Alon Wolf

I am a researcher at Tel Aviv University with a passion for AI and security vulnerabilities — and especially the intersection of both. My work spans generative models, diffusion-based editing, and understanding how AI systems can be probed, controlled, and secured.

Publications

SIGGRAPH 2026

Continuous Control of Editing Models via Adaptive-Origin Guidance

Alon Wolf^1,2, Chen Katzir², Kfir Aberman², Or Patashnik¹

¹Tel Aviv University ²Decart.ai

We introduce AdaOr (Adaptive-Origin Guidance), enabling smooth and continuous control over edit intensity in diffusion-based image and video editing models — without per-edit optimization or specialized datasets. By introducing a learnable identity instruction and interpolating the CFG origin, AdaOr allows seamless linear interpolation between the input and any target edit.

🌐 Project Page 📄 arXiv ✍️ Blog Post