IMAGGarment
IMAGGarment is a fine-grained garment generation system designed for controllable fashion design and digital apparel applications. Accepted by TVCG 2026, it enables precise and customizable synthesis of clothing by combining user inputs such as sketches, color references, and logo specifications. The software addresses the challenge of multi-conditional controllability through a two-stage training strategy that separately models global appearance and local details. The first stage utilizes a global appearance model with a mixed attention module and color adapter to encode silhouettes and colors jointly. The second stage employs a local enhancement model with an adaptive appearance-aware module to integrate user-defined logos and spatial constraints, ensuring accurate placement and visual consistency. This architecture supports end-to-end inference for unified generation with high realism. IMAGGarment is compatible with Python 3.8 and PyTorch 2.0, requiring CUDA 11.8 or higher. It relies on pre-trained stable