Handmade soap has two kinds of variation that product photography must not confuse. Each cut face can be unique, but the listing still promises a defined formula, batch, size, weight, color range, decoration, packaging, and condition. A beautiful generated swirl can therefore look category-correct while representing no bar a customer can receive.
This test sent one fictional cream-and-purple soap-stack 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 swirl, saturation, cut form, botanical-like decoration, bar count, props, crop, and light. No real soap, batch record, formula, ingredient list, process, dimensions, net weight, color standard, variation rule, stamp, packaging, cure/condition record, safety test, claims file, or product data was supplied.
Evidence boundary: this is one text prompt and one displayed output per model. The observations describe visible fictional soap-like bars—not exact formula, ingredients, process, batch, scent, net weight, dimensions, color accuracy, cure, safety, performance, model reliability, current price/value, shoot replacement, or preservation of a real soap SKU.
Quick answer
- Cleanest displayed cream-and-purple swirl: Seedream 4.5 in this run.
- Fullest displayed rustic prop scene: Nano Banana 2 in this run.
- Most saturated displayed purple treatment: GPT Image 2 in this run.
- Clearest displayed horizontal botanical-like band: FLUX.2 Pro in this run.
- What all four establish: each route made one plausible fictional artisan-soap concept.
- What none establish: an exact bar or batch, ingredient identity, formulation, process, dimensions, weight, cure, condition, safety, performance, claims, or repeated reliability.
One output per route does not establish a universal winner or that soap is a reliable category. Organic-looking swirls and plant-like fragments are visual cues, not evidence of cold-process manufacture, lavender, essential oil, natural ingredients, or a saleable batch.
First define what the customer receives
| Offer type | Fidelity target | Required evidence |
|---|---|---|
| Exact or one-of-one bar | The photographed individual bar with its current cut face, top, decoration, stamp, condition, dimensions, and weight. | Current multi-angle item photography, measurements, weight, formula/batch link, condition, packaging, fulfillment link, and exact-item disclosure. |
| Batch-representative bar | A representative unit within one disclosed handmade or natural-variation range. | Multiple batch samples, minimum/typical/range for dimensions and weight, allowed swirl/color/cut/top/embed/stamp variation, cosmetic tolerance, packaging, batch/lot mapping, and representative-photo disclosure. |
| Fixed molded or packaged SKU | One repeatable molded/stamped bar or packaged set with defined count, label, and configuration. | Approved sample, mold/stamp, dimensions, weight, color, surface, packaging artwork/copy, set count, accessories, tolerances, and current SKU record. |
Do not use one flawless fictional bar to imply that every naturally variable unit has the same face, top, botanical placement, dimensions, or color split.
The controlled concept brief
The original article did not publish the exact prompt. A bounded version matching the displayed subject is:
Fictional brand-free stack of three hand-cut soap-like bars with cream and muted purple marbling, sparse unidentified dried botanical-like fragments on top, matte irregular cut faces, simple warm wood surface, soft directional light, square artisan-product photograph. No ingredient name, scent claim, process claim, stamp, logo, label, package, person, lather, water, price, badge, or extra product.
This prompt can test art direction. It cannot preserve a real soap because the models invent the bar count, dimensions, cut, swirl, layer, color, opacity, surface, top, plant-like fragments, cracks, crumbs, moisture-facing cues, stamp, stack contact, hidden sides, packaging, and condition.
Four first-hand outputs
Seedream 4.5 produced the cleanest displayed cream-and-purple marbling and a close view of matte-looking cut faces, rough edges, and scattered plant-like fragments. Those are useful styling observations. They do not identify lavender, prove the fragments are embedded, establish a cold-process method, or match a real bar's swirl, dimensions, weight, condition, formula, or batch.
Nano Banana 2 made the fullest rustic-maker scene, adding containers, loose plant-like material, linen, and densely decorated soap-like bars. The presentation can imply ingredient identity and included props even though neither was sourced. An ecommerce asset needs a clear product-only view and an explicit check that no jar, textile, scoop, botanical, or extra bar is presented as included.
GPT Image 2 used the most saturated displayed purple treatment across its marbling and decoration. That supports a relative on-page color observation only. A real SKU needs neutral-light sources, an approved color range, batch-aware comparison, and device-aware review; the image cannot identify a colorant, fragrance, ingredient, or formulation.
FLUX.2 Pro emphasized a horizontal band of plant-like fragments between cream and purple layers, with a warm rustic presentation. It also displays four bars rather than the concept brief's three, changing the visible quantity. The band differs from a freeform in-the-bar swirl and would represent a different product design if the sold bar did not have that layer. It cannot establish ingredient identity, safe use, adhesion, shedding, cut quality, formula, or batch.
What this comparison supports
| Question | What the four images show | What remains untested |
|---|---|---|
| Can the routes make an artisan-soap concept? | Yes, once each for this fictional brief. | Reliability across seeds, recipes, batches, cuts, references, packages, views, prompts, and current routes. |
| Do visible swirl and styling treatments differ? | Yes; marbling, layers, saturation, fragments, bar count, cut, props, crop, and light vary. | Exact bar/batch match, calibrated color, dimensions, weight, formula, process, cure, condition, and production tolerance. |
| Are the plant-like fragments lavender or an ingredient? | Not established; they are visual fragments in fictional images. | Botanical identity, inclusion in formula, fragrance, allergen/sensitization data, supplier, safety, labeling, and allowed decoration. |
| Is a real soap SKU preserved? | No evidence; no product source was supplied. | Exact or representative cut, swirl, top, stamp, weight, dimensions, packaging, set count, lot, inventory condition, and fulfillment mapping. |
| Are quality, safety, or performance claims established? | No. | Process, cure, pH, microbial quality, lather, cleansing, moisturizing/skin effects, suitability, testing, certification, shelf life, and warranty. |
The old article cited an unsourced 78% survey, called soap reliable, claimed ideal color-true lighting, said AI beats makers' cameras, named winners, and promised reference-based exactness. The defensible conclusion is narrower: four routes made attractive fictional soap concepts once, and their displayed swirl, saturation, layer, cut, fragment, prop, and lighting treatments differ.
Soap fidelity checklist
Build the rejection sheet from current exact-item or representative-batch photography, formula and batch records, dimensions and weights, approved variation/condition standards, color and surface sources, stamp/mold records, packaging artwork and copy, ingredient/safety/claims files, fulfillment mapping, and current product data:
- Offer and batch: exact bar, batch representative, or fixed packaged SKU; batch/lot, pour/cut date, inventory state, representative range, substitutions, season/aging variation, fulfillment link, and disclosure.
- Bar geometry: unit/set count, length, width, height, shape, mold, cut angle, corners, bevel, top profile, layers, inclusions, cavities, holes, cracks, chips, crumbs, warp, shrinkage, and scale.
- Face and decoration: exact or allowed swirl/marble/layer pattern, color split, opacity, stamp/logo, emboss/deboss, top texture, piping, dusting, seeds/grains/herbs/flowers/spices-facing fragments, placement, density, adhesion, shedding, and cosmetic tolerance.
- Surface and condition: matte/gloss/translucent-facing appearance, smooth/rough/exfoliating-facing texture, soda-ash-facing film, glycerin-dew-facing droplets, discoloration, spots, oxidation-facing change, fading, moisture, bloom/residue, damage, cure/age state, wrapper contact, and lot variation.
- Color and scent-facing presentation: neutral approved color range, lot/age/light variation, white balance, saturation, background cast, screen caveat, and no visual inference of fragrance, essential oil, strength, permanence, or ingredient identity.
- Packaging and set: naked/wrapped/boxed, paper/band/film, window, seal, label copy/art, ingredient panel, net quantity, barcode/lot/date, claims/badges, unit count, dish/bag/accessories, props, packaging condition, and sold configuration.
- Measurements and fulfillment: per-unit and set dimensions, labeled/actual net weight and tolerance, variation over time, scale, quantity, inventory/lot, packaging, shipping state, heat/moisture protection, substitutions, thumbnail mapping, and returns record.
- Claims and rights: ingredient/INCI-facing identity, fragrance/essential oil, allergens, process, cure, pH, natural/organic, vegan/cruelty-free, palm-free, recycled/sustainable/certified, handmade/origin, lather, cleansing, moisturizing/therapeutic/skin effects, suitability, safety, shelf life, delivery, and warranty remain in approved records.
If the scene looks artisanal but the bar count, cut, swirl, layer, color, decoration, stamp, dimensions, weight, batch range, packaging, included items, or claims differ, it is the wrong soap asset.
A batch-aware soap workflow
1. Capture authoritative sources
For an exact bar, photograph current front, back, sides, top, base, cut face, decoration, stamp, condition, packaging, and scale; record dimensions and weight. For a representative batch, photograph several units and document the minimum, typical, and range for cut, swirl/layer, color, top, decoration, stamp, dimensions, weight, condition, packaging, and cosmetic variation.
2. Build the scene separately
Generate or photograph an empty wood, stone, tile, linen, sink, shower, shelf, or editorial plate at the final crop. Composite approved soap photography, then build contact, crumbs, shadow, reflection, and occlusion deliberately. Use only approved props; keep ingredient suggestions and non-included items out of the factual hero.
3. Test a constrained reference edit
masonry image "Change only the environment to a simple warm wood surface with soft directional light. Keep the supplied soap unchanged: exact unit count, dimensions, cut, every visible swirl and layer, colors, opacity, top, botanical-facing fragments, stamp, edges, chips, cracks, surface, condition, wrapper/label, packaging, and sold configuration. Add one natural contact shadow. Add no ingredient, flower, jar, dish, towel, lather, water, person, text, badge, or extra bar." \ --model gemini-3.1-flash-image-preview \ --ref ./approved-soap-batch.png \ --aspect 1:1 \ --output soap-scene-candidate.png
The instruction is not a bar, batch, formula, ingredient, color, weight, packaging, or SKU lock. Compare the candidate with current exact-item or batch sources, measurements, weights, variation rules, packaging, formula, safety/claims files, and fulfillment data. Composite approved product photography when exactness matters.
For a fictional concept:
masonry image "Fictional brand-free stack of three hand-cut soap-like bars with cream and muted purple marbling, sparse unidentified dried botanical-like fragments on top, matte irregular cut faces, warm wood surface, soft directional light, no ingredient name, scent claim, process claim, stamp, logo, label, package, person, lather, water, price, badge, or extra product" \ --model seedream-4-5 \ --aspect 1:1 \ --output soap-concept.png
Soap acceptance sheet
| Area | Pass condition |
|---|---|
| Source match | Candidate is compared with current exact-item or representative-batch front, back, sides, top, base, cut faces, decoration, stamp, condition, scale, measurements, weights, variation rules, packaging, formula, claims, fulfillment, and product data. |
| Product and batch | Exact or allowed count, dimensions, weight, cut, shape, top, swirl/layer, color, opacity, fragments, stamp, surface, condition, cosmetic tolerance, batch/lot, substitutions, and representative disclosure match. |
| Packaging and fulfillment | Wrapper/box/band/film, label art/copy, ingredients, net quantity, barcode/lot/date, claims/badges, set count, accessories, packaging condition, inventory mapping, shipping state, and sold configuration match. |
| Color and inspectability | Neutral-light product view, approved batch color range, white balance, saturation, scale, crop, contact, shadow, surface detail, no obstruction, no misleading ingredient prop, and no implied included item are approved. |
| Claims and rights | No ingredient, botanical, fragrance, allergen, process, cure, pH, natural/organic, vegan/cruelty-free, palm-free, sustainable/certified, handmade/origin, performance, therapeutic/skin, safety, shelf-life, delivery, or warranty claim is inferred from appearance. |
Bottom line
These four outputs are useful as a swirl, layer, color, cut, decoration, prop, and lighting art-direction comparison. They are not evidence that soap is a reliable AI category, the visual fragments are lavender, a reference preserves a batch, a fictional bar is ready to list, AI beats a maker's camera, or one route is always the best value.
Define whether the offer is an exact bar, representative batch, or fixed packaged SKU. Use fictional concepts to choose a scene; use current product photography, formula and batch records, measurements and weights, variation/condition standards, packaging, claims and safety files, fulfillment mapping, 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.
