On-pet photography has two independent truth problems: the animal must remain plausible and authorized, and the wearable product must remain the correct SKU with a fit that does not imply unsupported safety. A friendly expression and detailed fur can distract from the wrong buckle, adjustment range, D-ring, tag, bandana construction, or neck clearance.
This test sent one fictional golden-retriever collar-and-bandana brief through Seedream 4.5, Nano Banana 2, GPT Image 2, and FLUX.2 Pro. The four original outputs below show useful differences in visible fur, crop, expression, collar, tag, bandana, and porch treatment. No real pet, identity release, measurements, product, size chart, closure/hardware specification, fit standard, safety data, packaging, or product record was supplied.
Evidence boundary: this is one text prompt and one displayed output per model. The observations describe visible fictional on-pet concepts—not pet identity, breed verification, anatomical correctness, measured fit, comfort, safety, product reliability, rights clearance, price/value, shoot replacement, or preservation of a real SKU.
Quick answer
- Most detailed displayed fur and close product texture: Seedream 4.5 in this run.
- Fullest displayed porch composition: Nano Banana 2 in this run.
- Cleanest displayed warm on-pet concept: GPT Image 2 in this run.
- Softest displayed fur and product detail: FLUX.2 Pro in this run.
- What all four establish: each route made one plausible fictional dog-with-gear concept.
- What none establish: the correct pet, breed, anatomy, collar, adjustment, buckle, D-ring, tag, bandana, size, fit, comfort, safety, sold configuration, or repeated reliability.
One output per route does not establish a universal winner or a solved common-breed case. A still can look plausible while hiding occluded anatomy and inventing the product geometry that matters most.
The controlled concept brief
The original article did not publish the exact prompt. A bounded version matching the displayed subject is:
Fictional adult golden-retriever-like dog sitting calmly on a porch, wearing a fictional brand-free tan leather-look collar with one brass-look buckle, one adjustment keeper, one D-ring, one blank round tag, and a separate green plaid bandana tied with a visible safe-looking gap. Soft warm morning light, square pet-product photograph. Show four paws where the crop permits. No leash, harness, words, logo, claim, child, other animal, food, toy, watermark, or duplicate gear.
This prompt can test composition and visible pet/gear cues. It cannot preserve an animal or SKU because the model invents the body, pose, occluded anatomy, collar dimensions, hole spacing, hardware, fastening path, tag, bandana, fit, and hidden surfaces. “Safe-looking” is only visual direction, not safety validation.
Four first-hand outputs
Seedream 4.5 produced the closest crop and most pronounced displayed fur strands, with warm light, visible eyes, a tan collar-like strap, gold-toned hardware/tag, and green plaid bandana. A long loose strap tail projects camera-left beyond the apparent keeper area. Without an approved product and fastening diagram, this image cannot establish whether that path is intentional, correctly adjusted, retained, or acceptable for snag risk. It also does not establish animal anatomy, collar width, closure construction, hardware placement, tag engraving, bandana attachment, neck clearance, or comfort.
Nano Banana 2 made the fullest displayed sitting-dog composition, exposing more of the body and paws than the close crops. The collar-like strap and plaid bandana appear seated around the neck at article scale. Because no pet measurements, product dimensions, adjustment rules, or fit standard were supplied, that appearance cannot establish correct anatomy or fit.
GPT Image 2 produced a clean warm portrait with a visible face, fur, collar-like strap, tag, and plaid bandana. It is easy to read as a finished ad, which increases the need for deliberate review: clean presentation is not evidence that the animal, hardware, adjustment, drape, or intended-use fit is correct.
FLUX.2 Pro made a plausible overall pet portrait with softer fur, strap, tag, and bandana detail. Finer hardware and fit cues are less inspectable at article scale. Full-resolution review still needs to trace the complete collar path, closure, keepers, D-ring, tag attachment, fabric edge, knot/fastening, neck contact, and every visible body part.
What this comparison supports
| Question | What the four images show | What remains untested |
|---|---|---|
| Can the routes make an on-pet concept? | Yes, once each for this fictional brief. | Reliability across seeds, animals, poses, gear, crops, references, prompts, and current routes. |
| Do visible pet treatments differ? | Yes; fur detail, expression, crop, pose visibility, warmth, and softness vary. | Identity, breed, age, anatomy, proportions, motion, gait, health, behavior, comfort, and veterinary assessment. |
| Do visible product treatments differ? | Yes; straps, buckles, tags, hardware, plaid, drape, and placement vary. | Exact SKU, dimensions, material, construction, adjustment, load path, closure security, attachment, fit, safety, and repeatability. |
| Is a real product fit preserved? | No evidence; no pet, product, or measurements were supplied. | Neck/body measurements, size selection, hole/slider position, gap, pressure, airway, escape, snag, chafe, coat, load, and intended use. |
| Are performance or safety facts established? | No. | Strength, breakaway behavior, visibility, weather/chew resistance, comfort, suitability, care, compliance, supervision, warranty, and every marketed claim. |
The old article called the anatomy correct, product fit reliable, and common breeds solved. The defensible conclusion is narrower: four routes made attractive fictional dog-with-gear concepts once, and their displayed fur, crop, and product treatments differ.
Pet-product fit and safety checklist
Build the rejection sheet from authorized pet sources, measurements, approved product photography, drawings, specifications, size chart, fit instructions, packaging, test data, and current product data:
- Pet identity and rights: authorized source, owner/model release where required, species/breed description, age class, coat, distinguishing marks, body condition, and allowed edits.
- Visible animal: head, eyes, pupils/catchlights, ears, muzzle, nose, mouth/teeth/tongue, neck, torso, spine, joints, legs, paws/toes, tail, coat direction, occlusion, pose, balance, and ground contact.
- Product identity: SKU, size, colorway, material-facing cues, strap width/thickness, edge/lining, pattern, stitching, holes, keepers, buckle/clip, D-ring, tag, sliders, adjusters, fasteners, labels, marks, and wear.
- Adjustment and fit: measured neck/body range, selected size, adjustment point, overlap, keeper path, gap, orientation, bandana dimensions and fastening, coat compression, skin contact, airway clearance, and manufacturer fit instructions.
- Load and intended use: decorative collar, identification collar, leash attachment, restraint, harness, apparel, or accessory; actual load path, rated hardware, attachment point, supervision, environmental and activity limits.
- Animal-safety review: escape, choke, snag, pinch, chafe, pressure, heat, visibility, entanglement, chew/ingestion, sharp edge, allergy/irritation, mobility, behavior, distress, and appropriate human/veterinary or product-safety review.
- Scene and behavior: approved pose and expression, leash/handler logic, surface traction, weather, temperature, traffic/water/fire hazards, other animals/people, props, rewards, and no implied unsafe use.
- Packaging and claims: box/bag, insert, instructions, size chart, care, warnings, included items, sold configuration, strength, breakaway, reflective/visibility, waterproofing, durability, comfort, safety, suitability, origin, certification, and warranty remain in approved data and tests.
If the pet looks happy but the SKU, adjustment, closure, attachment, fit, intended use, warning, or safety implication differs, it is the wrong pet-product asset.
A measured on-pet workflow
1. Capture authoritative sources
Use authorized front, back, both sides, top, face, full-body, neck/body, pose, and identifying views of the pet. Photograph the product flat, open, closed, inside/outside, hardware, adjustment range, holes/sliders, fastening path, label, marks, packaging, and scale. Add pet measurements, size selection, fit instructions, intended use, safety limits, test data, and current product data.
2. Build the scene separately
Generate or photograph an empty porch, studio, park, home, travel, seasonal, or editorial plate at the final crop. Composite approved pet and product photography when possible, then build contact, shadow, fur occlusion, strap contact, and background depth deliberately. This keeps identity, product, and fit reviewable.
3. Test a constrained two-reference edit
masonry image "Place the authorized pet from the first reference on a simple warm porch and fit only the approved collar and bandana from the second reference according to the supplied measured size. Keep the pet unchanged: identity, coat, marks, face, eyes, ears, muzzle, body proportions, legs, paws, tail, and pose. Keep the product unchanged: strap width, color, stitching, holes, keeper, buckle, D-ring, tag, hardware, bandana pattern, edge, dimensions, and fastening. Preserve the approved adjustment point and visible fit gap. Add no leash, logo, words, claim, child, other animal, or duplicate gear." \ --model gemini-3.1-flash-image-preview \ --ref ./authorized-pet-reference.png \ --ref ./approved-collar-bandana-product.png \ --aspect 1:1 \ --output on-pet-scene-candidate.png
The instruction is not an identity, product, fit, or safety lock. Compare the candidate against every authorized pet view, product source, measurement, fit instruction, intended-use rule, and safety record. Composite approved pet/product photography or run a supervised live fit when exactness or safety matters.
For a fictional concept:
masonry image "Fictional adult golden-retriever-like dog sitting calmly on a porch, fictional tan leather-look collar with one brass-look buckle, one keeper, one D-ring, blank round tag, separate green plaid bandana, soft morning light, no leash, harness, words, logo, claim, child, other animal, food, toy, or duplicate gear" \ --model seedream-4-5 \ --aspect 1:1 \ --output pet-product-concept.png
Pet-product acceptance sheet
| Area | Pass condition |
|---|---|
| Source match | Candidate is compared with authorized pet views and approved flat/open/closed product, hardware, adjustment, fit, label, packaging, scale, measurements, size chart, intended use, safety data, and product data. |
| Pet and anatomy | Identity, marks, coat, face, eyes, ears, muzzle, neck, torso, joints, legs, paws, tail, proportions, occlusion, pose, balance, contact, and behavior pass qualified review. |
| SKU and fit | Size/colorway, strap, material-facing cues, stitching, holes, keepers, closure, adjusters, D-ring, tag, bandana, fastening path, adjustment point, gap, orientation, contact, and measured fit match. |
| Intended use and safety | Load path, attachment, supervision, activity/environment limits, escape, choke, snag, pinch, chafe, pressure, heat, visibility, entanglement, ingestion, irritation, mobility, and warnings are reviewed against approved evidence. |
| Packaging and delivery | Instructions, size chart, care, warnings, box/bag, insert, included items, sold configuration, crop, dimensions, responsive variants, color, retouching, and format are approved. |
| Claims and rights | No breed, anatomy, health, comfort, strength, breakaway, reflective, waterproof, durable, safe, suitable, compliant, origin, certification, warranty, affiliation, or rights-clearance claim is inferred from appearance. |
Bottom line
These four outputs are useful as an on-pet art-direction comparison. They are not evidence that a breed is solved, anatomy is correct, wearable fit is safe, a fictional product is ready to list, or one route is always the best value.
Use fictional concepts to choose the mood. Use authorized pet sources, the approved SKU, measurements, adjustment and fit instructions, intended use, safety evidence, packaging, and product data 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.
