Pet products are sold on the animal. A collar, a bandana, a harness, a coat, the shot that sells it is the one with a happy dog wearing it, and that shot is brutal to produce: live-animal shoots run from a few hundred to several thousand dollars per dog, trained pet models are scarce, and you cannot direct a golden retriever the way you direct a hand model. It is the perfect case for AI, if AI can render a believable animal, because the failure mode is unforgiving: a dog with dead eyes or a warped leg is worse than no photo at all.
So I tested it. I ran one brief, a golden retriever wearing a tan leather collar with a brass tag and a green plaid bandana on a porch, through four of the strongest image models with the same prompt: Nano Banana 2, GPT Image 2, Seedream 4.5, and FLUX.2 Pro. The uncanny pet I feared did not show up: all four produced a believable dog with natural eyes and correct anatomy, and one rendered fur the others could not touch. This is the pet 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, 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 detailed fur, the crispest product detail, a premium golden-hour macro.
- The good news: all four rendered a believable golden retriever, natural eyes, correct anatomy, no uncanny tell, with the collar and bandana fitting believably.
- The caveat: this is a popular breed, the forgiving case. Rarer breeds, mixes, unusual poses, or your specific pet need a reference photo and a careful look at the eyes and legs.
If you only remember one thing: for a common breed AI handles the on-pet shot beautifully, so choose on fur detail and use a reference photo when the pet, or the product, has to be exactly yours.
The test, model by model
One brief, four models, same prompt. I judged the animal first, eyes, fur, anatomy, the uncanny tells, then how the collar and bandana fit.
Seedream 4.5 made the most alive animal, and on a pet that is everything. The fur is rendered hair by hair in warm light, the eyes are natural and bright with no uncanny vacancy, and the product detail is the sharpest of the four: the collar buckle, the engraved brass tag, and the plaid weave of the bandana all read crisply. This is the same material strength it shows on flowers and perfume, and with an organic subject and no fine text to garble, it has nothing to trip on. Best result, lowest cost of the photoreal options.
Nano Banana 2 produced the most complete, ready-to-use shot: a full-body golden retriever on a porch with natural eyes, correct anatomy down to the paws, and a relaxed, believable pose. The collar sits right on the neck and the bandana drapes naturally. It is not the macro showpiece Seedream made, but as a finished on-pet product shot with no anatomy or eye tells, it is excellent and a third of GPT's cost.
GPT Image 2 gave a clean, warm result: natural eyes, correct anatomy, and a slightly more polished studio finish, with the collar and bandana sitting believably. It is a reliable on-pet shot with no uncanny tells, at the highest price of the four. Where it tends to lead, dense text and exact layout, does not come into play on a pet, so here it is solid rather than distinctive.
FLUX.2 Pro produced a convincing dog at the lowest cost: natural eyes, correct anatomy, and a believable collar and bandana fit, in a softer rendering with less fur detail than Seedream's or Nano's. It is the usual FLUX tradeoff, a strong overall image with the finest texture dialed back. For a quick on-pet catalog or social shot it is solid and cheap.
The comparison
| Model | Animal realism (eyes, fur, anatomy) | Product fit | Fur detail | Rough cost/image |
|---|---|---|---|---|
| Seedream 4.5 | Best, fully believable | Crispest detail | Best, hair-level | ~4.8 credits |
| Nano Banana 2 | Believable, complete | Natural fit | Good | ~9.3 credits |
| GPT Image 2 | Believable, polished | Natural fit | Good | ~26.4 credits |
| FLUX.2 Pro | Believable, softer | Natural fit | Softer | ~3.6 credits |
Credit costs are first-hand from this test on Masonry; per-image rates move, so check current pricing.
Why pets are easier than you would think, with a catch
After the packaging test, where a barcode is always fake, a believable dog is a relief, and it fits the pattern this series keeps finding: organic subjects are where AI is strongest.
Popular breeds are a solved case. A golden retriever is one of the most photographed animals on earth, so the models have an enormous amount to draw on, and it shows: natural eyes, correct anatomy, believable fur, on all four. The uncanny-pet fear, dead eyes, warped legs, waxy coat, comes from harder cases, and it did not appear here.
The catch is everything off the common path. Rarer breeds, mixes, action poses, two pets interacting, or a precise product fit are where the eyes and anatomy can still break. The eyes and the legs are the first things to check, because they are the tells that read as wrong instantly. Test your specific case rather than assuming the golden retriever result generalizes.
And the pet is generic, not yours. As everywhere in this series, a text prompt gives you a believable animal, not your customer's actual dog or your real product. The pet-brand workflow that scales is to start from a reference photo of a pet you have rights to and place your product on it, which holds the same animal across colorways and SKUs. For the product, feed a reference too.
How to shoot your pet line without the dog wrangling
The workflow is the roundup approach, applied to a living subject. For concepts and common breeds, trust the result and choose on fur detail. Check the eyes and legs first, since those are the uncanny tells. And for a specific pet or an exact product, feed a reference photo so the animal and the gear are yours rather than a plausible invention, which is also how you keep one consistent dog across a whole collection.
With the Masonry CLI you can compare the fur and fit across models, or pass a reference pet and your real product to keep both consistent:
masonry image "golden retriever on a porch wearing a leather collar and plaid bandana, soft morning light, photoreal" --model seedream-4-5 masonry image "put this exact collar and bandana on this dog, natural fit, photoreal" --ref ./real-dog.png --ref ./real-collar.png --model gpt-image-2
The bottom line
Pet products are an organic case, and organic is where AI is strongest. For a popular breed all four models produced a believable dog with natural eyes, correct anatomy, and a believable product fit, so the choice comes down to fur: Seedream 4.5 for the most detailed, alive coat at the lowest cost, with Nano Banana 2 and GPT Image 2 close behind. Check the eyes and legs, test anything off the common path, and use a reference photo when the pet or the product has to be exactly 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 pet shots from one place with the Masonry CLI.
