Page 504 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Model Training Pose-Agnostic Garment Transfer: Uses Stable Diffusion to warp garments
and Fine-Tun- realistically across body types.
ing Contrastive Language-Image Pre-training (CLIP)-Guided Fine-Tuning:
Preserves traditional textile patterns (Banarasi, Kanjeevaram, etc.) during
virtual try-ons.
Trust Mechanism: AI models incorporate real-world garment physics (wrin-
kles, elasticity) to build user confidence in how a garment will fit and feel. Fit
scoring and visual comparison tools increase alignment with actual prod-
ucts.
StyleGAN3 + ViT (Vision Transformer): Creates high-fidelity textures for
Indian fabrics while retaining weave details.
Diffusion Priors for Fabric Simulation: Models wrinkles, elasticity, and phys-
ics-aware garment movement.
Text-to-Design (Imagen, Parti): Enables on-demand, AI-generated garments,
blending modern cuts with traditional motifs.
Transformer-based Forecasting Models: Predict consumer preferences
based on trends, demographics, regional festivals, and past behavior.

Testbeds or Looklet dressing room demo [2]
Pilot Deploy- Virtual makeup demo [3]
ments
Outfit Anyone: Ultra-high quality virtual try-on for Any Clothing and Any
Person [4]
Outfit Anyone in the Wild: Get rid of Annoying Restrictions for Virtual Try-on
Task [5]

Code reposi- IDM-VTON: Improving Diffusion Models for Authentic Virtual Try-on in the
tories Wild dataset [6]
CatVTON: Concatenation Virtual Tryon dataset [7]
Multimodal-Conditioned Latent Diffusion Models for Fashion Image Editing
(Garment Design) [8]
Generative Pretrained Hierarchical Transformer for Time Series Forecasting
(Forecasting Modeling) [9]

2� Use Case Description

2�1 Description

The Indian fashion industry blends traditional craftsmanship with modern designs, serving
diverse consumer preferences across regions. However, both designers and online shoppers
struggle to visualize how outfits, ranging from sarees and sherwanis to contemporary dresses
and suits, will appear on real bodies, resulting in high return rates. Additionally, a lack of
demand forecasting leads to overproduction and inventory waste. This creates significant
environmental and financial costs for brands and artisans alike.

To address these challenges, we propose an AI-driven virtual try-on, garment generation, and
demand forecasting system. Using cutting-edge generative AI models like Stable Diffusion
and Transformer-based networks, users can digitally try on garments with realistic simulation of
fabric drape, fit, and style. AI-generated garment designs based on user inputs, such as photos,
text prompts, or sketches, support both consumer personalization and designer creativity.

468