We gave five current image models the same source product and a fixed brief in three scenes: a neutral studio pedestal, a dark reflective surface, and a person holding the product. That produced 13 real Masonry outputs. Every model kept the visible label words. None kept the product pixel-identical, and none earned a spot as the factual primary image.
The short answer, by scene:
| Scene | Closest to the source | Best-looking alternative | Fails the brief |
|---|---|---|---|
| Neutral pedestal (4 models) | Nano Banana 2 | Nano Banana Pro (more proportion drift) | Qwen Image Edit Plus |
| Dark reflective surface (6 images) | Nano Banana 2, 3 of 3 on core identity | Seedream 5 Pro (strongest reflections) | GPT Image 2 missed the dark scene |
| Held in a hand (3 models) | GPT Image 2 (closest grip and framing) | Nano Banana 2 (tighter grip than asked) | Qwen Image Edit Plus (stylized, cap grip) |
Nano Banana 2 is the most consistent starting point across all three. GPT Image 2 keeps text crisp but redraws package geometry in edits, and handled the hand brief best. All of them need merchant review before publishing.
How to read this: NORTHLINE is a fictional product we made for these tests, so we could publish every output. All 13 images are unselected Masonry job returns from August 14 to 16, 2026. Sample sizes are small and unequal (one to three per model per scene), so we show raw counts, not rates. This is a workflow example, not a conversion study or a universal leaderboard.
The source and what cannot change
The first two scenes use a straight-on studio packshot of one clear cylindrical serum bottle. The acceptance sheet locked six visible attributes before generation.
| Product truth | Fixed acceptance rule |
|---|---|
| Silhouette | One clear cylindrical bottle with the same shoulder, base, and proportions |
| Liquid | Vivid orange color and approximately the same fill line |
| Cap | One transparent cylindrical over-cap with no tint or decoration |
| Pump | Three visible brushed-silver tiers with the same relative geometry |
| Label | One centered white rectangle with the same size and placement |
| Text | NORTHLINE, VITAMIN C, and 30 ML, each appearing exactly once |
The hand scene uses a second fictional product, a tall sage bottle, because a small serum bottle held at mid-torso is too small to review. A real SKU needs more evidence than one front photo: other angles, dimensions, material data, approved artwork, packaging, variant, included items, and rights to the source photography.
Scene 1: neutral studio pedestal, four models
Every model received this prompt:
Using the supplied reference image as the immutable product, create a square premium ecommerce secondary-gallery photograph. Place the exact bottle centered on a pale warm-gray limestone pedestal against a softly lit neutral studio wall. Keep the camera straight-on and the product fully visible. Change only the background, surface, and lighting. Preserve exactly: cylindrical clear-glass silhouette and proportions, orange liquid color and fill line, transparent over-cap, three-tier brushed-silver pump geometry, white rectangular label size and placement, and every character of NORTHLINE, VITAMIN C, 30 ML. Do not add or remove product parts. No fruit, leaves, water, hands, box, extra bottle, claims, icons, headline, price, watermark, or extra text.
The live Masonry routes were gemini-3.1-flash-image-preview (Nano Banana 2), gemini-3-pro-image-preview (Nano Banana Pro), gpt-image-2, and qwen-image-edit-plus. The first, second, and fourth runs used seed 20260814; GPT Image 2 did not expose a seed parameter in its live route. All used one reference image and a square 1024-pixel output.
Masonry's CLI is asynchronous: masonry image returns a job ID, then the completed file is downloaded separately.
# Originally run on gemini-3.1-flash-image-preview (Nano Banana 2), which Masonry retired in November 2026; gemini-nano-banana-2.1 replaced it. masonry image "<fixed prompt above>" \ --model gemini-nano-banana-2.1 \ --ref ./northline-source.png \ --aspect 1:1 \ --seed 20260814 masonry job wait <job-id> masonry job download <job-id> --output ./nano-banana-2.png
Run masonry models list --type image before putting a model key into a durable production script. A Seedream 5 Pro attempt in this scene failed before generation for insufficient credits, so Seedream appears only in Scene 2.
Nano Banana 2: closest product relationship in this run
This candidate made the smallest overall product change in our visual review. It is the best starting point of these four for a secondary gallery slot, subject to full-resolution approval. “Closest” does not mean identical: the cap height, pump rendering, orange fill, base thickness, and reflections move relative to the source.
Nano Banana Pro: strongest finished scene, more proportion drift
The lighting, surface, and product separation are strong. This is the most campaign-ready composition in the run, but the product is less source-faithful than its polish suggests. That makes it a good example of why aesthetics and SKU accuracy need separate scores.
GPT Image 2: excellent text, material product redraw
GPT Image 2 demonstrates that text accuracy alone is not product accuracy. The candidate is coherent and commercially polished, but it depicts a materially different package geometry. We would reject it for a factual PDP slot and consider only the environment for compositing.
Qwen Image Edit Plus: visible words survive, identity drifts
This is a useful failure case. The result looks like the same product family at a glance, yet it changes the visual identity and ignores the requested pale neutral treatment. It should be rejected rather than rescued with a tighter crop.
Scene 1 scorecard: product truth before beauty
| Model | Exact text | Color / fill | Label geometry | Bottle / pump geometry | Brief adherence | Merchant disposition |
|---|---|---|---|---|---|---|
| Nano Banana 2 | Pass | Review | Review | Review | Pass | Secondary candidate after approval |
| Nano Banana Pro | Pass | Review | Review | Review | Pass | Campaign candidate after approval |
| GPT Image 2 | Pass | Review | Fail | Fail | Pass | Keep scene; composite approved product |
| Qwen Image Edit Plus | Pass | Review | Fail | Fail | Fail | Reject |
One candidate per model cannot show variance. Scene 2 adds repeat attempts for exactly that reason.
The model documentation supports why these routes belong on a current shortlist; it is not evidence of exact SKU fidelity. Google documents image editing and multiple reference inputs for Gemini image generation. OpenAI documents image inputs and edits for GPT Image 2. Qwen publishes its image-editing model and workflow. Capabilities are not acceptance results.
Scene 2: transparent and reflective, six candidates
Clear and mirrored products are harder than they look, because the product is not only its outline. The cap edge, liquid boundary, internal pump, metal finish, and the floor reflection all shape what the shopper believes is being sold. We kept the same serum and asked for a darker, reflective scene:
Edit the supplied fictional product reference into a square ecommerce secondary image on a polished dark cobalt stone surface with a soft blue-gray gradient backdrop. Keep the product front-facing at the same scale and preserve the exact clear cylindrical bottle and clear cap, the three-tier brushed-silver pump, circular nozzle, orange transparent liquid and fill line, white rectangular label, and the exact three text lines NORTHLINE, VITAMIN C, and 30 ML. Add one soft camera-left stripbox highlight, a natural contact shadow, and one controlled mirror reflection directly below the bottle. Do not add props, hands, droplets, bubbles, extra products, extra text, logos, or claims. Do not change bottle geometry, cap height, pump tiers, nozzle, liquid color, fill level, label shape, spelling, spacing, or quantity.
Nano Banana 2 ran three seeds (41101 to 41103), Seedream 5 Pro two (51201, 51202), and GPT Image 2 one. Credits ran out before the counts were equal, which is why the table shows denominators.
| Model | Candidates | Core identity | Exact text | Dark reflective scene | Approved as factual primary | Practical read |
|---|---|---|---|---|---|---|
| Nano Banana 2 | 3 | 3 / 3 | 3 / 3 | 3 / 3 | 0 / 3 | Most consistent starting point for secondary creative |
| Seedream 5 Pro | 2 | 2 / 2 | 2 / 2 | 2 / 2 | 0 / 2 | Keep the scene; composite the approved product |
| GPT Image 2 | 1 | 1 / 1 | 1 / 1 | 0 / 1 | 0 / 1 | Clean text; one sample, weaker scene match |
"3 / 3" is not a reliability claim. The narrower result: in this setup, every Nano Banana 2 attempt kept the core attributes, and all six images still needed a strict product review. Transparent and reflective products fail in their own ways:
- Edge drift: a clear cap or glass base widens, tapers, or grows a second rim.
- Invented internals: pumps, tubes, bubbles, or seams appear because the model completes what it expects to see.
- Reflection mismatch: a floor reflection can repeat a wrong label or imply a second product.
- Highlights hiding geometry: a strong strip highlight can mask a changed shoulder or wall thickness.
- Material collapse: clear acrylic, coated glass, chrome, and satin metal all turn into the same generic gloss.
A generated reflection is a scene treatment, not evidence of finish or condition. Review the product first, then the reflection. For category-specific checks, see the glassware, perfume, and jewelry guides.
Scene 3: held in a hand, three models
A person holding the product adds scale, grip, occlusion, contact shadow, and anatomy to the same decision. We switched to the tall sage bottle and asked for a trail walker holding it at mid-torso, fingers on the side edges only, so the whole front stays readable.
Use the supplied fictional product source as immutable product-identity evidence. Create one square photorealistic ecommerce secondary lifestyle image of an adult trail walker at a quiet overlook in soft early-morning light, framed from shoulders to hips. The person holds the exact bottle upright at mid-torso with one natural hand: palm behind the bottle and fingers contacting only the outer side edges so the entire front, vertical NORTHLINE text, and 750 ML remain visible. Preserve exactly one tall sage cylindrical bottle, its silhouette and proportions, rounded shoulders, black cylindrical cap, single black circular loop on camera-right, bottom seam and base, sage color and matte finish, exact vertical NORTHLINE text, exact 750 ML text, and quantity one. Change only the person, environment, framing, and scene lighting. Add no second bottle, straw, handle, sticker, logo, claim, price, watermark, extra text, condensation, liquid, damage, or invented product feature.
| Model | Visible identity and text | Requested grip | Photoreal scene | Disposition |
|---|---|---|---|---|
| GPT Image 2 | Pass; exact geometry unproven | Closest match | Pass | Secondary review |
| Nano Banana 2 | Pass; exact geometry unproven | Plausible, tighter than asked | Pass | Secondary review |
| Qwen Image Edit Plus | Name only; geometry fails | Fail, cap grip | Fail, stylized | Reject |
Hands make fidelity harder. Fingers cover the edges you need to compare, a convincing person can make the product look bigger or smaller, a front-only source does not authorize the sides or back, and the setting can imply a use the product never claimed. When a believable grip cannot be built without inventing product geometry, change the composition: put the exact product on a nearby surface, or photograph the real interaction. If the job is a video ad rather than a still, use the real-product AI UGC workflow.
What merchants are actually trying to avoid
Merchant discussions repeatedly describe the same operational problem: a generator produces a beautiful lifestyle scene but shifts a label, changes a product color, morphs a bottle, or makes multiple channel images look like different SKUs. Those threads are qualitative intent evidence, not market sizing or conversion proof, but the concerns are concrete enough to design a useful test around.
- A Shopify merchant asked which edits are safe because improved photos can still make a product look unlike what the customer receives. Read the product-photo discussion.
- Another merchant trying to create lifestyle images with real products called out product consistency as the hard part. Read the lifestyle-image discussion.
The business task is not “generate a pretty bottle.” It is “create more usable placements without increasing customer uncertainty or review cost.”
A merchant workflow that measures accepted assets
- Keep one approved source of truth. Store the original packshot, artwork, rights record, dimensions, materials, variant data, and packaging version with the SKU.
- Declare invariants before generation. List the geometry, colors, marks, text, claims, quantity, and included items that cannot change. Separately list what may change: background, surface, crop, or lighting.
- Use one fixed test matrix. Hold source, prompt, ratio, size, and candidate count constant. If a model does not support a control such as seed, record that difference.
- Review blind at full resolution. Score product truth before scene quality. Compare overlays or flicker views for geometry, zoom into labels and hardware, and review the actual mobile PDP.
- Choose a disposition. Approve as secondary creative, composite the approved product into the generated scene, or reject. Keep the factual primary image untouched.
- Measure cost per accepted asset. Track attempts, credits or spend, review minutes, retouching time, accepted outputs, downstream conversion, returns, and “not as described” contacts where traffic is sufficient.
If you need a complete production sequence rather than a model comparison, use the supplier-photo to ecommerce image-set workflow. If the product passes fidelity review but still looks pasted onto the scene, the product-grounding prompt test separates contact, shadow, shared light, and floor physics from product redraw. If you are tempted to synthesize missing sides from that one source, first inspect the three-view product-angle consistency test: plausible new views are not evidence of hidden product geometry. If the approved recipe must cross several SKUs, the bulk product-photography workflow shows why one route, prompt, aspect, and seed still need separate product and set review. The broader product-photography model comparison maps failure modes across thirty categories; this same-SKU test supplies the direct reference benchmark that page deliberately did not claim.
If fidelity testing is only one step in a larger merchant job, use the AI for ecommerce workflow map to route catalog, PDP, acquisition, lifecycle, evidence, and wholesale work by its authority record and commercial finish line.
Publishing and platform handoff
Keep generated copy out of the bitmap whenever possible. Add approved claims, price, CTA, and legal copy deterministically after the image passes product review. Test the asset in the real crop and placement rather than approving an isolated file.
When the secondary gallery asset must communicate dimensions, capacity, or included items, use the fact-safe product listing infographic workflow to separate the approved SKU record from the generated layout draft.
When several exact products need to be evaluated against the same buyer requirements, use the AI product-comparison visual workflow to keep product sources, field definitions, prices, decision logic, and destinations independently reviewable.
When feedback identifies which product detail buyers struggle to inspect, the AI review-mining to creative-brief workflow shows how to code that evidence without letting anecdotes override this source-fidelity gate.
For Google Merchant Center, current guidance says images created with generative AI must preserve the IPTC DigitalSourceType value TrainedAlgorithmicMedia. That metadata requirement does not excuse a visually incorrect SKU or variant. Review Google's current AI-generated image requirements before export, because policies can change. The Merchant Center checks for Shopify stores continues from the local candidate into exact item IDs, public image URLs, delivered metadata, landing-page preselection, processing, and rollback.
Use the AI product-photo trust test to set release guardrails, the Shopify variant-image workflow when adjacent colors need separate sources and assignments, and the platform rules hub before syndication. Then run the controlled source through Masonry's AI product photography studio or automate the matrix with the Masonry CLI.
Bottom line
Across three scenes and 13 outputs, every model kept the visible label words and none kept the exact product. Nano Banana 2 was the most consistent starting point. Nano Banana Pro and Seedream 5 Pro made the most polished scenes at the cost of proportion drift. GPT Image 2 handled the person-holding-product brief best. Keep the approved packshot factual, treat generated scenes as candidates, composite the real product when pixels matter, and optimize for accepted assets rather than generated ones. The next production step is a controlled one-SKU ad creative matrix that keeps the offer and copy fixed while testing two genuinely different visual hypotheses.
