Image editing with diffusion models in the domain of fashion

Keserič, Luka (author), Peer, Peter (mentor), Štruc, Vitomir (comentor)

Image editing, specifically appearance or style transfer, is an area of computer vision that has seen significant growth due to recent advancements in image diffusion models. There are many different approaches to appearance transfer with fine-tuning, external network adapters and tuning-free methods. In this work, we focus on the latter, where we leverage the existing Stable Diffusion network with some modifications to the internal processing of the attention maps and latent vectors, without modifying the model weights. We propose an appearance transfer method based on partial masking and timestep-controlled appearance modulation, controlling the area and amount of appearance we transfer. The results on the benchmarks outperform current baseline models, which shows the method’s potential for future improvements.

• Language: English
• Keywords: diffusion models, generative models, computer vision, image editing, fashion editing
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2026
• PID: 20.500.12556/RUL-180098
• COBISS.SI-ID: 271564035
• Publication date in RUL: 02.03.2026

Secondary language

• Language: Slovenian
• Title: Urejanje slik z difuzijskimi modeli na področju mode

Abstract

Urejanje slik, zlasti prenos videza ali stila, je področje v domeni računalni- škega vida, ki je zaradi nedavnih napredkov v difuzijskih modelih za slike doživelo velik porast priljubljenosti. Obstaja veliko različnih pristopov k prenosu stila z doučevanjemnjem, zunanjimi omrežnimi adapterji, pa tudi z metodami brez doučevanja. V tem delu se osredotočamo na slednje, kjer izkoriščamo obstoječe omrežje Stable Diffusion z nekaj spremembami mehanizma pozornosti in obdelave latentnih vektorjev, brez spreminjanja uteži modela. Predlagamo novo metodo prenosa videza, ki temelji na delnem maskiranju in časovno nadzorovani modulaciji, s čimer nadzorujemo območje in količino videza, ki ga želimo prenesti. Rezultati na primerjalnih testih presegajo trenutne osnovne modele, kar kaže na potencial metode za prihodnje izboljšave.

• Keywords: difuzijski modeli, generativni modeli, računalniški vid, urejanje slik, modno urejanje