Fabric lies under zoom
Zoom shows the truth: knit direction, slub, sheen. Models 'improve' texture into smooth synthetics that sell a different product than the one shipped — the single biggest driver of 'not as described' returns.
AI product photography for fashion: color-true fabric, one silhouette, every colorway
Fashion returns are dominated by two words: color-wise. A catalog needs the same garment rendered honestly across every colorway — knit grain, drape, print placement and hardware identical — while flat-lay-to-worn variants and marketplace compliance (Shopee now requires a separate size-chart image in many apparel/footwear sub-categories) eat studio time per SKU. Berrys generates from your actual garment photos — image-to-image, not prompt-to-imagination — so the fabric is your fabric and only the scene changes.
Reviewed October 2026 · Prepared by Berrys · Technical specifications & Canvas node workflows
Fashion marketplace image rules
| Parameter / Slot | Marketplace & Production Requirement |
|---|---|
| Shopee cover | Solid background, product ≥60% frame — models ARE allowed on apparel/beauty/sports covers; separate size-chart image required in footwear/apparel sub-categories (from mid-2025) |
| Shopee gallery | Up to 9 images, different angle each; lifestyle and close-ups allowed after the cover |
| TikTok Shop main | Front view of the actual product, pure white bg, ≥600×600 (800×800 some sites), no repeated angles |
| Amazon main | Pure RGB 255 white · product ≥85% · one unit · 1,000px+ longest side for zoom, 2,000px+ recommended |
| Return drivers | Color difference, missing scale cues, print/embellishment not matching — all three are image problems before they're product problems |
Where generic generation fails fashion
Drape is physics, not style
The same fabric behaves differently on different bodies and poses; free generation rewrites drape entirely. Flat-lay and dress-form references anchor the silhouette while scenes vary around it.
Print placement continuity
A floral whose pattern jumps between front and side renders kills trust. SKU referencing keeps the print mapping stable across the angle set.
Hardware & small logos
Zipper pulls, buckles, brand tabs — the small metal-and-text bits models smear first. Preserved-region handling keeps them true.
Colorway math
'Sage' rendered three shades off across five images is five reasons to return. One color-locked graph per colorway means the only variable is your reference dye card.
How Berrys runs an apparel catalog
Garment-anchored generation
Every scene branches from your real flat-lay/dress-form photo; the piece is never re-imagined, only re-staged.
Lighting-locked color truth
One saved lighting recipe across the catalog, so colorway differences are real, not session drift. Proof against physical swatches before publishing.
The whole family in one run
List and Router nodes fan a style's color/size grid through the same graph — consistent output is the default state, not heroic effort.
Production workflows in Berrys Canvas
Flat-lay → full PDP set
1. File Node: Upload high-resolution flat-lay or dress-form reference under 5600K light.
2. Brand System Node: Lock fabric knit texture, drape physics, and stitch vectors.
3. Image Node (Nano Banana 2.1): Render 2048px front hero on pure white (1:1 square and 4:5 vertical).
4. Fan-out: Duplicate to back view, 45° side, and macro fabric weave detail.
Multi-colorway dye card fan-out
1. File Node: Connect approved garment silhouette.
2. Brand System Node: Map Pantone swatches for all 8 colorways.
3. Image Node (Nano Banana 2.1): Render all colorways under identical studio lighting, proofed against physical swatches.
On-model scene branch
1. File Node: Connect locked garment master.
2. Brand System Node: Select model persona and studio/editorial scene grammar.
3. Image Node (Flux 3): Render worn compositions preserving zipper pulls, buttons, and print scale.
Berrys Plugins · Model Context Protocol (MCP)
Create in chat. Keep it on your canvas.
Already working in an AI agent? Connect Berrys to Claude, Cursor, ChatGPT, Codex, or Hermes. Your agent plans the fashion & apparel suite and runs the shots on your Berrys canvas, with saved Brand Memory and credit approval in chat.
Connector URL:
https://app.berrys.io/api/mcpRemote MCPDrop-in Agent Brief (Fashion & apparel)
Using Berrys, create a complete fashion PDP suite for my apparel reference in ./assets: 1. File node: Load flat-lay or ghost-mannequin garment photo. 2. Brand System node: Set neutral studio lighting. Lock fabric knit grain, seam alignment, and exact Pantone color. 3. Image node (Nano Banana 2.1): Render 2048px front packshot on solid white (1:1 and 4:5 crops). 4. Macro pass: Render 2048px close-up on fabric weave, button stitching, and zipper hardware. 5. Secondary angles: Render clean back view and 45° angle. Estimate credit cost and request approval before generating.
Compliance & production boundaries
Fashion's rule mirrors every marketplace: show the item the buyer receives. Generated scenes must never invent fabric behavior, hardware, or color the physical product can't deliver. Where a platform calls for the actual product image (main slots), the Berrys workflow starts from an actual photograph of the actual garment — the canvas scales truth, it doesn't replace it.
When NOT to use AI generation for this
- Fit claims: a render cannot promise sizing a physical sample hasn't confirmed
- Sheer or texture-sensitive fabrics where screen-only judgment misleads — macro-reference first
- Any worn scene where the real product's drape would differ from the render
Frequently asked questions
Can AI keep fabric texture realistic?
Only image-to-image from your actual garment photo can — which is how Berrys works. The knit, drape and print come from your reference; only lighting and context are generated.
Does it support on-model imagery?
Yes, with a discipline: worn scenes branch from the locked garment so cloth stays true. For standardized posing across a collection, use our free [AI Model Pose Reference Library](/for/poses) (100 commercial poses with ControlNet depth maps and 500 camera angles) to lock posture and camera perspective. For fit-critical categories treat model renders as supplementary and keep at least one flat reference shot per SKU.
How do you keep colorways consistent?
One lighting recipe across the family plus per-color SKU nodes, then proof against physical swatches. Color truth is verified, not assumed — the same standard a studio owes you.
What about Shopee/Amazon rules for fashion?
Shopee: square images, cover ≥60% frame, models allowed on apparel covers, separate size-chart image required in footwear/apparel sub-categories. Amazon: pure-white main, ≥85% fill, 1,000px+ zoom. The canvas exports each platform's crops from one master set.
Related tool comparisons & guides
In-Depth Guides & Research from Berrys Journal
Create on-brand assets without booking a studio
Start free on Berrys Canvas — 50 credits, no card required — or send Berrys Studio a brief and get finished campaign visuals with agency art direction.