moda

name: moda description: > Skill for high-fidelity garment reproduction in AI image generation. Should be used when the goal is to accurately describe clothing items, fabrics, textures, draping, layered constructions (such as lace over lining), transparency, fit, and silhouette for maximum visual accuracy. Provides a structured vocabulary for textile properties, construction details, and light interaction that AI models respond to with higher precision. Relevant for: fashion photography, e-commerce product shots, lookbooks, catalog images, virtual try-on, any scenario involving clothing or textile reproduction.

Fashion Garment Precision

Core Philosophy

Clothing is not a surface — it is a system of layers, structures, and behaviors.

AI models tend to treat garments as flat textures painted onto a body. Real clothing is a 3D construction with internal structure, multiple fabric layers, and physics-driven draping. The more precisely you describe how the fabric behaves, not just how it looks, the more accurate the output.

The goal of this skill is to give the AI enough structural understanding of a garment to reproduce it as a real piece of clothing — not as a painted texture.

🚨 Two Modes: Reference Image vs. Pure Creation

This is the most critical decision before writing any garment description.

Mode 1 — Reference Image Available (EDIT mode)

When a reference photo exists, the image is the authority for the garment. Text description works AGAINST fidelity if it is too detailed.

Why: AI image models weight text and image inputs simultaneously. If text describes the garment in high detail, the model re-generates the garment from text and treats the reference image as secondary. The result: a garment that matches the words, not the photo.

Rule: Use a Fidelity Lock FIRST, then a MAXIMUM of 2 reinforcement sentences. The reinforcement must follow this distinction:

🚨 REGRA CRÍTICA PARA TEXTURAS E ESTAMPAS: Nunca descreva como o padrão ou textura se parece visualmente (ex: "listrado", "zigzag", "xadrez"). O modelo nano é excelente em puxar a imagem original, mas se você ditar o padrão em texto, ele sobrescreve a imagem por uma versão genérica da IA. Para o reforço: descreva apenas a natureza física da peça (ex: "heavy winter knitwear", "fine modal tricot", "cotton poplin").

Fidelity Lock template (Mode 1 — goes FIRST in the prompt):
REFERENCE IMAGE IS THE AUTHORITY FOR THE GARMENT. Reproduce the clothing 
exactly as shown in the attached reference photo. Follow the texture and 
stitch pattern exactly as shown — do not invent or describe the pattern, 
copy it from the reference image. Only the model, pose, and background are new.

Reinforcement only (max 2 sentences): [fiber/material] + [construction type] + [precise color].

Mode 2 — No Reference Image (CREATE mode)

When no reference photo exists, use the full 3-Dimension framework below. This is the correct time for detailed garment description.

The 3 Dimensions of Garment Description (Mode 2 — no reference image)

Every garment must be described across 3 independent dimensions:

1. Material (What it's made of)

The fiber, weave, and finish of the fabric.

2. Construction (How it's built)

The seams, closures, layers, and structural elements.

3. Behavior (How it moves)

The draping, fall, cling, and reaction to the body and gravity.

⚠️ Most failed garment prompts only describe Dimension 1. Dimensions 2 and 3 are what separate a "fabric texture" from a "real piece of clothing."

Dimension 1: Material — The Fabric Vocabulary

Base Fabric Types and Their Visual Signatures

Use these precise terms instead of generic ones. Each carries specific visual characteristics that AI models interpret differently.

Woven Fabrics (structured, less stretch)

Knit Fabrics (stretch, drape)

Technical / Synthetic Fabrics

Surface Treatments and Finishes

These modifiers change how the base fabric behaves visually:

Dimension 2: Construction — How It's Built

Structural Elements Vocabulary

Describing construction tells the AI about the garment's 3D architecture.

Seams and Joining

Closures

Necklines, Collars, and Sleeves

Instead of vague terms, use precise construction language:

Dimension 3: Behavior — How It Moves

This is the most neglected and most impactful dimension. Fabric behavior is what makes a garment look "worn" vs "rendered."

Draping Vocabulary

Fold and Crease Behavior

Body Interaction

Complex Constructions: Multi-Layer Garments

This is where most AI prompts fail. When a garment has multiple layers, you must describe each layer independently and then describe their relationship.

The Layer Stack Pattern

Layer 1 (innermost/base): [fabric] + [opacity] + [fit]
Layer 2 (overlay): [fabric] + [transparency level] + [behavior relative to Layer 1]
Relationship: [how they interact visually]

Common Multi-Layer Cases

Lace Over Lining

Base layer: opaque [color] stretch jersey lining that follows the body contour.
Overlay: [color] floral lace with [scalloped/straight] edges, sitting slightly 
away from the lining. The lace pattern is visible against the contrasting 
lining beneath. Where the lace overlay meets skin (neckline, sleeves), 
the skin is partially visible through the lace openwork.

Key instruction: Explicitly describe what is visible through the lace — skin or lining — at each zone of the garment.

Matching Sets / Two-Piece Layers (e.g., Coat + Scarf)

When a garment consists of two separate but matching pieces (same material/pattern), you must force the physical separation so the AI doesn't merge them into one blob.

Base piece: [color] [fabric] [garment type, e.g., cardigan/coat] worn open or wrapped.
Accessory piece: matching [color] [fabric] [garment type, e.g., scarf] draped 
AROUND THE NECK, physically separate from the coat. The scarf crosses over the 
chest forming distinct layers on top of the base piece.

Key instruction: Use words that denote physical layering and separation: "draped around," "layered over," "physically separate," "distinct piece."

Sheer Over Opaque

Base layer: fitted [color] camisole/slip acting as modesty lining.
Overlay: semi-transparent [fabric] in [color] that reveals the outline 
and color of the layer beneath. Neither fully transparent nor fully opaque — 
a veiled, diffused visibility of the base layer.

Mesh Panel Construction

Main body: opaque [fabric] in [color].
Mesh panels: [location — shoulders/sides/yoke] in tonal or contrast mesh, 
creating a peek of skin within a structured frame. The transition between 
opaque and mesh has [clean seam / raw edge / binding].

Lined Blazer / Jacket (Partially Visible)

Outer shell: structured [fabric] in [color] with [lapel type].
Inner lining: visible at inner facing, cuffs (when rolled), and hem edge. 
Lining in [contrasting color/pattern] — partially visible as a flash of 
detail when the jacket moves or is worn open.

Layered Tulle / Organza (Volume Building)

Multiple layers of [color] tulle stacked to build volume and opacity 
gradient. Innermost layer is most opaque; outer layers become increasingly 
sheer. Each layer's edge is visible as a softer, ghosted duplicate. 
Creating a cloud-like, dimensional volume that gets more transparent 
toward the edges.

Embroidery / Appliqué on Base

Base fabric: [type] in [color], smooth and even.
Surface detail: [embroidered/appliquéd] [motif description] in [thread/fabric type]. 
The embroidery creates raised texture — catching light on the top of the 
stitches while casting micro-shadows around the edges. The thread has 
[matte/sheen/metallic] finish.

Light Interaction by Fabric Type

How the AI renders light on fabric is critical. Use these cues:

The Light Response Matrix

💡 Regra do par câmera × textura: Para detalhe de malha/trama, sempre especifique "macro lens" no prompt. Para fashion shot com textura visível, use "85mm f/1.8" — essa lente comprime a profundidade e destaca a superfície do tecido enquanto desfoca o fundo com bokeh natural.

Common AI Failures with Clothing & How to Fix

The Fidelity Lock System

When reproducing a garment from a reference image, use a Fidelity Lock — a structured instruction block that tells the AI exactly what to preserve and what to change.

Full Fidelity Lock Template

GARMENT FIDELITY LOCK:
- Fabric: [exact type] with [texture description]
- Color: [precise color, not generic — e.g., "dusty sage" not "green"]
- Pattern: [exact pattern description following fabric contours]
- Construction: [key details — neckline, sleeves, closure, hem]
- Fit: [how it relates to the body — loose/fitted/oversized]
- Draping: [how the fabric falls and folds]
- Light response: [matte/sheen/metallic — how it reacts to light]

CHANGE FREELY:
- Model appearance, pose, expression
- Background/setting
- Lighting direction and mood
- Camera angle

Minimal Fidelity Lock (Quick Version)

Keep the clothing exactly as shown: [fabric type], [color], [key construction detail], 
[draping behavior]. Pattern follows fabric folds naturally. 
Everything else (model, background, pose) is completely new.

Garment Description Checklist

Before finalizing any prompt involving clothing, verify:

• Fabric type specified? (not just "dress" but "cotton poplin dress")
• Behavior described? (drapes, clings, holds structure, billows)
• Construction details present? (at least neckline + closure + hem)
• Light interaction defined? (matte, sheen, transparent)
• Multi-layer garments: each layer described independently?
• Transparency zones explicit? (where skin vs lining is visible)
• Pattern behavior specified? (follows folds or remains flat)
• Fit described? (oversized, fitted, relaxed, structured)
• Wear state indicated? (brand new, lived-in, distressed)

Textile Surface Vocabulary (Mode 1 — Reinforcement tokens for textured garments)

💡 Thinking Mode: Para padrões de tricô complexos (Aran, Fair Isle, openwork), o Nano Banana Pro planeja as sombras e a profundidade do ponto antes de renderizar os pixels. Isso evita o padrão "colado" onde a trama parece uma textura plana sobre o corpo. Para ativar: mencione "complex textile construction" ou "structured knitwear physics" no prompt — o modelo prioriza o raciocínio estrutural.

When to use this section: Quando a referência visual existe mas o modelo gera a textura errada — loops inflados, ponto distorcido, peso visual incorreto. Use os tokens desta seção como reforço cirúrgico no prompt (máximo 1–2 frases), NUNCA descrevendo o padrão visual.

A Regra Anti-Inflação

O Nano Banana tende a "inflaro ponto" de peças texturizadas — loops ficam gordos, redondos e 3D quando a peça real é plana e regular. Isso acontece porque o modelo interpreta "knitwear" como estrutura volumosa por padrão.

Solução: usar tokens que descrevem a SUPERFÍCIE e o PESO da construção, não a aparência do ponto.

Crochê

⚠️ Anti-trap específico do crochê: Nunca use loops, bumps, rounds, bobbles, 3D, dimensional, puffy, raised rounds — esses termos ativam a síntese volumosa do Nano.

Vocabulary avançado (Precision Knitwear):

Tricô

⚠️ Anti-trap específico do tricô: Evite cable a menos que a peça seja realmente de tranças — o Nano gera tranças agressivas mesmo com referência. Para tricô grosso sem padrão visual, use thick flat-knit ou heavy solid-color knit.

Vocabulary avançado por tipo de ponto:

Yarn weight como controle de escala do ponto:

Tecido Plano Texturizado

Tokens de Intensidade de Textura

Quando você precisar calibrar o QUANTO de textura aparece na superfície:

Textural Anchor — Técnica Multi-Referência (recomendada para textura crítica)

O Nano Banana 2 suporta até 14 imagens de referência na mesma sessão. Para transferência fiel de textura de tecido, use a técnica Textural Anchor:

Setup de referências:

💡 Na app Studio Local: o pool de referência da sessão alimenta automaticamente o contexto. Imagens geradas que o usuário aprova e adiciona ao pool se tornam autoridades adicionais de consistência — funciona como LoRA dinâmico por sessão.

Limite prático: Enviar no máximo 8 imagens ao agente de contexto (thumbnails). Para geração final, usar até 14 refs de alta qualidade.

Template de prompt com Textural Anchor:

IMAGE 1 is the composition to preserve. IMAGE 2 is the exclusive texture 
authority. Apply the exact fabric surface from IMAGE 2 to the garment in 
IMAGE 1: replicate stitch structure, loop size, yarn plies, row regularity, 
and surface flatness exactly as shown. Apply only to the garment surface. 
Keep model, pose, composition, lighting, and scene from IMAGE 1 unchanged.

Para geração do zero (sem imagem de cena):

REFERENCE IMAGE IS THE TEXTURE AUTHORITY. Reproduce the exact fabric 
construction as shown: [flat uniform crochet / brioche / Aran cable — 
descrever a vibe sem o padrão visual]. Generate [modelo + cena + pose].

⚠️ Limitação do Gemini App (chat): O app pode ficar "preso" na primeira imagem durante iterações longas. Solução: reiniciar o chat com todas as referências novamente é mais eficiente que insistir em prompts correção. Na app Studio Local esse problema é mitigado pelo gerenciamento explícito do pool.

Thinking Level por Tipo de Textura

Integração com Nano Banana 2: o Thinking Mode executa Plan → Evaluate → Improve antes de renderizar pixels. Para texturas complexas, isso reduz drasticamente a inflação de ponto e a geração de textura "colada".

⚠️ Antes de usar Grain Injection (skill realismo, Lever 8) para corrigir AI Softness em texturas: tente subir o thinking de MINIMAL para MEDIUM. Frequentemente resolve sem pós-processamento.

Integration with Other Skills

This skill works as a precision layer for garment description. Combine with:

• Skill realismo: Apply realism levers to make the fashion photo feel authentic (natural lighting, organic composition, imperfect focus)
• Workflow /edit-image: Feed garment descriptions from this skill into the base prompt structure
• Workflow /puxar-foto-shopee: After downloading reference images, use this vocabulary to describe garments with maximum precision

Recommended flow:

1. Analyze the garment (visual reference or mental image)
2. Describe using the 3 Dimensions (Material → Construction → Behavior)
3. For complex garments, apply the Layer Stack Pattern
4. Add a Fidelity Lock if reproducing from reference
5. Combine with realismo skill for authentic final output