals raw clay. And the handle, the part AI was supposed to fumble, is properly pulled and joined, with a comfortable curve. This is the material and organic strength it shows on soap and textiles, applied to glaze. Best result, lowest cost of the photoreal options.
Nano Banana 2 produced the most convincingly hand-thrown mug: the throwing rings on the clay body are visible, the reactive glaze breaks naturally over the speckle, and the handle is believable and well joined. Styled with linen and a plant, it is a complete, rustic-maker shot. Slightly more muted than Seedream's vivid drip, but the most authentic throwing surface, at a third of GPT's cost.
GPT Image 2 gave a clean, correct mug with a tidy horizontal color break, a speckled lower body, an authentic unglazed foot, and a well-attached handle. It is a touch rounder and more uniform than the hand-thrown look of Nano, but accurate and believable. As with the other forgiving categories, its higher price buys no real advantage here.
FLUX.2 Pro produced a glossy, believable mug with a nicely pooled glaze band, visible throwing rings, and a properly joined handle on a rounded belly form. The glaze has real glossy depth, and at the lowest cost it is a strong glaze-forward result. None of FLUX's usual softness on the finest detail mattered on a surface this organic.
The comparison
| Model | Reactive glaze | Handle (the weak spot) | Thrown details | Rough cost/image |
|---|---|---|---|---|
| Seedream 4.5 | Best, vivid drip | Believable | Best, foot + rim | ~4.8 credits |
| Nano Banana 2 | Coherent, muted | Believable | Best throwing marks | ~9.3 credits |
| GPT Image 2 | Clean color break | Believable | Clean foot | ~26.4 credits |
| FLUX.2 Pro | Glossy, pooled | Believable | Visible rings | ~3.6 credits |
Credit costs are first-hand from this test on Masonry; per-image rates move, so check current pricing.
Why pottery quietly caught up
Pottery is a useful checkpoint in this series, because it shows a documented AI weakness that has closed.
The glaze was never the problem. A reactive glaze is glossy, organic, material-driven, which is exactly this generation's strength, so the color break, the drip, the glossy-over-matte contrast all came out well, Seedream best. If glaze were the only test, pottery would be trivially easy.
The handle was the problem, and it is mostly fixed. The earlier knock on AI pottery was the construction, the handle attached awkwardly because the model had not learned how a pulled handle joins a thrown body. In this test, all four attached a believable, comfortable handle. It is the same arc as handbag construction, the structural details that used to give AI away have largely caught up for standard forms. Unusual handles, lids, spouts, or sculptural pieces are still worth checking.
But the glaze is never your glaze. Reactive glazes are unique to each firing, so the generated one is a plausible invention, not your specific piece. For a true SKU, generate from a reference of the real mug. For concept and lifestyle, a prompt is plenty.
How to shoot your pottery line without a studio
The workflow is the roundup approach, applied to a category where per-piece photography never paid off. Trust the glaze and the handle, and choose on glaze detail, Seedream for the most vivid drip. Lean into the rustic, hand-thrown styling buyers expect, which all four render well. And for an exact piece, feed a reference photo so the glaze break and form are truly yours, then let AI produce the lifestyle and seasonal shots a small studio could never afford per mug.
With the Masonry CLI you can compare the glaze across models, or pass your real mug as a reference to keep the exact piece:
masonry image "handmade stoneware mug, reactive teal drip glaze over speckled clay, wooden table, soft light, photoreal" --model seedream-4-5 masonry image "place this exact mug on a sunlit windowsill, keep the glaze and form" --ref ./real-mug.png --model gemini-3.1-flash-image-preview
The bottom line
Pottery is a category that quietly caught up. The reactive glaze that is the visual hero of a mug came out beautifully on all four models, and the handle that used to expose AI was believable on every one, with Seedream 4.5 producing an Etsy-ready drip glaze at the lowest cost. The only thing to remember is that a reactive glaze is a unique invention, so use a reference photo when the exact piece 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 pottery from one place with the Masonry CLI.
