Welcome to our Project Page!

Dressing the Imagination - A Dataset for AI-Powered Translation of Text into Fashion Outfits and A Novel NeRA Adapter for Enhanced Feature Adaptation

Gayatri Deshmukh

Independent Researcher

Somsubhra De

IIT Madras

Chirag Sehgal

DTU

Jishu Sen Gupta

IIT BHU

Sparsh Mittal

IIT Roorkee

Accepted at WACV 2026 🌵

Dataset Code Pre-print

prompt details — *From imagination to illustration — your fashion ideas, perfectly sketched!*

generated sketch — *From imagination to illustration — your fashion ideas, perfectly sketched!*

✅ Key Contributions to #AI4Fashion

We present FLORA (Fashion Language Outfit Representation for Apparel Generation), the 1st curated dataset of fashion outfit sketches paired with rich, industry-grade textual descriptions. FLORA aims to advance AI-driven fashion design and assist designers and end-users in bringing creative ideas to life.
We propose NeRA (Nonlinear low-rank Expressive Representation Adapter), a novel parameter-efficient adapter based on Kolmogorov-Arnold Networks (KANs).

FLORA Dataset

A figure and the generated textual description pair using GPT-4o

Offers a sketch-centric benchmark with $4,330$ high-quality pairs of fashion outfit sketches and textual descriptions.
Each image is a clean, single-outfit sketch (no background clutter, no watermarks!).

Garment type & silhouette
Style and construction details
Fabrics, textures, and patterns
Pose and figure proportions
Accessories and overall aesthetic

Models fine-tuned on FLORA generate more accurate and stylistically nuanced fashion images from text inputs.

dataset styles — Diversity of `FLORA`, selected classes from each of the 9 categories

NeRA: Nonlinear Low-Rank Expressive Representation Adapter

Standard PEFT methods (LoRA, LoKR, DoRA, LoHA) use mainly linear or shallow MLP-based adapters, which are limited in capturing complex, nonlinear relationships in fine-grained domains like fashion.
NeRA is a new adapter module that:
- Uses KAN-style learnable spline or RBF basis functions instead of fixed activations.
- Plugs into existing diffusion/transformer layers similarly to LoRA, with frozen backbone weights.
- Provides more expressive, flexible feature transformations.

Qualitative Results

Qualitative comparison of zero-shot, MLP-based adapters and the proposed NeRA. *Colored arrows mark whether the prompt elements written in that color were correctly generated.*

t-SNE visualization of the FLUX feature space

NeRA yields clearer, better-separated clusters than other adapters, indicating stronger semantic organization and disentanglement of fashion concepts.

NeRA shows higher CLIPSIM and lower FID scores, indicating better text-image alignment and visual quality.

BibTeX

Please cite our paper if you find it useful in your work.

@inproceedings{Deshmukh_2026_WACV,
  author    = {Deshmukh, Gayatri and De, Somsubhra and Sehgal, Chirag and Gupta, Jishu Sen and Mittal, Sparsh},
  title     = {Dressing the Imagination: A Dataset for AI-Powered Translation of Text into Fashion Outfits and A Novel NeRA Adapter for Enhanced Feature Adaptation},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  month     = {March},
  year      = {2026},
}