🔍 Trying to fine-tune Stable Diffusion on images with a specific style? You'll get better results by also adapting the noise distribution to the style! Check our #ICCV2023 paper: 📑openaccess.thecvf.com/content/ICCV20… Collaboration with Largo.ai, Dr. Radhakrishna Achanta and Prof. Sabine Süsstrunk

I presented our paper VETIM at #BMVC2023 three weeks ago! 🔍 It is possible to learn a single token representing a concept in Stable Diffusion using supervision only at the CLIP output and without mimicking visual features from sample images. Check it out: ivrl.github.io/vetim/

✨ Excited to share our work “Exploiting the Signal-Leak Bias in Diffusion Models”! 🖼️ We turn a bias contained in diffusion models into a method to generate images in a desired style or color, or with a more natural color distribution. ivrl.github.io/signal-leak-bi… (1/8) ⬇️

🔥 Just released the code for our #WACV2024 paper "Exploiting the Signal-Leak Bias in Diffusion Models" 📄 🔗 Project page: ivrl.github.io/signal-leak-bi… 👩‍💻 Code: github.com/IVRL/signal-le…