Bridging Structure and Emotion: A Generative Framework for Accessible and Expressive Knit Design

Knit pattern design is challenging, requiring both technical skill and creative vision. We introduce an accessible generative framework using Stable Diffusion fine-tuned with LoRA and an interactive web interface. Our system utilizes a dual-label dataset (over 3000 images) annotated with both technical attributes (stitch type, complexity) and semantic/emotional descriptors (e.g., cozy, elegant) to ensure outputs are structurally coherent and stylistically diverse. The LoRA-tuned model exhibits strong domain specialization, producing clearer stitch definition and fewer non-knitting artifacts than the baseline. We develop an interactive platform featuring an emotion-technical balance slider. This framework positions generative AI as a creative partner that lowers technical barriers and expands the expressive possibilities of textile design.